THE

COOKBOOK
Methodologies and tools that reduce analytics
cycle time while improving quality.

Christopher Bergh, Gil Benghiat and Eran Strod

The DataOps Cookbook
© 2019 DataKitchen, Inc. All Rights Reserved.

To order additional copies of this book:

[email protected]

DataKitchen Headquarters:
101 Main Street, 14th Floor
Cambridge, MA 02142

Our Mailing Address:

One Broadway, 14th Floor
Cambridge, MA 02142

Printed in the United States of America

Cover design and layout by Ariel Plotkin-Gould
 Contents
FOREWORD 1

THE DATAOPS MANIFESTO 3

DataOps Principles 3
1. Continually satisfy your customer 3
2. Value working analytics 3
3. Embrace change 3
4. It’s a team sport 4
5. Daily interactions 4
6. Self-organize 4
7. Reduce heroism 4
8. Reflect 4
9. Analytics is code 4
10. Orchestrate 4
11. Make it reproducible 4
12. Disposable environments 4
13. Simplicity 4
14. Analytics is manufacturing 5
15. Quality is paramount 5
16. Monitor quality and performance 5
17. Reuse 5
18. Improve cycle times 5

EIGHT CHALLENGES OF DATA ANALYTICS 9

1 – The Goalposts Keep Moving 10
2 – Data Lives in Silos 10
3 – Data Formats are not Optimized 11
4 – Data Errors 11
5 – Bad Data Ruins Good Reports 12
6 – Data Pipeline Maintenance Never Ends 12
7 – Manual Process Fatigue 12
8 – The Trap of “Hope and Heroism” 12
Overcoming the Challenges 13
WHAT IS DATAOPS 17
Delivering Analytics at Amazon Speed 17
Analytics in the On-Demand Economy 18
The Seven Steps to Implement DataOps 27
Step 1 - Add Data and Logic Tests 27
Step 2 - Use a Version Control System 28
Step 3 - Branch and Merge 29
Step 4 - Use Multiple Environments 30
Step 6 - Parameterize Your Processing 31
Step 7: Work Without Fear or Heroism 32
 DataOps is NOT Just DevOps for Data 35
DataOps Resolves the Struggle Between
Centralization and Freedom in Analytics 47
DATAOPS FOR THE CHIEF DATA OFFICER 57
Prove Your Awesomeness with Data:
The CDO DataOps Dashboard 57
Surviving Your Second Year as CDO 62
CAOs and CDOs: Earn the Trust of your CEO 66
The Four Stage Journey to Analytics Excellence 68
DATAOPS FOR THE DATA ENGINEER AND THE DATA SCIENTIST 73
DataOps Puts Agility into Agile Data Warehousing 73
Speed Up Innovation with DataOps 76
How to Inspire Code Reuse in Data Analytics 79
What Data Scientists Really Need 81
DATAOPS FOR DATA QUALITY 85
Disband Your Impact Review Board:
Automate Analytics Testing 85
Build Trust Through Test Automation and Monitoring 94
How Data Analytics Professionals Can Sleep Better 100
DATAOPS AND YOUR CAREER 105
DataOps Engineer Will Be the Sexiest Job in Analytics 105
Building a DataOps Team 107
DATAOPS EXAMPLES AND CASE STUDIES 113
Grow Sales Using a DataOps-Powered
Customer Data Platform 113
Achieving Growth Targets by Implementing
a DataOps-Powered Customer Data Platform 117
How a Mixed Martial Arts Fighter Would
Approach Data Analytics 121
Reinvent Marketing Automation with
the DataKitchen DataOps Platform 123
Meeting the Product Launch Challenge with DataOps 125
ADDITIONAL RECIPES 129

DATAOPS RESOURCES 133

ABOUT THE AUTHORS 134

 Foreword

In the early 2000s, Chris and Gil worked at a company that specialized in analytics for the
pharmaceutical industry. It was a small company that offered a full suite of services related
to analytics — data engineering, data integration, visualization and what is now called “data
science.” Their customers were marketing and sales executives who tend to be challenging
because they are busy, need fast answers and don’t understand or care about the underlying
mechanics of analytics. They are business people, not technologists.

When a request from a customer came in, Chris and Gil would gather their team of engi-
neers, data scientists and consultants to plan out the how to get the project done. After days
of planning, they would propose their project plan to the customer. “It will take two weeks.”
The customer would shoot back, “I need it in two hours!”

Walking back to their office, tail between their legs, they would pick up the phone. It was a
customer boiling over with anger. There was a data error. If it wasn’t fixed immediately the
customer would find a different vendor.

The company had hired a bunch of smart people to deliver these services. “ I want to
innovate — Can I try out this new open source tool,” the team members would ask. “No,” the
managers would have to answer. “We can’t afford to introduce technical risk.”

They lived this life for many years. How do you create innovative data analytics? How do you
not have embarrassing errors? How do you let your team easily try new ideas? There had to
be a better way.

They found their answer by studying the software and manufacturing industries which had
been struggling with these same issues for decades. They discovered that data-analytics
cycle time and quality can be optimized with a combination of tools and methodologies that
they now call DataOps. They decided to start a new company. The new organization adopted
the kitchen metaphor for data analytics. After all, cooking up charts and graphs requires the
right ingredients and recipes.

The DataOps Cookbook • 1

 The DataKitchen founders (Chris, Gil and Eric) built their own DataOps Platform. What hap-
pened next was remarkable. When analytics go faster, users embrace analytics and innovate.
Rapid, high-quality analytics can unlock an organization’s creative potential.

Having experienced this transformation, the DataKitchen founders sought a way to help
other data professionals. There are so many talented people stuck in no-win situations. This
book is for data professionals who are living the nightmare of slow, buggy analytics and
frustrated users. It will explain why working weekends isn’t the answer. It provides you with
practical steps that you can take tomorrow to improve your analytics cycle time.

DataKitchen markets a DataOps Platform that will help analytics organizations implement
DataOps. However, this book isn’t really about us and our product. It is about you, your
challenges, your potential and getting your analytics team back on track.

The values and principles that are central to DataOps are listed in the DataOps Manifesto
which you can read below. If you agree with it, please join the thousands of others who
share these beliefs by signing the manifesto. There may be aspects of the manifesto that
require further explanation. Please read on. By the end of this book, it should all make sense.

You’ll also notice that we’ve included some real recipes in this book. These are some of our
favorites. We hope you enjoy them!

Please reach out to us at [email protected] with any comments or questions.

Chris, Gil and Eran

2 • The DataOps Cookbook

 The DataOps Manifesto

Background
Through firsthand experience working with data across organizations, tools, and industries
we have uncovered a better way to develop and deliver analytics that we call DataOps.
Whether referred to as data science, data engineering, data management, big data, business
intelligence, or the like, through our work we have come to value in analytics:
• Individuals and interactions over processes and tools
• Working analytics over comprehensive documentation
• Customer collaboration over contract negotiation
• Experimentation, iteration, and feedback over extensive upfront design
• Cross-functional ownership of operations over siloed responsibilities

DataOps Principles
1. CONTINUALLY SATISFY YOUR CUSTOMER
Our highest priority is to satisfy the customer through the early and continuous delivery of
valuable analytic insights from a couple of minutes to weeks.

2. VALUE WORKING ANALYTICS

We believe the primary measure of data analytics performance is the degree to which
insightful analytics are delivered, incorporating accurate data, atop robust frameworks and
systems.

3. EMBRACE CHANGE
We welcome evolving customer needs, and in fact, we embrace them to generate competi-
tive advantage. We believe that the most efficient, effective, and agile method of communi-
cation with customers is face-to-face conversation.

The DataOps Cookbook • 3

 4. IT’S A TEAM SPORT
Analytic teams will always have a variety of roles, skills, favorite tools, and titles.

5. DAILY INTERACTIONS
Customers, analytic teams, and operations must work together daily throughout the project.

6. SELF-ORGANIZE
We believe that the best analytic insight, algorithms, architectures, requirements, and de-
signs emerge from self-organizing teams.

7. REDUCE HEROISM
As the pace and breadth of need for analytic insights ever increases, we believe analytic
teams should strive to reduce heroism and create sustainable and scalable data analytic
teams and processes.

8. REFLECT
Analytic teams should fine-tune their operational performance by self-reflecting, at regular
intervals, on feedback provided by their customers, themselves, and operational statistics.

9. ANALYTICS IS CODE
Analytic teams use a variety of individual tools to access, integrate, model, and visualize data.
Fundamentally, each of these tools generates code and configuration which describes the
actions taken upon data to deliver insight.

10. ORCHESTRATE
The beginning-to-end orchestration of data, tools, code, environments, and the analytic
team’s work is a key driver of analytic success.

11. MAKE IT REPRODUCIBLE

Reproducible results are required and therefore we version everything: data, low-level
hardware and software configurations, and the code and configuration specific to each tool
in the toolchain.

12. DISPOSABLE ENVIRONMENTS

We believe it is important to minimize the cost for analytic team members to experiment by
giving them easy to create, isolated, safe, and disposable technical environments that reflect
their production environment.

13. SIMPLICITY
We believe that continuous attention to technical excellence and good design enhances
agility; likewise simplicity—the art of maximizing the amount of work not done—is essential.

4 • The DataOps Cookbook

14. ANALYTICS IS MANUFACTURING
Analytic pipelines are analogous to lean manufacturing lines. We believe a fundamental con-
cept of DataOps is a focus on process-thinking aimed at achieving continuous efficiencies in
the manufacture of analytic insight.

15. QUALITY IS PARAMOUNT

Analytic pipelines should be built with a foundation capable of automated detection of ab-
normalities (jidoka) and security issues in code, configuration, and data, and should provide
continuous feedback to operators for error avoidance (poka yoke).

16. MONITOR QUALITY AND PERFORMANCE

Our goal is to have performance, security and quality measures that are monitored continu-
ously to detect unexpected variation and generate operational statistics.

17. REUSE
We believe a foundational aspect of analytic insight manufacturing efficiency is to avoid the
repetition of previous work by the individual or team.

18. IMPROVE CYCLE TIMES

We should strive to minimize the time and effort to turn a customer need into an analytic
idea, create it in development, release it as a repeatable production process, and finally
refactor and reuse that product.

Join the Thousands of People Who Have Already Signed The Manifesto

The DataOps Cookbook • 5

6 • The DataOps Cookbook
The DataOps Cookbook • 7
 DataOps Ribeye
by Christopher Bergh

INGREDIENTS AND TOOLS

• Rib eye steaks 1 ½-2” thick
• Large baking potatoes
• Corn on the husk
• Olive oil
• Ground sea salt & Fresh coarse ground pepper or Penzey’s Chicago Steak
Seasoning
• Green Egg or another charcoal grill
• Instapen thermometer

PREPARE CORN AND POTATOES

1. Season/marinade steaks to your liking, but I now like the ‘TRex way of
prepping/grilling my steaks (see below): coat the steaks with olive oil, liberally
apply salt and pepper (or Penzey’s Chicago Steak Seasoning) and rub the
mixture into the steaks on both sides.
2. Soak corn in water for at least 30 minutes or more.
3. Brush potatoes with olive oil and salt on both sides. Heat Green Egg to 400
degrees with a few chunks of Hickory wood and put potatoes directly on grill
grid, direct heat, turning once after 30 minutes. About 20-25 minutes before
the potatoes are done, put the corn on the grill grid. Turn the corn about
every 8-10 minutes or just enough to keep the corn from getting burned (a
little charring of the corn is okay, though).
4. Remove the potatoes and corn and wrap in aluminum foil and place in a
warming oven or drawer at 170 degrees.

PREPARE THE STEAKS

1. TRex method of grilling/cooking steaks: Crank the draft door and daisy wheel
open all the way on the Egg for maximum temperature. When you get the
Green Egg up to at least over 600 degrees (be careful to “burp” the Egg at
this high temp), put the steaks on and sear for 90 seconds per side.
2. Take the steaks off the Egg and let sit/rest for 20 minutes. Shut down the
lower vent and Daisy Wheel to get the Green Egg back down to 425 degrees.
3. Throw some more chunks of Hickory wood on the fire and try and maintain a
temp around 400 degrees. After 10 minutes of letting the steaks sit/rest put
back on the Green Egg and cook for approximately 4-5 minutes per side for a
medium-rare/medium result. Check internal temperature of steak so that it is
120 degrees
8 • The DataOps Cookbook
 Eight Challenges of
Data Analytics

Companies increasingly look to analytics to drive growth strategies. As the leader of the
data-analytics team, you manage a group responsible for supplying business partners with
the analytic insights that can create a competitive edge. Customer and market opportunities
evolve quickly and drive a relentless series of questions. Analytics, by contrast, move slowly,
constrained by development cycles, limited resources and brittle IT systems. The gap be-
tween what users need and what IT can provide can be a source of conflict and frustration.
Inevitably this mismatch between expectations and capabilities can cause dissatisfaction,
leaving the data-analytics team in an unfortunate position and preventing a company from
fully realizing the strategic benefit of its data.

As a manager overseeing analytics, it’s your job to understand and address the factors that
prevent the data-analytics team from achieving peak levels of performance. If you talk to
your team, they will tell you exactly what is slowing them down. You’ll likely hear variations
of the following eight challenges:

The DataOps Cookbook • 9

 1 – The Goalposts Keep Moving
Users are demanding customers for a data-analytics team. Their requirements change
constantly. They require immediate responses, and no matter how much the analytics team
delivers, users keep generating new requests. It’s enough to overwhelm any data-analytics
team.

They don’t know what they want. Users are not data experts. They don’t know what insights
are possible until someone from your team shows them. Sometimes they don’t know what
they want until after they see it in production (and maybe not even then). Often, business
stakeholders do not know what they will need next week, let alone next quarter or next year.
It’s not their fault. It’s the nature of pursuing opportunities in a fast-paced marketplace.

They need everything ASAP. Business is a competitive endeavor. When an opportunity opens,
the company needs to move on it faster than the competition. When users bring a question
to the data-analytics team, they expect an immediate response. They can’t wait weeks or
months — the opportunity will close as the market seeks alternative solutions.

The questions never end. Sometimes providing business stakeholders with analytics generates
more questions than answers. Analytic insights enable users to understand the business
in new ways. This spurs creativity, which leads to requests for more analytics. A healthy
relationship between the analytics and users will foster a continuous series of questions that
drive demand for new analytics. However, this relationship can sour quickly if the delivery of
new analytics can’t meet the required time frames.

2 – Data Lives in Silos

In pursuit of business objectives, companies collect an enormous amount of data: orders,
deliveries, returns, website page views, mobile app navigations, downloads, clicks, metrics,
audio logs, social media and more. Further, this data can be combined with demographic,

10 • The DataOps Cookbook

psychographic or other third-party market data. All of this data is collected in separate da-
tabases which typically do not talk to each other. They utilize numerous platforms, APIs and
technologies. Accessing all of this data is a daunting task requiring such a wide range of skills
that it is rare to find a single person who can do it all. Integrating data from these myriad
sources becomes a major undertaking.

Business stakeholders want fast answers. Meanwhile, the data-analytics team has to work
with IT to gain access to operational systems, plan and implement architectural changes, and
develop/test/deploy new analytics. This process is complex, lengthy and subject to numer-
ous bottlenecks and blockages.

3 – Data Formats are not Optimized

Data in operational systems is usually not structured in a way that lends itself to the efficient
creation of analytics. For example, an ERP system might have a schema that is optimized for
inserts, updates, and for display in a web user interface. For operational systems, these are
the actions that need to happen in real time.

A database optimized for data analytics is structured to optimize reads and aggregations. It’s
also important for the schema of an analytics database to be easily understood by humans.
For example, the field names would be descriptive of their contents and data tables would
be linked in ways that make intuitive sense.

4 – Data Errors
Whether your data sources are internal or from external third parties, data will eventually
contain errors. Data errors can prevent your data pipeline from flowing correctly. Errors may
also be subtle, such as duplicate records or individual fields that contain erroneous data.
Data errors could be caused by a new algorithm that doesn’t work as expected, a database
schema change that broke one of your feeds, an IT failure or one of many other possibilities.
Data errors can be difficult to trace and resolve quickly.

The DataOps Cookbook • 11

 5 – Bad Data Ruins Good Reports
When data errors work their way through the data pipeline into published analytics, internal
stakeholders can become dissatisfied. This causes unplanned work, which diverts your key
contributors from the highest priority projects. Bad data also harms the hard-won credibil-
ity of the data-analytics team. If business colleagues repeatedly see bad data in analytics
reports, they might learn not to trust or value the work product of the data-analytics team.

6 – Data Pipeline Maintenance Never Ends

Data-analytics is a pipeline process that executes a set of operations and attempts to
produce a consistent output at a high level of quality. Every new or updated data source,
schema enhancement, analytics improvement or other change triggers an update to the
pipeline. The data-analytics team is continuously making changes and improvements to the
data pipeline. Each one of these changes must be made carefully so that it doesn’t break
operational analytics. The effort required to validate and verify changes often takes longer
than the time required to create the changes in the first place. You may not realize it, but
your analysts, data scientists and engineers may be spending 80% of their time updating,
maintaining and assuring the quality of the data pipeline. This is necessary work, but much of
it is behind the scenes and unappreciated when viewed against the growing backlog of new
requests from business customers.

7 – Manual Process Fatigue

Data integration, cleansing, transformation, quality assurance and deployment of new ana-
lytics must be performed flawlessly day in and day out. The data-analytics team may have
automated a portion of these tasks, but some teams perform numerous manual processes on
a regular basis. These rote procedures are error prone, time consuming and tedious.

Further, manual processes can also lead to high employee turnover. Many managers have
watched high-performing data-analytics team members burn out due to having to repeat-
edly execute manual data procedures. Manual processes strain the productivity of the data
team in numerous ways.

8 – The Trap of “Hope and Heroism”

The landscape is littered with projects and initiatives cancelled or deferred due to changing
requirements, slipped schedules, disappointed users, inflexibility, poor quality, low ROI, and
irrelevant features.

According to the research firm Gartner, Inc., half of all chief data officers (CDO) in large
organizations will not be deemed a success in their role. Per Forrester Research, 60% of the
data and analytics decision-makers surveyed said they are not very confident in their analytics
insights. Only ten percent responded that their organizations sufficiently manage the quality
of data and analytics. Just sixteen percent believe they perform well in producing accurate
models.

12 • The DataOps Cookbook

Many CDO’s and data-analytics professionals respond to these challenges in one of three
ways:

Heroism - Data-analytics teams work long hours to compensate for the gap between perfor-
mance and expectations. When a deliverable is met, the data-analytics team is considered
heroes. However, yesterday’s heroes are quickly forgotten when there is a new deliverable
to meet. Also, this strategy is difficult to sustain over a long period of time, and it, ultimately,
just resets expectations at a higher level without providing additional resources. The heroism
approach is also difficult to scale up as an organization grows.

Hope - When a deadline must be met, it is tempting to just quickly produce a solution with
minimal testing, push it out to the users and hope it does not break. This approach has inher-
ent risks. Eventually, a deliverable will contain data errors, upsetting the users and harming
the hard-won credibility of the data-analytics team.

Caution - The team decides to give each data-analytics project a longer development and
test schedule. Effectively, this is a decision to deliver higher quality, but fewer features to
users. One difficulty with this approach is that users often don’t know what they want until
they see it, so a detailed specification might change considerably by the end of a project. The
slow and methodical approach might also make the users unhappy because the analytics are
delivered more slowly than their stated delivery requirements and as requests pile up, the
data-analytics team risks being viewed as bureaucratic and inefficient.

None of these approaches adequately serve the needs of both users and data-analytics
professionals, but there is a way out of this bind. The challenges above are not unique to
analytics, and in fact, are shared by other organizations.

Overcoming the
Challenges
Some say that an analytics
team can overcome these
challenges by buying a
new tool. While it is true
that new tools are helpful,
they are not enough by
themselves. You cannot
truly transform your staff
into a high-performance
team without an overhaul
of the methodologies and
processes that guide your
workflows. In this book, we will discuss how to combine tools and new processes in a way
that improves the productivity of your data analytics team by orders of magnitude.

The DataOps Cookbook • 13

14 • The DataOps Cookbook
The DataOps Cookbook • 15
 DataOps Chili Mole
by Gil Benghiat

INGREDIENTS AND TOOLS

• 1 pound ground beef (90% lean or leaner)
• 1 diced onion
• 2 diced Jalapeño or 2 Serrano or 1 habanero pepper
• 2 pounds frozen sweet corn
• 28 oz can crushed tomatoes
• 15.5 oz can of black beans, NOT drained
• 15.5 oz can of black beans, drained
• 15.5 oz can of red beans, drained
• 15.5 oz can of white beans, drained
• 7 cloves of crushed garlic
• ½ cup sugar
• 3 ½ table spoons of unsweetened cocoa powder
• 3 table spoons of chili powder
• 1 teaspoon of salt

INSTRUCTIONS
1. Crock Pot: Combine all ingredients and cook on high for 5-8 hours.
Stir occasionally.
2. Stove Top: Combine ground beef, onion, and pepper. Cook on medium high
until beef is cooked through. Add the remaining ingredients and cook on
low-simmer for 1-2 hours. Stir occasionally.
3. For vegan chili: Substitute 5 tablespoons of canola oil for the ground beef.
4. Serve with rice.

16 • The DataOps Cookbook

 What Is DataOps

You can view DataOps in the context of a century-long evolution of ideas that improve
how people manage complex systems. It started with pioneers like W. Edwards Deming and
statistical process control - gradually these ideas crossed into the technology space in the
form of Agile, DevOps and now, DataOps. In the next section we will examine how these
methodologies impact productivity, quality and reliability in data analytics.

Delivering Analytics at Amazon Speed

The world changed in February 2005 when Amazon Prime brought flat-rate, unlimited, two-
day shipping into a world where people expected to pay extra to receive packages in four
to six business days. Since its launch, Amazon Prime has completely transformed the retail
market, making low-cost, predictable shipping an integral part of consumer expectations.
This business model, which some have called the “on-demand economy,” is popping up in
many industries and markets across the globe.

For example, some may remember video stores where movies were rented for later viewing.
Today, 65 percent of global respondents to a recent Nielsen survey watch video on demand
(VOD), many of them daily. With VOD, a person’s desire to watch a movie is fulfilled within
seconds. Amazon participates in the VOD market with their Amazon Prime Video service.

Instant fulfillment of customer orders seems to be part of Amazon’s business model. They
have even brought that capability to IT. About 10 years ago, Amazon Web Services (AWS)
began offering computing, storage, and other IT infrastructure on an as-needed basis.
Whether the need is for one server or thousands and whether for hours, days, or months,
you only pay for what you use, and the resources are available in just a few minutes.

To successfully compete in today’s on-demand economy, companies need to deliver their

products and services just as Amazon has done—in other words, at Amazon speed. What
might be surprising to many is how the expectations of instant fulfillment are crossing over
into data analytics, which, along with everything else in the digital economy, is now expected
to happen at Amazon speed and with Amazon predictability.

The DataOps Cookbook • 17

 A typical example: the VP of sales enters the office of the chief data officer (CDO). She’d like
to cross-reference the customer database with some third-party consumer data. The CDO
asks for time to study the problem and, days later, has planned the project. Resources will be
allocated and configured, schemas will be updated, reports will be elegantly designed, and
the delivery pipeline will be thoroughly tested. The changes will take several weeks. “Not
acceptable,” the VP of sales fires back. The new analytics are needed for a meeting with the
board later in the week. “The competition is ahead of us; we can’t wait weeks.” This scenario
is playing out in one form or another in corporations around the globe.

ANALYTICS IN THE ON-DEMAND ECONOMY

Analytics must be delivered rapidly in order to meet user expectations in the on-demand
economy. This is simply not possible with an approach that depends upon “hope and hero-
ism” or “caution.”

In order to deliver value consistently, quickly and accurately, data-analytics teams must learn
to create and publish analytics in a new way. We call this new approach DataOps. DataOps
is a combination of tools and methods, which streamline the development of new analytics
while ensuring impeccable data quality. DataOps helps shorten the cycle time for producing
analytic value and innovation, while avoiding the trap of “hope, heroism and caution.”

18 • The DataOps Cookbook

Data Analytics Can Learn from Agile
If you were managing 100 software developers, you would have to choose the best way
to maximize their productivity. Since the dawn of the computer era many software project
management approaches have been tried. The waterfall model dominated software project
management up until the 1990’s. In the early days of computing, project management
was adapted from the manufacturing and construction industries, which required detailed
planning and a great degree of structure. Projects were organized into phases (conception,
initiation, analysis, design, construction, testing, production/implementation and mainte-
nance) and progressed through these phases sequentially. Once a phase was done, the team
moved forward to the next phase.

Figure 1: In Agile development, a burndown chart shows work remaining over time.

The waterfall model is better suited to situations where the requirements are fixed and well
understood up front. This is nothing like the technology industry where the competitive
environment evolves rapidly. In the 1980’s a typical software project required about 12
calendar months. In technology-driven businesses (i.e. nearly everyone these days) custom-
ers demand new features and services, and competitive pressures change priorities on a
seemingly daily basis. The waterfall model has no mechanism to respond to these changes.
In waterfall, changes trigger a seemingly endless cycle of replanning causing delays and
resulting in project budget overruns.

In the early 2000’s, the software industry embraced a new approach to code production
called Agile Development. Agile is an umbrella term for several different iterative and incre-
mental software development methodologies.

In Agile Software Development, the team and its processes and tools are organized around
the goal of publishing releases to the users every few weeks (or at most every few months).
A development cycle is called a sprint or an iteration. At the beginning of an iteration, the
team commits to completing working and (the most) valuable changes to the code base.
Features are associated with user stories, which help the development team understand the

The DataOps Cookbook • 19

 context behind requirements. User stories include descriptions of features and acceptance
criteria. The Agile methodology is particularly good for non-sequential product develop-
ment where market requirements are quickly evolving. This is similar to the data-analytics
environment where each new analysis and report of the data inspires requests for additional
queries.

Agile is widely credited with boosting software productivity. One study sponsored by the
Central Ohio Agile Association and Columbus Executive Agile Special Interest Group found
that Agile projects were completed 31 percent faster and with a 75 percent lower defect
rate than the industry norm. The vast majority of companies are getting on-board. In a survey
of 400 IT professionals by TechBeacon, two-thirds described their company as either “pure
agile” or “leaning towards agile. Among the remaining one third of companies, most use a
hybrid approach, leaving only nine percent using a pure waterfall approach.

In an increasingly competitive marketplace, Agile methods allow companies to become more

responsive to customer requirements and accelerate time to market. Agile also improves ROI
because features delivered in each iteration can be immediately monetized instead of wait-
ing months for a big release. Agile is the major reason that release frequency improved from
around 3 months in the 1990’s to about 3 weeks in the 2000’s. However, improvements
didn’t stop there. Today releases are occurring every few seconds using an updated approach
which builds upon Agile.

Figure 2: The decrease in release cycle time as software development evolved

from waterfall to Agile to DevOps.

Data Analytics Can Learn from DevOps

Before the advent of on-demand cloud services, the various groups in software develop-
ment (design, development, test, quality, support, …) had to set-up their own infrastructure.
Whatever components were needed (physical servers, networks, storage, software, …) had
to be ordered, installed, configured and managed by the IT department. Servers would be
ordered at different times and from different vendors, each slightly different from the other.

20 • The DataOps Cookbook

Depending on the task at hand, different machines could have a different array of software
applications with revisions of each app being continuously updated. Target devices could
range from embedded IoT (Internet of Things) to the largest, most powerful severs. With
all of this variability, it was quite common for individuals within the company to be running
code in different environments. Outside the four walls of the company, customers could be
running in yet another environment. This situation presented challenges.

The software development pipeline can be organized as follows: planning, resourcing,

development, testing, quality assurance, and deployment. Continuous delivery requires
automation from planning all the way through to delivery/deployment. Previously, software
development, testing, quality assurance, and customers could each be running in different
environments. This hampered their ability to communicate and led to misunderstandings and
delays.

If, for example, the customer reported a problem, it might not be replicable in the support,
test or development groups due to differences in the hardware and software environments
being run. This lack of alignment fostered misunderstandings and delays and often led to a
lack of trust and communication between the various stakeholders.

In a complex world requiring the physical provisioning of servers, installation of stacks and
frameworks, and numerous target devices, the standardization and control of the run-time
environment has been difficult and slow. It became necessary to break down barriers be-
tween the respective teams in the software development pipeline. This merging of develop-
ment and IT/Operations is widely known as DevOps, which also has had enormous impact
on the world of software development. DevOps improves collaboration between employees
from the planning through the deployment phases of software. It seeks to reduce time to
deployment, decrease time to market, minimize defects, and shorten the time required to fix
problems.

About a decade ago, Amazon Web Services (AWS) and other cloud providers, began offering
computing, storage and other IT resources as an on-demand service. No more waiting weeks
or months for the IT department to fulfill a request for servers. Cloud providers now allow
you to order computing services, paying only for what you use, whether that is one proces-
sor for an hour or thousands of processors for months. These on-demand cloud services
have enabled developers to write
code that provisions processing
resources with strictly specified
environments, on-demand, in just
a few minutes. This capability has
been called Infrastructure as Code
(IaC). IaC has made it possible for
everyone in the software devel-
opment pipeline, all the different
groups mentioned above, to use
an identical environment tailored
to application requirements. With
IaC, design, test, QA and support

The DataOps Cookbook • 21

 can easily get on the same page. This leads to much better collaboration between the groups
and breaks down barriers that prevented open communication. In other words, no more
finger pointing.

With IT infrastructure being defined by code, the hard divisions between IT operations and
software development are able to blur. The merger of development and operations is how
the term DevOps originated.

With the automated provisioning of resources, DevOps paved the way for a fully automated
test and release process. The process of deploying code that used to take weeks, could now
be completed in minutes. Major organizations including Amazon, Facebook and Netflix are
now operating this way. At a recent conference, Amazon disclosed that their AWS team per-
forms 50,000,000 code releases per year. This is more than one per second! This methodol-
ogy of rapid releases is called continuous delivery or alternatively, continuous deployment, when
new features (and fixes) are not only delivered internally but fully deployed to customers.

DevOps starts with continuous delivery and Agile development and adds automated provi-
sioning of resources (infrastructure as code) and cloud services (platform as a service) to en-
sure that the same environment is being utilized at every stage of the software development
pipeline. The cloud provides a natural platform that allows individuals to create and define
identical run-time environments. DevOps is beginning to achieve critical mass in terms of its
adoption within the world of software development.

DevOps improves collaboration between employees from the planning through the deploy-
ment phases of software. It seeks to reduce time to deployment, decrease time to market,
minimize defects, and shorten the time required to fix problems.

The impact of DevOps on development organizations was shown in a 2014 survey, “The
2014 State of DevOps Report” by Puppet Labs, IT Revolution Press and ThoughtWorks,
based on 9,200 survey responses from technical professionals. The survey found that IT or-
ganizations implementing DevOps were deploying code 30 times more frequently and with
50 percent fewer failures. Further, companies with these higher performing IT organizations
tended to have stronger business performance, greater productivity, higher profitability and
larger market share. In other words, DevOps is not just something that engineers are doing
off in a dark corner. It is a core competency that helps good companies become better.

The Data analytics team transforms raw data into actionable information that improves
decision making and provides market insight. Imagine an organization with the best data
analytics in the industry. That organization would have a tremendous advantage over com-
petitors. That could be you.

Data Analytics Can Learn from Manufacturing

What could data analytics professionals possibly learn from industrial manufacturers? It
turns out, a lot. Automotive giant Toyota pioneered a set of methods, later folded into a dis-
cipline called lean manufacturing, in which employees focus relentlessly on improving quality
and reducing non-value-add activities. This culture enabled Toyota to grow into the one of
the world’s leading car companies. The Agile and DevOps methods that have led to stellar
improvements in coding velocity are really just an example of lean manufacturing principles
applied to software development.

22 • The DataOps Cookbook

Conceptually, manufacturing is a pipeline process. Raw materials enter the manufacturing
floor through the stock room, flow to different work stations as work-in-progress and exit
as finished goods. In data-analytics, data progresses through a series of steps and exits in
the form of reports, models and visualizations. Each step takes an input from the previous
step, executes a complex procedure or set of instructions and creates output for the subse-
quent step. At an abstract level, the data-analytics pipeline is analogous to a manufacturing
process. Like manufacturing, data analytics executes a set of operations and attempts to
produce a consistent output at a high level of quality. In addition to lean-manufacturing-in-
spired methods like Agile and DevOps, there is one more useful tool that can be taken from
manufacturing and applied to data-analytics process improvement.

W. Edwards Deming championed statistical process control (SPC) as a method to improve

manufacturing quality. SPC uses real-time product or process measurements to monitor and
control quality during manufacturing processes. If the process measurements are maintained
within specific limits, then the manufacturing process is deemed to be functioning properly.
When SPC is applied to the data-analytics pipeline, it leads to remarkable improvements in
efficiency and quality. For example, Google executes over one hundred million automated
test scripts per day to validate any new code released by software developers. In the Google
consumer surveys group, code is deployed to customers eight minutes after a software engi-
neer finishes writing and testing it.

In data analytics, tests should verify that the results of each intermediate step in the
production of analytics matches expectations. Even very simple tests can be useful. For
example, a simple row-count test could catch an error in a join that inadvertently produces a
Cartesian product. Tests can also detect unexpected trends in data, which might be flagged
as warnings. Imagine that the number of customer transactions exceeds its historical average
by 50%. Perhaps that is an anomaly that upon investigation would lead to insight about
business seasonality.

Tests in data analytics can be applied to data or models either at the input or output of a
phase in the analytics pipeline. Tests can also verify business logic.

The DataOps Cookbook • 23

 Business logic tests validate assumptions about the data. For example:
• Customer Validation – Each customer should exist in a dimension table
• Data Validation – At least 90 percent of data should match entries in a dimension table

Input tests check data prior to each stage in the analytics pipeline. For example:
• Count Verification – Check that row counts are in the right range, ...
• Conformity – US Zip5 codes are five digits, US phone numbers are 10 digits, ...
• History – The number of prospects always increases, ...
• Balance – Week over week, sales should not vary by more than 10%, ...
• Temporal Consistency – Transaction dates are in the past, end dates are later than start
dates, ...
• Application Consistency – Body temperature is within a range around 98.6F/37C, ...
• Field Validation – All required fields are present, correctly entered, ...

Output tests check the results of an operation, like a Cartesian join. For example:
• Completeness – Number of customer prospects should increase with time
• Range Verification – Number of physicians in the US is less than 1.5 million

The data analytics pipeline is a complex process with steps often too numerous to be moni-
tored manually. SPC allows the data analytics team to monitor the pipeline end-to-end from
a big-picture perspective, ensuring that everything is operating as expected. As an automat-
ed test suite grows and matures, the quality of the analytics is assured without adding cost.
This makes it possible for the data analytics team to move quickly — enhancing analytics to
address new challenges and queries — without sacrificing quality.

Figure 3: DataOps has evolved from lean manufacturing and

software methodologies.

24 • The DataOps Cookbook

DataOps for Data Analytics
The speed and flexibility achieved by Agile and DevOps, and the quality control attained by
SPC, can be applied to data analytics. Leading edge proponents of this approach are calling it
DataOps. DataOps, simply stated, is Agile development and DevOps with statistical process
control, for data analytics. DataOps applies Agile methods, DevOps, and manufacturing
quality principles, methodologies and tools, to the data-analytics pipeline. The result is a
rapid-response, flexible and robust data-analytics capability, which is able to keep up with
the creativity of internal stakeholders and users.

DataOps is an analytic development method that emphasizes communication, collaboration,

integration, automation, measurement and cooperation between data scientists, analysts,
data/ETL (extract, transform, load) engineers, information technology (IT), and quality assur-
ance/governance. The method acknowledges the interdependence of the entire end-to-end
analytic process. It aims to help organizations rapidly produce insight, turn that insight into
operational tools, and continuously improve analytic operations and performance. It enables
the whole analytic team involved in the analytic process to follow the values laid out in the
DataOps Manifesto.

When DataOps is implemented correctly, it addresses many of the issues discussed earlier
that have plagued data-analytics teams.

Table 1: Challenge/DataOps Approach

DataOps views the data-analytics pipeline as a process and as such focuses on how to make
the entire process run more rapidly and with higher quality, rather than optimizing the pro-
ductivity of any single individual or tool by itself.

The DataOps Cookbook • 25

 Key Benefits of DataOps
DataOps can accelerate the ability of data-analytics teams to create and publish new
analytics to users. It requires an Agile mindset and must also be supported by an automated
platform which incorporates existing tools into a DataOps development pipeline. DataOps
spans the entire analytic process, from data acquisition to insight delivery. Its goal is to
achieve more insight and better analysis, while still being faster, cheaper and higher quality.

The key business benefits of adopting DataOps are:

• Reduce time to insight
• Improve analytic quality
• Lower the marginal cost to ask the next business question
• Improve analytic team morale by going beyond hope, heroism and caution
• Promote team efficiency through agile process, reuse and refactoring

DataKitchen markets an automated DataOps platform that helps companies accelerate their
DataOps implementation, but this book is about DataOps not us. This book is not trying to
sell you anything. You can implement DataOps all by yourself, using your existing tools, by
implementing the seven steps described in the next section. If you desire assistance, there is
an ecosystem of DataOps vendors who offer a variety of innovative solutions and services.

26 • The DataOps Cookbook

The Seven Steps to Implement DataOps
Data analytics has become business critical, but requirements quickly evolve and data-an-
alytics teams that respond to these challenges in the traditional ways often end up facing
disappointed users. DataOps offers a more effective approach that optimizes the productivi-
ty of the data analytics pipeline by an order of magnitude.

Imagine the next time that the Vice President of Marketing requests a new customer
segmentation, by tomorrow. With DataOps, the data-analytics team can respond ‘yes’ with
complete confidence that the changes can be accomplished quickly, efficiently and robustly.
How then does an organization implement DataOps? You may be surprised to learn that an
analytics team can migrate to DataOps in seven simple steps.

STEP 1 - ADD DATA AND LOGIC TESTS

If you make a change to an analytic pipeline, how do you know that you did not break any-
thing? Automated testing insures that a feature release is of high quality without requiring
time-consuming, manual testing. The idea in DataOps is that every time a data-analytics
team member makes a change, he or she adds a test for that change. Testing is added
incrementally, with the addition of each feature, so testing gradually improves and quality is
literally built in. In a big run, there could be hundreds of tests at each stage in the pipeline.

Adding tests in data analytics is analogous to the statistical process control that is imple-
mented in a manufacturing operations flow. Tests insure the integrity of the final output by
verifying that work-in-progress (the results of intermediate steps in the pipeline) matches
expectations. Testing can be applied to data, models and logic. The figure below shows
examples of tests in the data-analytics pipeline.

For every step in the data-analytics pipeline, there should be at least one test. The philoso-
phy is to start with simple tests and grow over time. Even a simple test will eventually catch
an error before it is released out to the users. For example, just making sure that row counts
are consistent throughout the process can be a very powerful test. One could easily make
a mistake on a join and make a cross product which fails to execute correctly. A simple row-
count test would quickly catch that.

Tests can detect warnings in addition to errors. A warning might be triggered if data exceeds
certain boundaries. For example, the number of customer transactions in a week may be
OK if it is within 90% of its historical average. If the transaction level exceeds that, then a
warning could be flagged. This might not be an error. It could be a seasonal occurrence for
example, but the reason would require investigation. Once recognized and understood, the
users of the data could be alerted.

DataOps is not about being perfect. In fact, it acknowledges that code is imperfect. It’s
natural that a data-analytics team will make a best effort, yet still miss something. If so, they
can determine the cause of the issue and add a test so that it never happens again. In a rapid
release environment, a fix can quickly propagate out to the users.

With a suite of tests in place, DataOps allows you to move fast because you can make
changes and quickly rerun the test suite. If the changes pass the tests, then the data-analyt-

The DataOps Cookbook • 27

 ics team member can be confident and release it. The knowledge is built into the system and
the process stays under control. Tests catch potential errors and warnings before they are
released so the quality remains high.

Automated tests continuously monitor the data pipeline for errors and anomalies. They work
nights, weekends and holidays without taking a break. If you build a DataOps dashboard, you
can view the high-level state of your data operations at any time. If warning and failure alerts
are automated, you don’t have to constantly check your dashboard. Automated testing frees
the data-analytics team from the drudgery of manual testing, so they can focus on higher
value-add activities.

Figure 4: Tests enable the data professional to apply statistical process controls
to the data pipeline

STEP 2 - USE A VERSION CONTROL SYSTEM

There are many processing steps that turn raw data into useful information for stakeholders.
To be valuable, data must progress through these steps, linked together in some way, with
the ultimate goal of producing a data-analytics output. Data may be preprocessed, cleaned,
checked, transformed, combined, analyzed, and reported. Conceptually, the data-analysis
pipeline is a set of stages implemented using a variety of tools including ETL tools, data
science tools, self-service data prep tools, reporting tools, visualization tools and more.
The stages may be executed serially, but many stages can be parallelized. The pipeline is
deterministic because the pipeline stages are defined by scripts, source code, algorithms,
html, configuration files, parameter files, containers and other files. All of these items are
essentially just code. Code controls the entire data-analytics pipeline from end to end in a
reproducible fashion.

The artifacts (files) that make this reproducibility possible are usually subject to continuous
improvement. Like other software projects, the source files associated with the data pipeline
should be maintained in a version control (source control) system such as Git. A version con-
trol tool helps teams of individuals organize and manage the changes and revisions to code.
It also keeps code in a known repository and facilitates disaster recovery. However, the most
important benefit of version control relates to a process change that it facilitates. It allows
data-analytics team members to branch and merge.

28 • The DataOps Cookbook

STEP 3 - BRANCH AND MERGE
In a typical software project, developers are continuously updating various code source files.
If a developer wants to work on a feature, he or she pulls a copy of all relevant code from the
version control tool and starts to develop changes on a local copy. This local copy is called a
branch. This approach can help data-analytics teams maintain many coding changes to the
data-analytics pipeline in parallel. When the changes to a branch are complete and tested,
the code from the branch is merged back into the trunk, where the code came from.

Branching and merging can be a major productivity boost for data analytics because it allows
teams to make changes to the same source code files in parallel without slowing each other
down. Each individual team member has control of his or her work environment. They can
run their own tests, make changes, take risks and experiment. If they wish, they can discard
their changes and start over. Another key to allowing team members to work well in parallel
relates to providing them with an isolated machine environment.

Figure 5: Branching and merging enables parallel development in data analytics.

The DataOps Cookbook • 29

 STEP 4 - USE MULTIPLE ENVIRONMENTS
Every data-analytics team member has their own development tools on their own laptop.
Version control tools allow team members to work on their own private copy of the source
code while still staying coordinated with the rest of the team. In data analytics, a team mem-
ber can’t be productive unless they also have a copy of the data that they need. Most use
cases can be covered in less than a Terabyte (TB). Historically, disk space has been prohibi-
tively expensive, but today, at less than $25 per TB per month (cloud storage), costs are now
less significant than the opportunity cost of a team member’s time. If the data set is still too
large, then a team member can take only the subset of data that is needed. Often the team
member only needs a representative copy of the data for testing or developing one set of
features.

When many team members work on the production database, it can lead to conflicts. A
database engineer changing a schema may break reports. A data scientist developing a new
model might get confused as new data flows in. Giving team members their own Environ-
ment isolates the rest of the organization from being impacted by their work.

STEP 5 - REUSE & CONTAINERIZE

Another productivity boosting method for teams is the ability to reuse and containerize
code. Each middle step in the data-analytics pipeline receives output from a prior stage and
provides input to the next stage. It is cumbersome to work with an entire data-analytics
pipeline as one monolith, so it is common to break it down into smaller components. It’s
easiest for other team members to reuse smaller components if they can be segmented or
containerized. One popular container technology is Docker.

Some steps in the data-analytics pipeline are messy and complicated. For example, one
operation might call a custom tool, run a python script, use FTP and other specialized
logic. This operation might be hard to set up (because it requires a specific set of tools)
and difficult to create (because it requires a specific skill set). This scenario is another
common use case for creating a container. Once the code is placed in a container, it is
much easier to use by other programmers who aren’t familiar with the custom tools inside
the container but know how to use the container’s external interfaces. It is also easier to
deploy that code to each environment.

30 • The DataOps Cookbook

STEP 6 - PARAMETERIZE YOUR PROCESSING
There are cases when the data-analytic pipeline needs to be flexible enough to incorporate
different run-time conditions. Which version of the raw data should be used? Is the data
directed to production or testing? Should records be filtered according to some criterion
(such as private health care data)? Should a specific set of processing steps in the workflow
be included or not? To increase development velocity, these options need to be built into the
pipeline. A robust pipeline design will allow the engineer or analyst to invoke or specify these
options using parameters. In software development, a parameter is some information (e.g. a
name, a number, an option) that is passed to a program that affects the way that it executes.
If the data-analytic pipeline is designed with the right flexibility, it will be ready to accommo-
date different run-time circumstances.

For example, imagine a pharmaceutical company that obtains prescription data from a
3rd party company. The data is incomplete, so the data producer uses algorithms to fill in
those gaps. In the course of improving their product, the data producer develops a different
algorithm to the fill in the gaps. The data has the same shape (rows and columns), but certain
fields are modified using the new algorithm. With the correct built-in parameters, an engi-
neer or analyst can easily build a parallel data mart with the new algorithm and have both
the old and new versions accessible through a parameter change.

The DataOps Cookbook • 31

 STEP 7: WORK WITHOUT FEAR OR HEROISM
Many data analytics professionals live in fear. In data analytics there are two common ways
to be professionally embarrassed (or get fired):
• Allow poor quality data to reach users
• Deploy changes that break production systems

Data engineers, scientists and analysts spend an excessive amount of time and energy
working to avoid these disastrous scenarios. They attempt “heroism” — working weekends.
They do a lot of hoping and praying. They devise creative ways to avoid overcommitting. The
problem is that heroic efforts are eventually overcome by circumstances. Without the right
controls in place, a problem will slip through and bring the company’s critical analytics to a halt.

The DataOps enterprise puts the right set of tools and processes in place to enable data and
new analytics to be deployed with a high level of quality. When an organization implements
DataOps, engineers, scientists and analysts can relax because quality is assured. They can
Work Without Fear or Heroism. DataOps accomplishes this by optimizing two key workflows.

The Value Pipeline

Data analytics seeks to extract value from data. We call this the Value Pipeline. The diagram
below shows the Value Pipeline progressing horizontally from left to right. Data enters the
pipeline and moves into production processing. Production is generally a series of stages:
access, transforms, models, visualizations, and reports. When data exits the pipeline, in the
form of useful analytics, value is created for the organization. DataOps utilizes toolchain
workflow automation to optimize operational efficiency in the Value Pipeline. Data in the
Value Pipeline is updated on a continuous basis, but code is kept constant. Step 2 in the
seven steps of implementing DataOps — using version control — serves as the foundation
for controlling the code deployed.

As mentioned above, the worst possible outcome is for poor quality data to enter the Value
Pipeline. DataOps prevents this by implementing data tests (step 1). Inspired by the statisti-
cal process control in a manufacturing workflow, data tests ensure that data values lay within
an acceptable statistical range. Data tests validate data values at the inputs and outputs of
each processing stage in the pipeline. For example, a US phone number should be ten digits.
Any other value is incorrect or requires normalization.

Once data tests are in place, they work 24x7 to guarantee the integrity of the Value Pipeline.
Quality becomes literally built in. If anomalous data flows through the pipeline, the data tests
catch it and take action — in most cases this means firing off an alert to the data analytics
team who can then investigate. The tests can even, in the spirit of auto manufacturing, “stop
the line.” Statistical process control eliminates the need to worry about what might happen.
With the right data tests in place, the data analytics team can Work Without Fear or Heroism.
This frees DataOps engineers to focus on their other major responsibility — the Innovation
Pipeline.

32 • The DataOps Cookbook

 Figure 6: The value pipeline

The Innovation Pipeline

The Innovation Pipeline seeks to improve analytics by implementing new ideas that yield
analytic insights. As the diagram illustrates, a new feature undergoes development before
it can be deployed to production systems. The Innovation Pipeline creates a feedback
loop. Innovation spurs new questions and ideas for enhanced analytics. This requires more
development leading to additional insight and innovation. During the development of new
features, code changes, but data is kept constant. Keeping data static prevents changes in
data from being falsely attributed to the impact of the new algorithm. A fixed data set can be
set-up when creating a development environment — step 4 in the seven steps of implement-
ing DataOps.

DataOps implements continuous deployment of new ideas by automating the workflow for
building and deploying new analytics. It reduces the overall cycle time of turning ideas into
innovation. While doing this, the development team must avoid introducing new analytics
that break production. The DataOps enterprise uses logic tests (step 1) to validate new code
before it is deployed. Logic tests ensure that data matches business assumptions. For exam-
ple, a field that identifies a customer should match an existing entry in a customer dimension
table. A mismatch should trigger some type of follow-up.

With logic tests in place, the development pipeline can be automated for continuous deploy-
ment, simplifying the release of new enhancements and enabling the data analytics team to
focus on the next valuable feature. With DataOps the dev team can deploy without worrying
about breaking the production systems — they can Work Without Fear or Heroism. This is a
key characteristic of a fulfilled, productive team.

The Value-Innovation Pipeline

In real world data analytics, the Value Pipeline and
Innovation Pipeline are not separate. The same
team is often responsible for both. The same assets
are leveraged. Events in one affect the other.
The two workflows are shown combined into the
Value-Innovation Pipeline in the figure below. The
Value-Innovation Pipeline captures the interplay
between development and production and between
data and code. DataOps breaks down this barri-
Figure 7: The innovation pipeline

The DataOps Cookbook • 33

 er so that cycle time, quality and creativity can all be improved. The DataOps enterprise
masters the orchestration of data to production and the deployment of new features both
while maintaining impeccable quality. Reduced cycle time enables DataOps engineers to
impact the organization in highly visible ways. Improved quality enables the team to move
forward with confidence. DataOps speeds the extraction of value from data and improves
the velocity of new development while ensuring the quality of data and code in production
systems. With confidence in the Value-Innovation pipeline that stems from DataOps, the
data analytics team avoids the anxiety and over-caution that characterizes a non-DataOps
enterprise. Work Without Fear or Heroism!

Figure 8: The Value and Innovation Pipelines illustrate how new analytics are
introduced into data operations.

34 • The DataOps Cookbook

DataOps is NOT Just DevOps for Data
One common misconception about DataOps is that it is just DevOps applied to data analyt-
ics. While a little semantically misleading, the name “DataOps” has one positive attribute.
It communicates that data analytics can achieve what software development attained with
DevOps. That is to say, DataOps can yield an order of magnitude improvement in quality
and cycle time when data teams utilize new tools and methodologies. The specific ways that
DataOps achieves these gains reflect the unique people, processes and tools characteristic
of data teams (versus software development teams using DevOps). Here’s our in-depth take
on both the pronounced and subtle differences between DataOps and DevOps.

Figure 9: DevOps is often depicted as an infinite loop, while DataOps is

illustrated as intersecting Value and Innovation Pipelines

The Intellectual Heritage of DataOps

DevOps is an approach to software development that accelerates the build lifecycle
(formerly known as release engineering) using automation. DevOps focuses on continuous
integration and continuous delivery of software by leveraging on-demand IT resources (infra-
structure as code) and by automating integration, test and deployment of code. This merging
of software development and IT operations (“DEVelopment” and “OPerationS”) reduces time
to deployment, decreases time to market, minimizes defects, and shortens the time required
to resolve issues.

Using DevOps, leading companies have been able to reduce their software release cycle time
from months to (literally) seconds. This has enabled them to grow and lead in fast-paced,
emerging markets. Companies like Google, Amazon and many others now release software
many times per day. By improving the quality and cycle time of code releases, DevOps de-
serves a lot of credit for these companies’ success.

Optimizing code builds and delivery is only one piece of the larger puzzle for data analyt-
ics. DataOps seeks to reduce the end-to-end cycle time of data analytics, from the origin
of ideas to the literal creation of charts, graphs and models that create value. The data
lifecycle relies upon people in addition to tools. For DataOps to be effective, it must manage
collaboration and innovation. To this end, DataOps introduces Agile Development into data
analytics so that data teams and users work together more efficiently and effectively.

In Agile Development, the data team publishes new or updated analytics in short increments
called “sprints.” With innovation occurring in rapid intervals, the team can continuously re-
assess its priorities and more easily adapt to evolving requirements. This type of responsive-

The DataOps Cookbook • 35

 ness is impossible using a Waterfall project management methodology which locks a team
into a long development cycle with one “big-bang” deliverable at the end.

Studies show that Agile software development projects complete faster and with fewer
defects when Agile Development replaces the traditional Waterfall sequential methodol-
ogy. The Agile methodology is particularly effective in environments where requirements
are quickly evolving  — a situation well known to data analytics professionals. In a DataOps
setting, Agile methods enable organizations to respond quickly to customer requirements
and accelerate time to value.

Figure 10: DataOps has evolved from lean manufacturing and

software methodologies.

Agile development and DevOps add significant value to data analytics, but there is one more
major component to DataOps. Whereas Agile and DevOps relate to analytics development
and deployment, data analytics also manages and orchestrates a data pipeline. Data con-
tinuously enters on one side of the pipeline, progresses through a series of steps and exits
in the form of reports, models and views. The data pipeline is the “operations” side of data
analytics. It is helpful to conceptualize the data pipeline as a manufacturing line where quali-
ty, efficiency, constraints and uptime must be managed. To fully embrace this manufacturing
mindset, we call this pipeline the “data factory.”

In DataOps, the flow of data through operations is an important area of focus. DataOps
orchestrates, monitors and manages the data factory. One particularly powerful lean-man-
ufacturing tool is statistical process control (SPC). SPC measures and monitors data and
operational characteristics of the data pipeline, ensuring that statistics remain within accept-
able ranges. When SPC is applied to data analytics, it leads to remarkable improvements in

36 • The DataOps Cookbook

efficiency, quality and transparency. With SPC in place, the data flowing through the opera-
tional system is verified to be working. If an anomaly occurs, the data analytics team will be
the first to know, through an automated alert.

While the name “DataOps” implies that it borrows most heavily from DevOps, it is all three
of these methodologies - Agile, DevOps and statistical process control — that comprise the
intellectual heritage of DataOps. Agile governs analytics development, DevOps optimizes
code verification, builds and delivery of new analytics and SPC orchestrates and monitors
the data factory. Figure 10 illustrates how Agile, DevOps and statistical process control flow
into DataOps.

You can view DataOps in the context of a century-long evolution of ideas that improve how
people manage complex systems. It started with pioneers like Demming and statistical pro-
cess control — gradually these ideas crossed into the technology space in the form of Agile,
DevOps and now, DataOps.

DevOps vs. DataOps — The Human Factor

As mentioned above, DataOps is as much about managing people as it is about tools. One subtle
difference between DataOps and DevOps relates to the needs and preferences of stakeholders.

Figure 11: DataOps and DevOps users have different mindsets

DevOps was created to serve the needs of software developers. Dev engineers love coding
and embrace technology. The requirement to learn a new language or deploy a new tool
is an opportunity, not a hassle. They take a professional interest in all the minute details of
code creation, integration and deployment. DevOps embraces complexity.

DataOps users are often the opposite of that. They are data scientists or analysts who are
focused on building and deploying models and visualizations. Scientists and analysts are
typically not as technically savvy as engineers. They focus on domain expertise. They are
interested in getting models to be more predictive or deciding how to best visually render
data. The technology used to create these models and visualizations is just a means to an
end. Data professionals are happiest using one or two tools — anything beyond that adds un-
welcome complexity. In extreme cases, the complexity grows beyond their ability to manage
it. DataOps accepts that data professionals live in a multi-tool, heterogeneous world and it
seeks to make that world more manageable for them.

The DataOps Cookbook • 37

 DevOps vs. DataOps - Process Differences
We can begin to understand the unique complexity facing data professionals by looking at
data analytics development and lifecycle processes. We find that data analytics professionals
face challenges both similar and unique relative to software developers.

The DevOps lifecycle is commonly illus-

trated using a diagram in the shape of an
infinite symbol — See Figure 12. The end
of the cycle (“plan”) feeds back to the be-
ginning (“create”), and the process iterates
indefinitely.

The DataOps lifecycle shares these

iterative properties, but an important
difference is that DataOps consists of two
Figure 12: The DevOps lifecycle is
active and intersecting pipelines (Figure
often depicted as an infinite loop
13). The data factory, described above, is
one pipeline. The other pipeline governs
how the data factory is updated — the creation and deployment of new analytics into the
data pipeline.

The data factory takes raw data sources as input and through a series of orchestrated steps
produces analytic insights that create “value” for the organization. We call this the “Value
Pipeline.” DataOps automates orchestration and, using SPC, monitors the quality of data
flowing through the Value Pipeline.

The “Innovation Pipeline” is the process by which new analytic ideas are introduced into the
Value Pipeline. The Innovation Pipeline conceptually resembles a DevOps development pro-
cess, but upon closer examination, several factors make the DataOps development process more
challenging than DevOps. Figure 13 shows a simplified view of the Value and Innovation Pipelines.

Figure 13: The DataOps lifecycle — the Value and Innovation Pipelines

38 • The DataOps Cookbook

DevOps vs. DataOps — Development and Deployment Processes
DataOps builds upon the DevOps development model. As shown in Figure 14, the DevOps
process flow includes a series of steps that are common to software development projects:
• Develop – create/modify an application
• Build – assemble application components
• Test – verify the application in a test environment
• Deploy – transition code into production
• Run – execute the application

DevOps introduces two foundational concepts: Continuous Integration (CI) and Continuous
Deployment (CD). CI continuously builds, integrates and tests new code in a development
environment. Build and test are automated so they can occur rapidly and repeatedly. This
allows issues to be identified and resolved quickly. Figure 14 illustrates how CI encompasses
the build and test process stages of DevOps.

Figure 14: Comparing the DataOps and DevOps processes

CD is an automated approach to deploying or delivering software. Once an application

passes all qualification tests, DevOps deploys it into production. Together CI and CD resolve
the main constraint hampering Agile development. Before DevOps, Agile created a rapid
succession of updates and innovations that would stall in a manual integration and deploy-
ment process. With automated CI and CD, DevOps has enabled companies to update their
software many times per day.

The Duality of Orchestration in DataOps

It’s important to note that “orchestration” occurs twice in the DataOps process shown in Fig-
ure 14. As we explained above, DataOps orchestrates the data factory (the Value Pipeline).
The data factory consists of a pipeline process with many steps. Imagine a complex directed
acyclic graph (DAG). The “orchestrator” could be a software entity which controls the execu-
tion of the steps, traverses the DAG, and handles exceptions. For example, the orchestrator
might create containers, invoke runtime processes with context-sensitive parameters, trans-
fer data from stage to stage, and “monitor” pipeline execution. Orchestration of the data
factory is the second “orchestration” in the DataOps process in Figure 15.

The DataOps Cookbook • 39

 Figure 15: DataOps orchestrates the data factory.

As noted above, the Innovation Pipeline has a representative copy of the data pipeline
which is used to test and verify new analytics before deployment into production. This is
the orchestration that occurs in conjunction with “testing” and prior to “deployment” of new
analytics — as shown in Figure 16.

Orchestration occurs in both the Value and Innovation Pipelines. Similarly, testing fulfills a
dual role in DataOps.

Figure 16: DataOps orchestration controls the numerous tools that access, transform,
model, visualize and report data

The Duality of Testing in DataOps

Tests in DataOps have a role in both the Value and Innovation Pipelines. In the Value
Pipeline, tests monitor the data values flowing through the data factory to catch anomalies
or flag data values outside statistical norms. In the Innovation Pipeline, tests validate new
analytics before deploying them.

40 • The DataOps Cookbook

In DataOps, tests target either data or code. Figure 17 below illustrates this concept. Data
that flows through the Value Pipeline is variable and subject to statistical process control
and monitoring. Tests target the data which is continuously changing. Analytics in the Value
Pipeline, on the other hand, are fixed and change only using a formal release process. In the
Value Pipeline, analytics are revision controlled to minimize any disruptions in service that
could affect the data factory.

In the Innovation Pipeline code is variable and data is fixed. The analytics are revised and
updated until complete. Once the sandbox (analytics development environment) is set-up,
the data doesn’t usually change. In the Innovation Pipeline, tests target the code (analytics),
not the data. All tests must pass before promoting (merging) new code into production. A
good test suite serves as an automated form of impact analysis that runs on any and every
code change before deployment.

Some tests are aimed at both data and code. For example, a test that makes sure that a
database has the right number of rows helps your data and code work together. Ultimately
both data tests and code tests need to come together in an integrated pipeline as shown
in Figure 13. DataOps enables code and data tests to work together so all around quality
remains high.

Figure 17: In DataOps, analytics quality is a function of data and code testing

DataOps Complexity — Sandbox Management

When an engineer joins a software development team, one of their first steps is to create a
“sandbox.” A sandbox is an isolated development environment where the engineer can write
and test new application features, without impacting teammates who are developing other
features in parallel. Sandbox creation in software development is typically straightforward
— the engineer usually receives a bunch of scripts from teammates and can configure a sand-
box in a day or two. This is the typical mindset of a team using DevOps.

The DataOps Cookbook • 41

 Sandboxes in data analytics are often more challenging from a tools and data perspective.
First of all, data teams collectively tend to use many more tools than typical software dev
teams. There are literally thousands of tools, languages and vendors for data engineering,
data science, BI, data visualization, and governance. Without the centralization that is char-
acteristic of most software development teams, data teams tend to naturally diverge with
different tools and data islands scattered across the enterprise.

Figure 18: A “sandbox” is an isolated development environment where the data

professional can write and test new analytics without impacting teammates.

DataOps Complexity - Test Data Management

In order to create a dev environment for analytics, you have to create a copy of the data
factory. This requires the data professional to replicate data which may have security,
governance or licensing restrictions. It may be impractical or expensive to copy the entire
data set, so some thought and care is required to construct a representative data set. Once
a multi-terabyte data set is sampled or filtered, it may have to be cleaned or redacted (have
sensitive information removed). The data also requires infrastructure which may not be easy
to replicate due to technical obstacles or license restrictions.

Figure 19: The concept of test data management is a first order problem in DataOps.

The concept of test data management is a first order problem in DataOps whereas in most
DevOps environments, it is an afterthought. To accelerate analytics development, DataOps
has to automate the creation of development environments with the needed data, software,
hardware and libraries so innovation keeps pace with Agile iterations.

42 • The DataOps Cookbook

DataOps Connects the Organization in Two Ways
DevOps strives to help development and operations (information technology) teams work
together in an integrated fashion. In DataOps, this concept is depicted in Figure 20. The
development team are the analysts, scientists, engineers, architects and others who create
data warehouses and analytics.

In data analytics, the operations team supports and monitors the data pipeline. This can be
IT, but it also includes customers — the users who create and consume analytics. DataOps
brings these groups together so they can work together more closely.

Figure 20: DataOps combines data analytics development and data operations

Freedom vs. Centralization

DataOps also brings the organization together across another dimension. A great deal of
data analytics development occurs in remote corners of the enterprise, close to business
units, using self-service tools like Tableau, Alteryx, or Excel. These local teams, engaged in
decentralized, distributed analytics creation play an essential role in delivering innovation to
users. Empowering these pockets of creativity maintains the enterprise’s competitiveness,
but frankly, a lack of top-down control can lead to unmanaged chaos.

Centralizing analytics development under the control of one group, such as IT, enables the
organization to standardize metrics, control data quality, enforce security and governance,
and eliminate islands of
data. The issue is that too
much centralization chokes
creativity.

One important benefit of

DataOps is its ability to har-
monize the back-and-forth
between the decentralized
and centralized development
of data analytics — the ten-
sion between centralization
and freedom. In a DataOps
Figure 21: DataOps brings together centralized and
enterprise, new analytics
distributed development

The DataOps Cookbook • 43

 originate and undergo refinement in the local pockets of innovation. When an idea proves
useful or is worthy of wider distribution, it is promoted to a centralized development group
who can more efficiently and robustly implement it at scale.

DataOps brings localized and centralized development together enabling organizations to

reap the efficiencies of centralization while preserving localized development — the tip of the
innovation spear. DataOps brings the enterprise together across two dimensions as shown in
Figure 22 —  development/operations as well as distributed/centralized development.

Figure 22: DataOps brings teams together across two dimensions  —  develop-
ment/operations as well as distributed/centralized development.

DataOps brings three cycles of innovation between core groups in the organization: central-
ized production teams, centralized data engineering/analytics/science/governance develop-
ment teams, and groups using self-service tools distributed into the lines business closest to
the customer. Figure 23 shows the interlocking cycles of innovation.

Figure 23: DataOps brings three cycles of innovation between

production, central data, and self-service teams.

44 • The DataOps Cookbook

Enterprise Example - Data Analytics Lifecycle Complexity
Having examined the DataOps development process at a high level, let’s look at the devel-
opment lifecycle in the enterprise context. Figure 24 illustrates the complexity of analytics
progression from inception to production. Analytics are first created and developed by an
individual and then merged into a team project. After completing unit acceptance testing
(UAT), analytics move into production. The goal of DataOps is to create analytics in the
individual development environment, advance into production, receive feedback from users
and then continuously improve through further iterations. This can be challenging due to the
differences in personnel, tools, code, versions, manual procedures/automation, hardware,
operating systems/libraries and target data. The columns in Figure 24 show the varied char-
acteristics for each of these four environments.

The challenge of pushing analytics into production across these four quite different envi-
ronments is daunting without DataOps. It requires a patchwork of manual operations and
scripts that are in themselves complex to manage. Human processes are error-prone so data
professionals compensate by working long hours, mistakenly relying on hope and heroism for
success. All of this results in unnecessary complexity, confusion and a great deal of wasted
time and energy. Slow progression through the lifecycle shown in Figure 24 coupled with
high-severity errors finding their way into production can leave a data analytics team little
time for innovation.

Figure 24: Data Analytics Development Lifecycle Complexities

Implementing DataOps
DataOps simplifies the complexity of data analytics creation and operations. It aligns data
analytics development with user priorities. It streamlines and automates the analytics
development lifecycle — from the creation of sandboxes to deployment. DataOps controls
and monitors the data factory so data quality remains high, keeping the data team focused
on adding value.

The DataOps Cookbook • 45

 You can get started with DataOps by implementing the seven steps in the last section. You
can also adopt a DataOps Platform which will support DataOps methods within the context
of your existing tools and infrastructure.

A DataOps Platform automates the steps and processes that comprise DataOps: sandbox
management, orchestration, monitoring, testing, deployment, the data factory, dashboards,
Agile, and more. A DataOps Platform is built for data professionals with the goal of simpli-
fying all of the tools, steps and processes that they need into an easy-to-use, configurable,
end-to-end system. This high degree of automation eliminates a great deal of manual work,
freeing up the team to create new and innovative analytics that maximize the value of an
organization’s data.

46 • The DataOps Cookbook

DataOps Resolves the Struggle Between
Centralization and Freedom in Analytics
Freedom and employee empowerment are essential to innovation, but a lack of top-down
control leads to chaos. Self-service tools enable data analysts to create new analytics very
quickly, but they can drift in different directions. Imagine a team of analysts building reports
that tally sales figures and each come up with a different result. One approach includes drop
shipments and sales from distributors/subsidiaries. Another report might consist of product
sales, but not services. These different approaches each have their use case, but from a man-
ager’s perspective inconsistency creates the appearance of inaccuracy. You can’t establish a
shared reality when everyone has different numbers.

Some managers respond to this challenge by centralizing analytics. With data and analytics
under the control of one group, such as IT, you can standardize metrics, control data quality,
enforce security and governance, and eliminate islands of data. All worthy endeavors, how-
ever forcing analytic updates through a heavyweight IT development process is a sure way to
stifle innovation. It is one of the reasons that some companies take three months to deploy
ten lines of SQL into production. Analytics have to be able to evolve and iterate quickly to
keep up with user demands and fast-paced markets. Managers instinctively understand that
data analytics teams must be free to innovate. The fast-growing self-service tools market
(Tableau, Looker, etc.) addresses this market.

Centralizing analytics brings it under control but granting analysts free reign is necessary to
stay competitive. How do you balance the need for centralization and freedom? How do you
empower your analysts to be innovative without drowning in the chaos and inconsistency
that a lack of centralized control inevitably produces? Visit any modern enterprise, and you
will find this challenge playing out repeatedly in budget discussions and hiring decisions. You
might say, it is a struggle between centralization and freedom.

The DataOps Cookbook • 47

 Figure 25: Data Supply, Data Engineering and Data Analysis work together in a
supply chain to fulfill the analytics requirements of customers (business users).

The Analytics Supply Chain

DataOps processes and tools offer you a way to harmonize these opposing forces, empow-
ering data analysts, while exerting a measured amount of centralization and control on your
end-to-end process. To explore these ideas further, we need to review the structure of the
analytics supply chain, and how the various roles relate to each other. The analytics organi-
zation in a DataOps enterprise consists of three essential job functions: data analysts, data
engineers and data suppliers. You can think of the three roles as groups forming a supply
chain. Data suppliers extract data for data engineers who create targeted data sets. Data
analysts consume these data sets and generate analytics for business use cases. As figure 25
shows, the three functions work together in an interlinked fashion with data and analytics
flowing to the right and feedback and requirements flowing to the left. Each group focuses
on its immediate customer (its right neighbor), but together they share the mission of deliv-
ering analytic insights to business users. While they share a common underlying mission, the
three groups operate in different business contexts.

Figure 26: Data suppliers, engineers and analysts use different cycle times driven
mainly by their tools, methods and proximity to demanding users.

48 • The DataOps Cookbook

Data Analysts
Data analysts directly support business users who work in fast-paced markets, which contin-
uously evolve. A successful analyst finds ways to respond quickly to user requests. If new an-
alytics are needed, the analyst pulls that together. New charts and graphs, updates, changes
to calculated fields, integrations of new data sets — top-performing analysts do whatever it
takes to address user requirements.

Analysts choose tools and processes oriented toward this business context. They use
powerful, self-service tools, such as Tableau, Alteryx, and Excel, to quickly create or iterate
on charts, graphs, and dashboards. They organize their work into daily sprints (figure 26), so
they can deliver value regularly and receive feedback from users immediately. Agile tools like
Jira are an excellent way to manage the productivity of analyst daily sprints.

The data analyst is the tip of the innovation spear. Organizations must give data analysts
maximum freedom to experiment. There are a lot more data in the world than companies
can analyze. Not everything can be placed in data warehouses. Not all data should be opera-
tionalized. Companies need data analysts to play around with different data sets to establish
what is predictive and relevant.

When Freedom is Not Free

As their body of analytics grows, data analysts can get bogged down in non-value-add tasks.
Self-service tools do not include mechanisms that promote and enable reuse. The data
analyst may end up copying a calculated field into many reports and then have to manage
changes to that algorithm manually. It becomes a revision control nightmare. New data sets
are often integrated using labor-intensive and error-prone manual steps. This becomes a
heavy burden on data analysts, consuming more than 75% of their time. Help and support
from data engineering addresses these challenges for the team.

Some companies mistakenly ask data engineering to create data sets for every idea. It is best
to let analysts lead on implementing new analytic ideas and proving them out before consid-
ering how data engineering can help. For example, consider the following:

1. Have the new analytics proven to be useful to many users?

2. Have calculations been reused in many reports/charts/dashboards?
3. Can automation reduce duplication of effort or manual integration errors?
4. Does data quality need improvement?

By this standard, the organization focuses its data engineering resources on those items
that give the most bang for the buck. Keep in mind that when analytics are moved into a data
warehouse, some of the benefits of centralization come at the expense of reduced freedom
— it is slower to update a data warehouse than a Tableau worksheet. It’s important to wait
until analytics have earned the right to make this transition. The value created by centralizing
must outweigh the restriction of freedom.

The DataOps Cookbook • 49

 Data Engineers
Data engineers choose tools and processes which facilitate their objectives — to produce
quality-checked data sets like data lakes, data warehouses and data marts for Data Analysts.
These data sets include field calculations that analysts can leverage, promoting reuse. Data
engineers can also automate data integration and other processes, minimizing manual steps
for the data analyst. With the added centralization offered by data engineering, analysts can
mitigate non-value add tasks and keep innovating rapidly.

Data engineers utilize programmable platforms such as AWS, S3, EC2 and Redshift. These
tools require programming in a high-level language and offer greater potential functionality
than the tools used by analysts. The relative complexity of the tools and scope of projects in
data engineering fit best in weekly Agile iterations (figure 26). DataOps platforms like Dat-
aKitchen enable the data engineer to streamline the quality control, orchestration and data
operations aspects of their duties. With automated support for agile development, impact
analysis, and data quality, the data engineer can stay focused on creating and improving data
sets for analysts.

After data sets have proven their value, it’s worth considering whether the benefits of fur-
ther centralization outweigh the cost of a further reduction in freedom. Data suppliers fulfill
the function of greater centralization by providing data sources or data extracts for data
engineering.

There are several reasons that a project may have earned the right to transition to data
suppliers. Analytics may provide functionality that executives wish to make available to the
entire corporation, not just one business unit. It could also be a case of standardization — for
example, the company wants to standardize on an algorithm for calculating market share. In
another example, perhaps data engineering has implemented quality control on a data set
and wishes to achieve efficiencies by pushing this functionality upstream to the data suppli-
er. A data supplier may be an external third party or an internal group, such as an IT master
data management (MDM) team.

Master Data Management

In MDM, an enterprise links all of its critical data to a common reference. For example,
in the pharmaceutical industry, there are 20-30 public data sets that describe physicians,
payers and products. The initial merging and mastery of the data sets may, and in some cases
should, be performed by the data engineering team, for example, for the purpose of business
analytics.

After the usefulness of the mastered data is established, the company might decide that the
data has broader uses. They may want the customer or partner list to be available for a portal
or tied into a billing system. This use case requires a higher standard of accuracy for the mas-
tered data than was necessary for the analytic data warehouse. It’s appropriate at this point
to consider moving the MDM to a data supplier, such as a corporate IT team, who are adept
at tackling more extensive, development initiatives. Put another way, initial data mastery
may have been good enough for analytic insights, but data must be perfect when it is being
used in a billing system. The data supplier takes the MDM to the next level.

50 • The DataOps Cookbook

 Table 2: Data Supply, Data Engineering and Data Analysis prefer different tools
and methods which influence their optimal cycle times

Data Suppliers
Projects transitioned to data suppliers tend to incorporate more process and tool complexity
than those in data engineering, leading to a more extended iteration period of one or more
months (figure 26). These projects use tools such as RDBS, MDM, Salesforce, Excel, sFTP,
etc., and rely upon waterfall project management and MS Project tracking. Table 2 summariz-
es tools and processes preferred by data suppliers as contrasted with engineers and analysts.

The Centralization-Freedom Spectrum

Data analysts, data engineers and data suppliers sit on a centralization-innovation (freedom)
spectrum with data suppliers offering the most centralization capabilities, data analysts
producing the fastest innovation and data engineering serving as a transition space in the
center. The characteristic strengths and weaknesses of each of these groups are strongly
influenced by their daily, weekly and monthly iteration periods. By organizing the various
groups in this way, the company has access to the full spectrum of development; fast innova-
tion, medium-scope development and longer-term, complex projects. For every need, the
project has a home. The principle of granting analysts as much freedom as possible ensures
that the innovation engine continues to turn. Waiting for analytics to stabilize and earn the
right to be transitioned ensures that centralization adds value where it should and doesn’t
infringe on freedom and creativity.

Figure 27: Data Suppliers, Data Engineers and Data Analysts sit on a spectrum of
centralization and innovation/freedom.

The DataOps Cookbook • 51

 The DataOps Framework for Innovation Management
The supply chain model that we have discussed illustrates how DataOps processes and tools
help enterprises empower data analysts while exerting a measured amount of centralization
and control on the end-to-end data pipeline. The special sauce behind DataOps is auto-
mated orchestration, continuous deployment and testing/monitoring of the data pipeline.
DataOps reduces manual effort, enforces data quality and streamlines the orchestration of
the data pipeline.

Figure 28: Tests verify that data rows, facts and dimensions match business logic
throughout the data pipeline

For example, Figure 28 shows how the DataOps platform orchestrates, tests and monitors
every step of the data operations pipeline, freeing up the team from significant manual
effort. The test verifies that the quantity of data matches business logic at each stage of the
data pipeline. If a problem occurs at any point in the pipeline, the analytics team is alerted and can
resolve the issue before it develops into an emergency. With 24x7 monitoring of the data pipeline,
the team can rest easy and focus on customer requirements for new/updated analytics.

52 • The DataOps Cookbook

 Figure 29: DataOps enables a superpower mindset —
make your customer awesome

Giving Your Team Superpowers

The goal of DataOps is to minimize overhead and free data engineers and analysts to focus
on delivering analytics to customers. With automation, DataOps enables data professionals
to improve their productivity by an order of magnitude. A single data engineer supports ten
data analysts, who in turn support 100 business professionals. It’s like gaining superpowers.
Data suppliers, engineers and analysts then concentrate their energy on their primary objec-
tive — making customers awesome.

The DataOps Cookbook • 53

54 • The DataOps Cookbook
The DataOps Cookbook • 55
 DataOps Chicken Wings
by Nick Bracy

INGREDIENTS AND TOOLS

• 1/4 cup of soy sauce
• 3 tbsp of sesame oil
• 1 1/2 tbsp siracha sauce
• Freshly grated ginger and garlic
(I use about 2-3 tsp of minced garlic — 2 cloves)
• Ginger (can be overpowering, a grated piece of ginger that is about a square inch
is good, about the size of the top part of your thumb—from knuckle to the top of
your thumb)
• 1 tbsp of rice wine vinegar
• A generous tbsp of organic honey or agave
• Black pepper to taste
• 1/2 cup of ketchup (optional)

INSTRUCTIONS
1. Place chicken drums and wings in a large zip-lock back, add marinade, seal
zip-lock bag, mix contents of the bag around gently (you don’t want to acci-
dentally open the bag and marinate your kitchen floor or counter), make sure
your chicken is well coated inside the bag.
2. Refrigerate your chicken in the marinade for 8-24 hours (You can also just
cook them right away if you don’t have the time)
3. Best slow cooked for 5-6 hours in a crockpot or on 225 degrees in a conven-
tional oven—use all the contents in the bag. (If you don’t have that kind of time,
bake at 400 degrees Fahrenheit.) 3.5 lbs. of chicken should bake for 55-60
minutes; 4.5 lbs. of chicken requires 60-65 minutes.

Recipe inspired by Joe DeFran

56 • The DataOps Cookbook

 DataOps for the Chief
Data Officer

Prove Your Awesomeness with Data:

The CDO DataOps Dashboard
Do you deserve a promotion? You may think to yourself that your work is exceptional. Could
you prove it?

As a Chief Data Officer (CDO) or Chief Analytics Officer (CAO), you serve as an advocate for
the benefits of data-driven decision making. Yet, many CDO’s are surprisingly unanalytical
about the activities relating to their own department. Why not use analytics to shine a light
on yourself?

Internal analytics could help you pinpoint areas of concern or provide a big-picture assess-
ment of the state of the analytics team. We call this set of analytics the CDO Dashboard. If
you are as good as you think you are, the CDO Dashboard will show how simply awesome
you are at what you do. You might find it helpful to share this information with your boss
when discussing the data analytics department and your plans to take it to the next level.
Below are some reports that you might consider including in your CDO dashboard:

The DataOps Cookbook • 57

 BURNDOWN CHART
The burndown chart graphically represents the completion of backlog tasks over time. It
shows whether a team is on schedule and sheds light on the productivity achieved in each
development iteration. It can also show a team’s accuracy in forecasting its own schedule.

Figure 30: Sample Burndown chart

VELOCITY CHART
The velocity chart shows the amount of work completed during each sprint — it displays how
much work the team is doing week in and week out. This chart can illustrate how improved
processes and indirect investments (training, tools, process improvements, …) increase veloc-
ity over time.

Figure 31: Sample Velocity chart

58 • The DataOps Cookbook

TORNADO REPORT
The Tornado Report is a stacked bar chart that displays a weekly representation of the oper-
ational impact of production issues and the time required to resolve them. The Tornado Re-
port provides an easy way to see how issues impacted projects and development resources.

Figure 32: Sample Tornado report

DATA ARRIVAL REPORT

A large organization might receive hundreds of data sets from suppliers and each one could
represent dozens of files. All of the data has to arrive error-free in order to, for example,
build the critical Friday afternoon report. The Data Arrival report tracks how vendors per-
form relative to their respective service level agreements (SLA).

The DataOps Cookbook • 59

 The Data Arrival report enables you to track data suppliers and quickly spot delivery issues.
Any partner that causes repeated delays can be targeted for coaching and management. The
Tornado Report mentioned above can help quantify how much time is spent managing these
issues in order to articulate impact. These numbers are quite useful when coaching a peer
organization or vendor to improve its quality.

Figure 33: Sample Data Arrival Report

TEST COVERAGE AND INVENTORY

The Test Coverage and Inventory Reports show the degree of test coverage of the data
analytics pipeline. It shows the percent of tables and data covered by tests and how test
coverage improves over time. The report can also provide details on each test. In a DataOps
enterprise, results from tests run on the production pipeline are linked to real-time alerts. If
a process fails with an error, the analytics team can troubleshoot the problem by examining
test coverage before or after the point of interest.

60 • The DataOps Cookbook

STATISTICAL PROCESS CONTROLS
The data analytics pipeline is a complex process with steps often too numerous to be mon-
itored manually. Statistical Process Control (SPC) allows the data analytics team to monitor
the pipeline end-to-end from a big-picture perspective, ensuring that everything is operating
as expected.

Figure 34: Sample Statistical Process Controls

NET PROMOTER SCORE

A Net Promoter Score is a customer satisfaction metric that gauges a team’s effectiveness.
For a data team, this is often a survey of internal users who are served by analytics. The Net
Promoter Score can show that the data analytics team is effective at meeting the needs of
its internal customer constituency or that
satisfaction is improving.

One of the main goals of analytics is to

improve decision-making. The CDO Dash-
board puts information at the fingertips of
executives, so they have a complete picture
of what is happening in the data analytics
domain. When it’s time to review perfor-
mance, the CDO Dashboard can help you
show others that the analytics department Figure 35: Net Promoter Score
is a well-oiled machine. Now, about that
promotion…

The DataOps Cookbook • 61

 Surviving Your Second Year as CDO
As the Chief Data Officer (or Chief Analytics Officer) of your company, you manage a team,
oversee a budget and a hold a mandate to set priorities and lead organizational change. The
bad news is that everything that could possibly go wrong from a security, governance and
risk perspective is your responsibility. If you do a perfect job, then no one on the manage-
ment team ever hears your name.

The average tenure of a CDO or CAO is about 2.5 years. In our conversations with data and
analytics executives, we find that CDOs and CAOs often fall short of expectations because
they fail to add sufficient value in an acceptable time frame. If you are a CDO looking to sur-
vive well beyond year two, we recommend avoiding three common traps that we have seen
ensnare even the best and brightest.

1) THE TRAP OF DATA DEFENSE

Babson College professor Tom Davenport classifies data and analytics projects as either
defense or offense. Data defense seeks to resolve issues, improve efficiency or mitigate risks.
Data quality, security, privacy, governance, compliance — these are all critically important
endeavors, but they are in essence, just enabling activities. You could think of data defense
as providing indirect value.

Data offense expands top-line revenue, builds the brand, grows the company and in general
puts points on the board. Using data analytics to help marketing and sales is data offense.
Companies may acknowledge the importance of defense, but they care passionately about
offense and focus on it daily. Data offense provides the organization with direct value and it
is what gets CDOs and CAOs promoted.

The challenge for a CDO is that data defense is hard. A company’s shortcomings in gover-
nance, security, privacy, or compliance may be glaringly obvious. In some cases, new regu-
lations like GDPR (General Data Protection Regulation, EU 2016/679) demand immediate

62 • The DataOps Cookbook

action. Data defense has a way of consuming more than its fair share of attention and staff.
If not put in perspective, data defense is a trap that can divert the CDO’s attention and
resources away from offensive activities that create value for the organization.

2) THE TRAP OF DEFERRED VALUE

Projects that implement new platforms and solutions can require months, if not years, of
integration and oversight. If conceived as a waterfall project, with a big-bang deliverable at
the end, these projects produce little to no value until they are complete. We call this the
trap of deferred value, and it is possibly the main reason that many CDOs never make it past
year three of their tenure.

In a fast-paced, competitive environment, an 18-month integration project can seem like the
remote future. Also, success is uncertain until you deliver. Your C-level peers know that big
software integration projects fail half the time. Projects frequently turn out to be more com-
plex than anticipated, and they often miss the mark. For example, you may have thought you
needed ten new capabilities, but your internal customers only really require seven, and two
of them were not on your original list. The issue is that you won’t know which seven features
are critical until around the time of your second annual performance review and by then it
might be too late to right the ship.

Figure 36: CDO’s often make the dual mistake of (1) focusing too much on delivering
indirect value (governance, security, privacy, or compliance, …) and (2) using a wa-
terfall project methodology which defers the delivery of value to the end of a long
project cycle. In the case shown, it takes several months to deliver direct value

The DataOps Cookbook • 63

 3) THE TRAP OF DATA VALUATION
Industry analysts and the media have long touted the strategic value of data. Following the
advice of analysts, a CDO may decide to embark on a project to quantify the monetary value
of the company’s data. This seems like a worthy endeavor that some say should attain a high
level of visibility.

A data valuation project can take months of effort and consumes the attention of the CDO
and her staff on what is essentially an internally-focused, intellectual exercise. In the end,
you have a beautiful PowerPoint presentation with detailed spreadsheets to back it up. Your
data has tremendous value that can and should be carried on the balance sheet. You tell every-
one all about it — why don’t they care?

Don’t confuse data valuation with data offense. Knowing the theoretical value of data is not
data offense. While data valuation may be useful and important in certain cases, it is often a
distraction. All of the time and resources devoted to creating and populating the valuation
model could have been spent on higher value-add activities.

DIRECT VS. INDIRECT VALUE

Investments in data analytics can create value either directly or indirectly. Sales growth is an
example of direct value. Indirect value lays the foundation for future growth and productiv-
ity. In both cases, value is delivered either quickly or in a longer time frame. One common
mistake is to focus too heavily on indirect-value, long-term projects.

Figure 37: DataOps uses an iterative product management methodology (Agile develop-
ment) that enables the CDO to rapidly deliver direct value (growing the top line).

64 • The DataOps Cookbook

That’s not to say indirect or long-term projects don’t have their place. They can be import-
ant and worthwhile. For example, a CDO may wisely invest in employee training or building
technical infrastructure. It’s essential to create the right mix, investing in enough indirect val-
ue-creators to support long-term growth and enough direct-value and short-term projects to
maintain a high level of visibility.

DATAOPS ACCELERATES VALUE CREATION

The trick is to reorganize the data and analytics teams to be responsive and adaptable to
the needs of internal customers and users. DataOps can help here. DataOps subscribes to
an Agile, iterative approach. Deliver something of value in a few weeks and build on that in
successive intervals. DataOps combines Agile with DevOps and lean manufacturing methods
to provide a data and analytics team with the processes and tools needed to accelerate the
value-creation cycle. Raise a glass to year two!

The DataOps Cookbook • 65

 CAOs and CDOs: Earn the Trust of your CEO
One of the greatest challenges in analytics is earning the trust of your organization’s CEO
and management team. A lot of people in the business world make decisions with their gut.
They rely on experience and intuition, but many companies would prefer to depend upon
data. You cannot walk through an airport these days and not see some version of an adver-
tisement saying we are the company who is going to help you to be more data-driven.

People do not always trust data. Imagine you are an executive and an employee walks into
your office and shows you charts and graphs that contradict strongly held assumptions
about your business. A lot of managers in this situation favor their own instincts. Data-ana-
lytics professionals, who tend to be doers, not talkers, are sometimes unable to convince an
organization to trust its data.

BUILDING TRUST IN THE DATA

We’ve discovered that the best way for data-analytics professionals to build trust with their
management team is to deliver value consistently, quickly and accurately. To accomplish this,
you need to create and publish analytics in a new way. We call this new approach DataOps.
DataOps is a combination of tools and methods, which streamline the development of new
analytics while ensuring impeccable data quality. DataOps helps shorten the cycle time for
producing analytic value and innovation, addressing some of the fundamental challenges
that prevent organizations from trusting their data.

EARN TRUST BY DELIVERING A JOURNEY OF VALUE

DataOps uses the Agile Development methodology to create valuable new analytics for the
business or organization. Agile accepts that people don’t necessarily know what they want
until they see it. The data-analytics team delivers new analytics in a short time frame and
receives immediate feedback from users. This tight feedback loop steers the development of
new analytics to the features that are most valuable to the business. Users aren’t expected
to take a leap of faith. They are gradually introduced to their own data and direct the journey

66 • The DataOps Cookbook

of each future improvement in analytics through their feedback. This feedback both im-
proves the analytics and draws them into the process as an active and invested stakeholder.
When users grow to appreciate the value provided, it is time to operationalize the analytics
and deliver them on a continuous basis.

EARN TRUST BY DELIVERING QUICKLY

In DataOps automated tests enable the data-analytics team to deploy new analytics quickly
and with confidence. Minimizing the cycle time of new analytics is critical to earning the
trust of users. The relevance of a question asked, at any given moment, decays rapidly as the
situation facing the organization quickly evolves. Customers and prospects do not stand still.
If analytics take too long, frustration builds, and the answer to a question might be delivered
long after the question has ceased to be relevant.

DataOps relies upon the data lake design pattern, which enables data analytics teams to up-
date schemas and transforms quickly to create new data warehouses that promptly address
pressing business questions. DataOps incorporates the continuous-deployment methodolo-
gy that is characteristic of DevOps. This reduces the cycle time of new analytics by an order
of magnitude. When users get used to quick answers, it builds trust in the data-analytics
team, and stimulates the type of creativity and teamwork that leads to breakthroughs.

EARN TRUST BY DELIVERING ACCURATELY

Trust is tough to earn and easy to squander. If bad data makes its way through your data
pipeline, the users might not ever again trust the data. DataOps tests and monitors business
logic and data validity by testing data at each stage of the data-analytics pipeline. We liken
this testing to the statistical process control used in lean manufacturing. The tests can start
simple, but over time they are expanded in breadth until they become a formidable check
on quality. If a problem occurs with an internal or external dataset, or at any processing
stage, the data-analytics team will be alerted immediately by the automated tests. DataOps
protects the integrity of the data, so the data itself is worthy of user trust.

THE BENEFIT OF TRUSTED ANALYTICS

Earning your organization’s trust makes the job of the data-analytics team a lot easier, but
much more is at stake. Companies that don’t trust their data will be outcompeted in the mar-
ketplace. Managers will make decisions based on instinct, past experience or preconceived
notions. In cases like this, managers sometimes develop different versions of reality and can’t
agree on the facts, let alone strategic plans.

A company that trusts its data develops a unified view of reality and can formulate a
shared vision of how to achieve its goals. Data-driven companies deliver higher growth and
ultimately higher valuations than their peers. As a CAO or CDO, leading the organization
to become more data-driven is your mission. DataOps makes that easier by helping the da-
ta-analytics team deliver quickly and robustly, creating value that is recognized and trusted
by the organization.

The DataOps Cookbook • 67

 The Four Stage Journey to Analytics Excellence
First, we walk, then we run. The same is true in data analytics. In our many discussions, we
have encountered companies that are just starting out with data analytics and others with
substantial organizations handling petabytes of data. Everyone that we meet is somewhere
along this spectrum of maturity. We’ve found that just because an enterprise’s data analytics
organization is large does not mean that it is excellent. In fact, the flaws in a process or meth-
odology become particularly noticeable when a team grows beyond the initial stages.

We view every company as being somewhere on a journey towards achieving excellence. In

our experience, the journey is divided into four stages. That said, some get there faster by
taking a shortcut. We’ll discuss the four-stage journey and the shortcut to excellence below.

STAGE 1 - DATA DESERT

Companies generate data from a variety of enterprise applications. This data can help organi-
zations gain a better understanding of customers, products, and markets. If your company is
not reaping value from your data, then you live in a data desert. In a data desert, the data is
underutilized or lays dormant. Like a mineral resource that remains in the ground, the data
could have enormous potential, but without data analytics that potential goes unrealized.

This situation could have implications for the company’s future. What if competitors have
devised a way to use data analytics to garner a competitive advantage? Without a compre-
hensive data strategy, a company risks missing the market.

STAGE 2- BOUTIQUE ANALYTICS

Some organizations are engaged in analytics but do so in a decentralized fashion or on a
small scale. Some enterprises are just getting started in analytics. Whether or not a person
has programming skills, it is possible to do a fair amount of analytics using everyday tools like
spreadsheets. One can accomplish even more using data visualization software. We call this
Boutique Analytics. In a boutique shop, data analytics professionals are akin to artisans.

68 • The DataOps Cookbook

Boutique Analytics tend to be ad hoc or create one-off reports that answer questions posed
by a manager. For example, a global enterprise may wish to know how much of its reve-
nue it derives from one customer. Data is exported from CRMs or operations systems and
pulled into a spreadsheet for analysis. The term Boutique Analytics may make it sound small
in scale, but some large enterprises are known to rely solely upon this approach. A large
enterprise might run weekly reports exporting sales data into a flat file. The global sales and
marketing team can then easily manipulate the data in a spreadsheet. The sharing of data
using flat files can be used to complement an enterprise’s operational analytics.

There is nothing inherently wrong with Boutique Analytics. It is a great way to explore the
best ways to deliver value based on data. The eventual goal should be to operationalize the
data and deliver that value on a regular basis. This can be time-consuming and error-prone if
executed manually.

STAGE 3- WATERFALL ANALYTICS

If an analytics initiative is successful and the team grows, a company will eventually begin
to manage analytics more formally. Companies usually have a deeply entrenched project
management culture based upon the methodology used by their research and development
teams. Often project management is based on the Waterfall method so it is natural for these
organizations to implement Waterfall Analytics.

In the Waterfall world, development cycles are long and rigidly controlled. Projects pass
through a set of sequential phases: architecture, design, test, deployment, and maintenance.
Changes in the project plan at any stage cause modifications to the scope, schedule or
budget of the project. As a result, Waterfall projects are resistant to change. This is wholly
appropriate when you are building a bridge or bringing a new drug to market, but in the field
of data analytics, changes in requirements occur on a continuous basis. Teams that use Wa-
terfall analytics often struggle with development cycle times that are much longer than their
users expect and demand. Waterfall analytics also tends to be labor intensive, which makes
every aspect of the process slow and susceptible to error. Most data-analytics teams today
are in the Waterfall analytics stage and are often unaware that there is a better way.

Table 3: Four Stages of Analytics

The DataOps Cookbook • 69

 STAGE 4 - DATAOPS ANALYTICS
DataOps is a new approach to data analytics, which is superior to Waterfall Analytics in
terms of flexibility, quality, and development cycle time. DataOps adopts key concepts from
lean manufacturing. It views data analytics as a continuously operating pipeline, which can
be automated, monitored and controlled. New analytics are created using Agile Develop-
ment, a methodology created in the software engineering field. Agile manages the devel-
opment of new analytics by delivering valuable features in short increments. This allows an
organization to quickly adapt to new requirements or change course based on the demands
of the marketplace. Analytics are deployed using the continuous deployment methodology
pioneered by DevOps. Automated orchestration replaces labor-intensive manual processes.
This means that new analytics can be published continuously, on-demand with minimal hu-
man intervention. Data quality flowing through the data analytics pipeline is monitored using
automated data and logic tests executed as part of the continuous deployment automation.
These tests are inspired by the statistical process control widely used in modern manufactur-
ing operations.

TAKE THE SHORTCUT

The mistake that many companies make is that they languish in stage 3. The better approach
is to take a shortcut, skip stage 3 entirely, and move directly to stage 4. If your organization
is already in stage 3, then it’s advantageous to advance as quickly as possible.

THE JOURNEY TO EXCELLENCE

DataOps provides the foundation for data analytics excellence. It streamlines the devel-
opment of new analytics, shortens cycle time, and automates the data-analytics pipeline,
freeing the team to focus on value-adding activities. It also controls the quality of data
flowing through the pipeline so users can trust their data. With DataOps in place, the team is
productive, responsive and efficient. They will race far ahead of competitors whose analytics
are less nimble and less impactful. DataOps shortens your journey to analytics excellence.

70 • The DataOps Cookbook

The DataOps Cookbook • 71
 DataOps Healthy Hearty Banana Oatmeal Bread
by Andrew Sadoway

INGREDIENTS AND TOOLS

• 1 1/4 cups whole wheat flour
• 1 1/4 old-fashioned oats
• 1 teaspoon baking powder
• 1/2 teaspoon baking soda
• 1/4 teaspoon salt
• 1 teaspoon cinnamon
• Dash of nutmeg (approx. 1/8 teaspoon)
• 3 medium-largeish bananas, mashed (defrosted from freezer OK)
• 1/2 cup low fat plain yogurt + 1 teaspoon vanilla
• 2 tablespoons honey, 1/3 cup brown sugar
• 1 large egg
• Splash of milk, as needed
• 3/4 cup dried cranberries
• 3/4 cup chopped walnuts + 1/4 cup chopped walnuts

INSTRUCTIONS
Preheat oven to 350. Combine dry ingredients (flour through nutmeg) in a small
bowl. In a separate bowl, mix together yogurt, vanilla, brown sugar and honey. Add
egg. Add mashed up bananas. Slowly fold dry ingredients into wet. Stir in cran-
berries and 3/4 cup walnuts gently. Pour mixture into buttered loaf pan. Sprinkle
remaining walnuts on top of loaf. Bake about 45 minutes, or until lightly browned
and knife comes out clean.

72 • The DataOps Cookbook

 DataOps for the Data
Engineer and the
Data Scientist

DataOps Puts Agility into Agile Data Warehousing

Data analytics professionals get used to being in no-win situations. Internal customers make
a simple request; for example, add a new file to the database. Users expect requests like
these to take days, yet, in many large organizations, they require months to complete. At
DataKitchen, we repeatedly hear from companies that they need to improve their cycle time
for new analytics. One approach, Agile Data Warehousing, applies Agile principles to data
warehouse projects in an attempt to speed innovation. However, many companies quickly
discover that simply implementing Scrum is not sufficient to attain results.

Imagine that you oversee a fifty-person team managing numerous large integrated databases
(DB) for a big insurance or financial services company. You have 300 terabytes (TB) of data
which you manage using a proprietary database. Between software, licensing, maintenance,
support and associated hardware, you pay $10M per year in annual fees. Even putting an-
other single CPU into production could cost hundreds of thousands of dollars.

Someday these large databases will

move to the cloud at a fraction of the
cost. New databases will be turned
on and off like light bulbs with the en-
terprise only paying for the resources
they consume. That’s a long-term
goal. In the short term, the team has
to produce results using the existing
platform.

You can’t afford separate instan-

tiations of the entire data set for
development, quality assurance (QA),
performance testing and production
so non-production machines are given

The DataOps Cookbook • 73

 subsets of the data. The necessity of provisioning physically separate hardware instantiations
is one barrier to greater Agility.

The machine environments are different and have to be managed and maintained separately.
New analytics are tested on each machine in turn — first in dev, then QA and finally produc-
tion. You may not catch every problem in dev and QA since they aren’t using the same data
and environment as production.

Running regression tests manually is time-consuming so it can’t be done often. This creates
risk whenever new code is deployed. Also, when changes are made on one machine they
have to be manually installed on the others. The steps in this procedure are detailed in a
30-page text document, which is updated by a committee through a cumbersome series of
reviews and meetings. It is a very siloed and fractured process, not to mention inefficient;
during upgrades, the DB is offline, so new work is temporarily on hold.

In our hypothetical company, the organization of the workforce is also a factor in slowing the
team’s velocity. Everyone is assigned a fixed role. Adding a table to a database involves sev-
eral discrete functions: a Data Quality person who analyzes the problem, a Schema/Architect
who designs the schema, an ETL engineer who writes the ETL, a Test Engineer that writes
tests and a Release Engineer who handles deployment. Each of these functions is performed
sequentially and requires considerable documentation and committee review before any
action is taken. Hand-off meetings mark the transition from one stage to the next.

The team wants to move faster but is prevented from doing so due to heavyweight process-
es, serialization of tasks, overhead, difficulty in coordination and lack of automation. They
need a way to increase collaboration and streamline the many inefficiencies of their current
process without having to abandon their existing tools.

HOW DATAOPS HELPS

DataOps is a new approach to data analytics that automates the orchestration of data to
production and the deployment of new features, both while maintaining impeccable quality.
DataOps does not mandate the use of any particular tool or technology, but support in the
following areas can be critical to Agile Data Warehousing in large teams, such as the one
described:

Shared Workspace – DataOps creates a shared workspace so team members have visibility
into each other’s work. This enables the team to work more collaboratively and seamlessly
outside the formal structure of the hand-off meeting. DataOps also streamlines documenta-
tion and reduces the need for formal meetings as a communication forum.

Orchestration – DataOps deploys code updates to each machine instantiation and auto-
mates the execution of tests along each stage of the data analytics pipeline. This includes
data and logic tests that validate both the production and feature deployment pipelines.
Tests are parameterized so they can run in the subset database of each particular machine
environment equally well. As the test suite improves, it grows to reflect the full breadth of
the production environment. Automated tests are run repeatedly so you can be confident
that new features have not broken old ones.

74 • The DataOps Cookbook

These tools and process changes together break down the organizational and technology
barriers that prevent the team from implementing Agile methods in data analytics. DataOps
unburdens the team from non-value-add tasks and empowers them to self-organize around
new creative initiatives. When the team is free to innovate, the continuous improvement
culture built into DataOps will begin working to reduce the cycle time of new analytics from
months to days (and less). This ultimately puts the Agility back into Agile Data Warehousing
by delivering high-quality analytics to users in a timely fashion.

The DataOps Cookbook • 75

 Speed Up Innovation with DataOps
LEVERAGING DATA LAKES, DATA WAREHOUSES AND
SCHEMAS FOR FASTER ANALYTICS
Analytics professionals often strain to make one change to their analytic pipeline per month.
DataOps increases their productivity by an order of magnitude. DataOps accelerates innova-
tion by automating and orchestrating the data analytics pipeline and speeding ideas to pro-
duction. It does this by applying Agile Development, DevOps and statistical process controls
to data analytics. This enables the DataOps Engineer to quickly respond to requests for new
analytics while guaranteeing a high level of quality. In order to understand this, it is helpful to
know a little about the role of data lakes, schemas and data warehouses in DataOps.

DATAOPS REQUIRES EASY ACCESS TO DATA

When data is moved from disparate silos into a common repository, it is much easier for
a data analytics team to work with it. The common store is called a data lake. To optimize
DataOps, it is often best to move data into a data lake using on-demand simple storage.

People often speak about data lakes as a repository for raw data. It can also be helpful to
move processed data into the data lake. There are several important advantages to using
data lakes. First and foremost, the data analytics team controls access to it. Nothing can
frustrate progress more than having to wait for access to an operational system (ERP, CRM,
MRP, …). Additionally, a data lake brings data together in one place. This makes it much eas-
ier to process. Imagine buying items at garage sales all over town and placing them in your
backyard. When you need the items, it is much easier to retrieve them from the backyard
rather than visiting each of the garage sale sites. A data lake serves as a common store for all
of the organization’s critical data. Easy, unrestricted access to data eliminates restrictions on
productivity that slow down the development of new analytics.

Note that if you put public company financial data in a data lake, everyone who has access to the
data lake is an “insider.” If you have confidential data, HIPAA data (Health Insurance Portability
and Accountability Act of 1996) or Personally identifiable information (PII) — these must be man-
aged in line with government regulations, which vary by country.

76 • The DataOps Cookbook

The structure of a data lake is designed to support efficient data access. This relates to how
data is organized and how software accesses it. A database schema establishes the relation-
ship between the entities of data.

UNDERSTANDING SCHEMAS
A database schema is a collection of tables. It dictates how the database is structured and
organized and how the various data relate to each other. Below is a schema that might be
used in a pharmaceutical-sales analytics use case. There are tables for products, payers,
period, prescribers and patients with an integer ID number for each row in each table. Each
sale recorded has been entered in the fact table with the corresponding IDs that identify the
product, payer, period, and prescriber respectively. Conceptually, the IDs are pointers into
the other tables.

Figure 38: The Schema of a Pharmaceutical-Sales Analytics System

The schema establishes the basic relationships between the data tables. A schema for an
operational system is optimized for inserts and updates. The schema for an analytics system,
like the star schema shown here, is optimized for reads, aggregations, and is easily under-
stood by people.

Suppose that you want to do analysis of patients based on their MSA (metropolitan service
area). An MSA is a metropolitan region usually clustered near a large city. For example,
Cambridge, Massachusetts is in the Greater Boston MSA. The prescriber table has a zip-code
field. You could create a zip-code-to-MSA lookup table or just add MSA as an attribute to
the patient table. Both of these are schema changes. In one case you add a table and in the
other case you add a column.

The DataOps Cookbook • 77

 TRANSFORMS CREATE DATA WAREHOUSES
The data lake provides easier access, but lacks the optimizations needed for visualizations
or modeling. For example, data often enters the data lake in the format of the source system
and not using an optimized schema that facilitates analysis. Data warehouses better address
analytic-specific requirements. For example, the data warehouse could have a schema that
supports specific visualization, modeling or other features.

You might hear the term data mart in relation to data analytics. Data marts are a streamlined
form of data warehouses. The two are conceptually very similar.

Data transforms (scripts, source code, algorithms, …) create data warehouses from data
lakes. In DataOps this process is optimized by keeping transform code in source control and
by automating the deployment of data warehouses. An automated deployment process is
significantly faster, more robust and more productive than a manual deployment process.

THE DATAOPS PIPELINE

The automation of the pipeline that transforms the schemas of data lakes, creating data
warehouses and data marts, is a key reason that DataOps is able to improve the speed and
quality of the data analytic pipeline. Without using a data lake, data is highly dispersed, and
difficult to access. Schemas of operational systems are difficult to navigate and most likely
not optimized for analytics.

DataOps moves the enterprise beyond slow, inflexible, disorganized and error-prone manual
processes. The DataOps pipeline leverages data lakes and transforms them into well-crafted
data warehouses using continuous deployment techniques. This speeds the creation and
deployment of new analytics by an order of magnitude. Additionally, the DataOps pipeline is
constantly monitored using statistical process control so the analytics team can be confident
of the quality of data flowing through the pipeline. Work Without Fear or Heroism. With
these tools and process improvements, DataOps compresses the cycle time of innovation
while ensuring the robustness of the analytic pipeline. Faster and higher quality analytics ul-
timately lead to better insights that enable an enterprise to thrive in a dynamic environment.

78 • The DataOps Cookbook

How to Inspire Code Reuse in Data Analytics
In DataOps, the data analytics team moves at lightning speed using highly optimized tools and pro-
cesses. One of the most important productivity tools is the ability to reuse and containerize code.

When we talk about reusing code, we mean reusing data analytics components. All of the
files that comprise the data analytics pipeline — scripts, source code, algorithms, html, con-
figuration files, parameter files — we think of these as code. Like other software develop-
ment, code reuse can significantly boost coding velocity.

Code reuse saves time and resources by leveraging existing tools, libraries or other code in
the extension or development of new code. If a software component has taken several
months to develop, it effectively saves the organization several months of development time
when another project reuses that component. This practice can be used to decrease projects
budgets. In other cases, code reuse makes it possible to complete projects that would have
been impossible if the team were forced to start from scratch.

Containers make code reuse much simpler. A container packages everything needed to run
a piece of software — code, runtimes, tools, libraries, configuration files — into a stand-alone
executable. Containers are somewhat like virtual machines but use fewer resources because
they do not include full operating systems. A given hardware server can run many more
containers than virtual machines.

A container eliminates the problem in which code runs on one machine, but not on another,
because of slight differences in the set-up and configuration of the two servers or software
environments. A container enables code to run the same way on every machine by auto-
mating the task of setting up and configuring a machine environment. This is one DataOps
techniques that facilitates moving code from development to production — the run-time
environment is the same for both. One popular open-source container technology is Docker.

Each step in the data-analytics pipeline is the output of the prior stage and the input to the
next stage. It is cumbersome to work with an entire data-analytics pipeline as one mono-
lith, so it is common to break it down into smaller components. On a practical level, smaller
components are much easier to reuse by other team members.

The DataOps Cookbook • 79

 Some steps in the data-analytics pipeline are messy and complicated. For example, one
operation might call a custom tool, run a python script, use FTP and other specialized logic.
This operation might be both hard to set up, because it requires a specific set of tools, and
difficult to create, because it requires a specific skill set. This scenario is another common
use case for creating a container. Once the code is placed in a container, it is much easier to
use by other programmers who aren’t familiar with the custom tools inside the container but
know how to use the container’s external interfaces. All of the complexity is embedded in-
side the container. It is also easier to deploy that code to different environments. Containers
make code reuse much more turnkey and allow developers much greater flexibility in sharing
their work with each other.

80 • The DataOps Cookbook

What Data Scientists Really Need
Kurt Cagle‘s perceptive analysis of data science titled “Why You Don’t Need Data Scientists”
explains the many reasons that data science falls short of the high expectations usually
placed on it:
• Fancy dashboards are pretty but are only as valuable as the data behind them. Data
quality often….stinks.
• Data sets are quirky and difficult to work with.
• Users/stakeholders know their business domain but little about what data can do for them.
• A multimillion-dollar initiative to rebuild the data pipeline from the ground up is gener-
ally off the table.
• The people who own the databases won’t give data scientists access.
• Everyone agrees that integrating data from disparate databases is really, really hard, but
in reality, it’s much harder than people think.

These are all excellent points and often the conversation ends here — in exasperation. We
can tell you that we have been there and have the PTSD to prove it. Fortunately, a few years
ago, we found a way out of what may seem at times like a no-win situation. We believe that the
secret to successful data science is a little about tools and a lot about people and processes.

DON’T BOIL THE OCEAN

Use Agile methods to create new analytics. Leverage infrastructure that enables teams to
work together in an Agile way. Start small and simple. Create something quickly that adds
value. Get feedback from your stakeholders. Repeat iteratively.

TESTS ARE BEST

Implement automated process controls that monitor data at every stage of your data analyt-
ics pipeline. Think of your data analytics as a lean manufacturing pipeline where the quality
of data cannot drift outside statistical and logical bounds. Let your tools work 24x7 so data
scientists can stay focused on creating analytics that add value.

The DataOps Cookbook • 81

 AUTOMATE AND ORCHESTRATE
Data Scientists spend 75% of their time doing data engineering. It’s about time that data
professionals took a page from DevOps. Automate workflow and the deployment of new
analytics. Orchestrate the end-to-end data pipeline so we stop sucking the life out of data
scientists. A single data engineer should be able to support ten data analysts and scientists,
who in turn should be supporting 100 business professionals. An automated pipeline can get
you there.

82 • The DataOps Cookbook

The DataOps Cookbook • 83
 DataOps No-Knead Bread
by Gil Benghiat

INGREDIENTS AND TOOLS

• 3 cups bread flour
• 1/4 teaspoons instant yeast
• ½ teaspoon salt
• 1 ½ cups very warm, almost hot water
• Oil for work surface (canola)

INSTRUCTIONS
1. Combine flour, yeast and salt in a large bowl and stir with your DataKitchen
spoon. Add water and stir until blended; dough will be shaggy. You may need
an extra ¼ cup of water to get all the flour to blend in. Cover bowl with plas-
tic wrap. Let dough rest at least 4 hours (12-18 hours is good too) at warm
room temperature, about 70 degrees.
2. Lightly oil a work surface and place dough on it; fold it over on itself once or
twice. Cover loosely with plastic wrap and let rest 30 minutes more. This is a
good time to turn the oven on to 425°F.
3. Put a 6-to-8-quart heavy covered pot (cast iron, enamel, Pyrex or ceramic)
in the oven as it heats. When dough is ready, carefully remove pot from
oven. Slide your hand under dough and put it into pot, seam side up. Shake
pan once or twice if dough is unevenly distributed; it will straighten out as it
bakes.
4. Cover with lid and bake 30 minutes, then remove lid and bake another 15 to
30 minutes, until loaf is beautifully browned. Cool on a rack.

NOTES
In a convection oven, cook 23 minutes with the lid on, and then 5 minutes with
the lid off.
You don’t need to pre-heat the pot. You can put the dough on a cookie sheet. The
only difference is the crust will not be as crunchy or as beautifully browned. You
can experiment with a round shape or Italian or French loaf shapes. The longer
shapes will take less time to cook.
You can also cook at a lower temperature (e.g. 350°F). In all cases, take the bread
out when the internal temperature reaches 190°F - 200°F. Use a meat thermom-
eter to check.

Derived from New York Times Speedy No-Knead Bread

84 • The DataOps Cookbook

 DataOps for
Data Quality

Disband Your Impact Review Board:

Automate Analytics Testing
Some companies take six months to write 20 lines of SQL and move it into production.

The last thing that an analytics professional wants to do is introduce a change that breaks
the system. Nobody wants to be the object of scorn, the butt of jokes, or a cautionary tale.
If that 20-line SQL change is misapplied, it can be a “career-limiting move” for an analytics
professional.

Analytics systems grow so large and complex that no single person in the company under-
stands them from end to end. A large company often institutes slow, bureaucratic proce-
dures for introducing new analytics in order to reduce fear and uncertainty. They create
a waterfall process with specific milestones. There is a lot of documentation, checks and
balances, and meetings — lots of meetings.

The DataOps Cookbook • 85

 IMPACT ANALYSIS
One of the bottlenecks in an analytics release process is called “impact analysis.” Impact
analysis gathers experts on all of the various subsystems (data feeds, databases, transforms,
data lakes/warehouses, tools, reports, …) so they can review the proverbial 20 lines of SQL
and try to anticipate if/how it will adversely impact data operations.

Imagine you are building technical systems that integrate data and do models and visu-
alizations. How does a change in one area affect other areas? In a traditional established
company, that information is locked in various people’s heads. The company may think it has
no choice but to gather these experts together in one room to discuss and analyze proposed
changes. This is called an “impact analysis meeting.” The process includes the company’s
most senior technical contributors; the backbone of data operations. Naturally, these individ-
uals are extremely busy and subject to high-priority interruptions. Sometimes it takes weeks
to gather them in one room. It can take additional weeks or months for them to approve a
change.

The impact analysis team is a critical bottleneck that slows down updates to analytics. A
DataOps approach to improving analytics cycle time adopts process optimization techniques
from the manufacturing field. In a factory environment, a small number of bottlenecks
often limit throughput. This is called the Theory of Constraints. Optimize the throughput of
bottlenecks and your end-to-end cycle time improves (check out “The Goal” by Eliyahu M.
Goldratt).

GET OUT OF YOUR HEAD

The Impact Analysis Meeting is a bottleneck because it relies upon your top technical ex-
perts — one of the most oversubscribed resources in the company. What if you could extract
all the knowledge and experience trapped in the brains of your company’s experts and code
it into a series of tests that would perform the impact analysis for you? This would give
you a quick way to test out changes to analytics without requiring bureaucratic procedures
and meetings. If the tests pass, you could deploy with confidence. No more waiting on the
impact review team. With a comprehensive test suite, you reduce reliance on the impact
analysis bottleneck and move a lot faster.

AUTOMATING IMPACT ANALYSIS

Manual testing moves the bottleneck from impact review to the testing team. Manual testing
is performed step-by-step, by a person. This tends to be expensive as it requires someone to
create an environment and run tests one at a time. It can also be prone to human error.

DataOps automates testing. Environments are spun up under machine control and test
scripts, written in advance, are executed in batch. Automated testing is much more cost-ef-
fective and reliable than manual testing, but the effectiveness of automated testing depends
on the quality and breadth of the tests. In a DataOps enterprise, members of the analytics
team spend 20% of their time writing tests. Whenever a problem is encountered, a new test
is added. New tests accompany every analytics update. The breadth and depth of the test
suite continuously grow.

86 • The DataOps Cookbook

One advantage of automated testing is that it’s easier to run so it’s executed repeatedly and
regularly. Manual testing is often too expensive and slow to run on a regular basis. To ensure
high quality, you have to be able to consistently and regularly test your data and code.

These concepts are new to many data teams, but they are well established in the software
industry. As figure 39 shows, the cycle time of software development releases has been
(and continues to be) reduced by orders of magnitude through automation and process
improvements. The automation of impact analysis can have a similar positive effect on your
organization’s analytics cycle time.

Figure 39: Software developers have reduced the cycle time for new releases
by orders of magnitude using automation and process improvements

ANALYTICS IS CODE
At this point some of you are thinking this has nothing to do with me. I am a data analyst/sci-
entist, not a coder. I am a tool expert. What I do is just a sophisticated form of configuration. This
is a common point of view in data analytics. However, it leads to a mindset that slows down
analytics cycle time.

Tools vendors have a business interest in perpetuating the myth that if you stay within the
well-defined boundaries of their tool, you are protected from the complexity of software
development. This is ill-considered.

Don’t get us wrong. We love our tools, but don’t buy into this falsehood.

The DataOps Cookbook • 87

 Figure 40: From data access to visualization to reports, there is code running at
every stage of the data operations pipeline

The $100B analytics market is divided into two segments: tools that create code and tools
that run code. The point is — data analytics is code. The data professional creates code and
must own, embrace and manage the complexity that comes along with it.

Figure 41: Tableau files are stored as XML, and can contain conditional
branches, loops and embedded code.

Figure 40 shows a data operations pipeline with code at every stage of the pipeline. Python,
SQL, R — these are all code. The tools of the trade (Informatica, Tableau, Excel, …) these too
are code. If you open an Informatica or Tableau file, it’s XML. It contains conditional branches
(if-then-else constructs), loops and you can embed Python or R in it.

Remember our 20-line SQL change that took six months to implement? The problem is that
analytics systems become so complex that they can easily break if someone makes one mis-
begotten change. The average data-analytics pipeline encompasses many tools (code genera-
tors) and runs lots of code. Between all of the code and people involved, data operations
becomes a combinatorially complex hairball of systems that could come crashing down with
one little mistake.

88 • The DataOps Cookbook

 Figure 42: Informatica presents a UI that creates ETL in an XML format that is
then converted to Java and executed on the machine.

For example, imagine that you have analytics that sorts customers into five bins based on
some conditional criterion. Deep inside your tool’s XML file is an if-then-else construct that
is responsible for sorting the customers correctly. You have numerous reports based off of
a template that contains this logic. They provide information to your business stakeholders:
top customers, middle customers, gainers, decliners, whales, profitable customers, …

There’s a team of IT engineers, database developers, data engineers, analysts and data
scientists that manage the end to end system that supports these analytics. One of these
individuals makes a change. They convert the sales volume field from an integer into a
decimal. Perhaps they convert a field that was US dollars into a different currency. Maybe
they rename a column. Everything in the analytics pipeline is so interdependent; the change
breaks all of the reports that contain the if-then-else logic upon which the original five
categories are built. All of a sudden, your five customer categories become one category, or
the wrong customers are sorted into the wrong bins. None of the dependent analytics are
correct, reports are showing incorrect data, and the VP of Sales is calling you hourly.

At an abstract level, every analytic insight produced, every deliverable, is an interconnected

chain of code modules delivering value. The data analytics pipeline is best represented by a
directed acyclic graph (DAG). For example, see Figure 43.

The DataOps Cookbook • 89

 Figure 43: The Directed Acyclic Graph (DAG) models the
steps in the data analytics pipeline

Whether you use an analytics tool like Informatica or Tableau, an Integrated Development
Environment (IDE) like Microsoft Visual Studio (Figure 44) or even a text editor like Notepad,
you are creating code. The code that you create interacts with all of the other code that
populates the DAG that represents your data pipeline.

To automate impact analysis, think of the end-to-end data pipeline holistically. Your test
suite should verify software entities on a stand-alone basis as well as how they interact.

Figure 44: Developers write SQL, Python and other code using an integrated
development environment or sometimes a simple editor like Notepad.

90 • The DataOps Cookbook

TYPES OF TESTS
The software industry has decades of experience ensuring that code behaves as expected.
Each type of test has a specific goal. If you spend any time discussing testing with your
peers, these terms are sure to come up:
• Unit Tests – testing aimed at each software component as a stand-alone entity
• Integration Tests – focus on the interaction between components to ensure that they
are interoperating correctly
• Functional Tests – verification against functional specification or user stories.
• Regression Tests – rerun every time a change is made to prove that an application is
still functioning
• Performance Tests – verify a system’s responsiveness, stability and availability under a
given workload
• Smoke Tests – quick, preliminary validation that the major system functions are opera-
tional

TESTS TARGET DATA OR CODE OR BOTH

It’s also helpful to frame the purpose and context of a test. Tests can target data or code
and run as part of the data operations pipeline. Location balance, historical balance and
statistical process controls (time balance) tests are directed at the data flowing through an
operations pipeline. The code that runs the data processing steps in the pipeline is fixed. The
code is tightly controlled and only changed via a release process. Data that moves through
operations, on the other hand, is variable. New data flows through the pipeline continuously.
As Figure 45 shows, the data operations pipeline delivers value to users. DataOps terms this
the Value Pipeline.

Figure 45: Data Operations: The Value Pipeline

The development of new analytics follows a different path, which is shown in Figure 46 as
the Innovation Pipeline. The Innovation Pipeline delivers new insights to the data operations
pipeline, regulated by the release process. To safely develop new code, the analyst needs an
isolated development environment. When creating new analytics, the developer creates an
environment analogous to the overall system. If the database is terabytes in size, the data
professional might copy it for test purposes. If the data is petabytes in size, it may make
sense to sample it; for example, take 10% of the overall data. If there are concerns about

The DataOps Cookbook • 91

 privacy or other regulations, then sensitive information is removed. Once the environment is
set up, the data typically remains stable.

Table 4: In the Value Pipeline code is fixed and data is variable. In the Innovation
Pipeline, data is fixed, and code is variable.

In the Innovation Pipeline code is variable, but data is fixed. Tests target the code, not the
data. The unit, integration, functional, performance and regression tests that were men-
tioned above are aimed at vetting new code. All tests are run before promoting (merging)
new code to production. Code changes should be managed using a version control system,
for example GIT. A good test suite serves as an automated form of impact analysis that can
be run on any and every code change before deployment.

Figure 46: New analytics are developed in

the Innovation Pipeline

Some tests are aimed at both data and code. For example, a test that makes sure that a
database has the right number of rows helps your data and code work together. Ultimately
both data tests and code tests need to come together in an integrated pipeline as shown
in Figure 47. DataOps enables code and data tests to work together so all around quality
remains high.

92 • The DataOps Cookbook

 Figure 47: Ultimately the Value and Innovation Pipelines work
together to maintain data and code quality

A unified, automated test suite that tests/monitors both production data and analytic code
is the linchpin that makes DataOps work. Robust and thorough testing removes or minimizes
the need to perform manual impact analysis, which avoids a bottleneck that slows innova-
tion. Removing constraints helps speed innovation and improve quality by minimizing analyt-
ics cycle time. With a highly optimized test process you’ll be able to expedite new analytics
into production with a high level of confidence.

20 new lines of SQL? You’ll have it right away.

The DataOps Cookbook • 93

 Build Trust Through Test Automation and Monitoring

“Trust takes years to build, seconds to break, and forever to repair.”

We recently talked to a data team in a financial services company that lost the trust of their
users. They lacked the resources to implement quality controls so bad data sometimes
leaked into user analytics. After several high-profile episodes, department heads hired their
own people to create reports. For a data-analytics team, this is the nightmare scenario, and it
could have been avoided.

Organizations trust their data when they believe it is accurate. A data team can struggle to
produce high-quality analytics when resources are limited, business logic keeps changing
and data sources have less-than-perfect quality themselves. Accurate data analytics are the
product of quality controls and sound processes.

The data team can’t spend 100% of its time checking data, but if data analysts or scientists
spend 10-20% of their time on quality, they can produce an automated testing and monitor-
ing system that does the work for them. Automated testing can work 24x7 to ensure that
bad data never reaches users, and when a mishap does occur, it helps to be able to assure
users that new tests can be written to make certain that an error never happens again. Auto-
mated testing and monitoring greatly multiplies the effort that a data team invests in quality.

94 • The DataOps Cookbook

DATA FLOW AS A PIPELINE
Think of data analytics as a manufacturing pipeline. There are inputs (data sources), process-
es (transformations) and outputs (analytics). A typical manufacturing process includes tests
at every step in the pipeline that attempt to identify problems as early as possible. As every
manufacturer knows, it is much more efficient and less expensive to catch a problem in
incoming inspection as opposed to finished goods.

Figure 48 depicts the data-analytics pipeline. In this diagram, databases are accessed and
then data is transformed in preparation for being input into models. Models output visualiza-
tions and reports that provide critical information to users.

Figure 48: Testing each stage of the data-analytic pipeline

Along the way, tests ask important questions. Are data inputs free from issues? Is business
logic correct? Are outputs consistent? As in lean manufacturing, tests are performed at every
step in the pipeline. For example, data input tests are analogous to manufacturing incoming
quality control. Figure 49 shows examples of data input, output and business logic tests.

Data input tests strive to prevent any bad data from being fed into subsequent pipeline
stages. Allowing bad data to progress through the pipeline wastes processing resources and
increases the risk of never catching an issue. It also focuses attention on the quality of data
sources, which must be actively managed — manufacturers call this supply chain management.

Data output tests verify that a pipeline stage executed correctly. Business logic tests validate
data against tried and true assumptions about the business. For example, perhaps all Europe-
an customers are assigned to a member of the Europe sales team.

Test results saved over time provide a way to check and monitor quality versus historical
levels.

The DataOps Cookbook • 95

 Figure 49: Tests validate data inputs and outputs and verify
that data is consistent with business logic.

FAILURE MODES
A disciplined data production process classifies failures according to severity level. Some
errors are fatal and require the data analytics pipeline to be stopped. In a manufacturing
setting, the most severe errors “stop the line.”

Some test failures are warnings. They require further investigation by a member of the data
analytics team. Was there a change in a data source? Or a redefinition that affects how data
is reported? A warning gives the data-an-
alytics team time to review the changes,
talk to domain experts, and find the root
cause of the anomaly.

Many test outputs will be informational.

They help the data engineer, who oversees
the pipeline, to monitor routine pipeline
activity or investigate failures.
Table 5: Actions required for
different failure modes
TYPES OF TESTS
The data team may sometimes feel that its work product is under a microscope. If the analyt-
ics look “off,” users can often tell immediately. They are experts in their own domain and will
often see problems in analytics with only a quick glance.

Finding issues before your internal customers do is critically important for the data team.
There are three basic types of tests that will help you find issues before anyone else: location
balance, historical balance and statistical process control.

96 • The DataOps Cookbook

LOCATION BALANCE TESTS
Location Balance tests ensure that data properties match business logic at each stage of
processing. For example, an application may expect 1 million rows of data to arrive via FTP.
The Location Balance test could verify that the correct quantity of data arrived initially, and
that the same quantity is present in the database, in other stages of the pipeline and finally,
in reports.

Figure 50: Location Balance Tests verify 1M rows in raw source data, and the
corresponding 1M rows / 300K facts / 700K dimension members in the database
schema, and 300K facts / 700K dimension members in a Tableau report

HISTORICAL BALANCE
Historical Balance tests compare current data to previous or expected values. These tests
rely upon historical values as a reference to determine whether data values are reasonable
(or within the range of reasonable). For example, a test can check the top fifty customers or
suppliers. Did their values unexpectedly or unreasonably go up or down relative to historical
values?

It’s not enough for analytics to be correct. Accurate analytics that “look wrong” to users raise
credibility questions. Figure 51 shows how a change in allocations of SKUs, moving from
pre-production to production, affects the sales volumes for product groups G1 and G2. You
can bet that the VP of sales will notice this change immediately and will report back that
the analytics look wrong. This is a common issue for analytics — the report is correct, but it
reflects poorly on the data team because it looks wrong to users. What has changed? When
confronted, the data-analytics team has no ready explanation. Guess who is in the hot seat.

Historical Balance tests could have alerted the data team ahead of time that product group
sales volumes had shifted unexpectedly. This would give the data-analytics team a chance to
investigate and communicate the change to users in advance. Instead of hurting credibility,
this episode could help build it by showing users that the reporting is under control and that
the data team is on top of changes that affect analytics. “Dear sales department, you may no-
tice a change in the sales volumes for G1 and G2. This is driven by a reassignment of SKUs within
the product groups.”

The DataOps Cookbook • 97

 Figure 51: It’s not enough for analytics to be correct. Accurate analytics that
“look wrong” to users raise credibility questions.

STATISTICAL PROCESS CONTROL

Lean manufacturing operations measure and monitor every aspect of their process in order
to detect issues as early as possible. These are called Time Balance tests or more commonly,
statistical process control (SPC). SPC tests repeatedly measure an aspect of the data pipeline
screening for error or warning patterns. SPC offers a critical tool for the data team to catch
failures before users see them in reports.

Figure 52: Statistical Process Control tests apply numerical

criteria to data-analytics pipeline measurements

98 • The DataOps Cookbook

NOTIFICATIONS
A complex process could have thousands of tests running continuously. When an error or
warning occurs, a person on the data team should be alerted in real-time through email, text
or a notification service like slack. This frees the data team from the distraction of having
to periodically poll test results. If and when an event takes place, they’ll be notified and can
take action.

Figure 53: Example data test result notification email

Automated tests and alerts enforce quality and greatly lessen the day-to-day burden of
monitoring the pipeline. The organization’s trust in data is built and maintained by producing
consistent, high-quality analytics that help users understand their operational environment.
That trust is critical to the success of an analytics initiative. After all, trust in the data is really
trust in the data team.

The DataOps Cookbook • 99

 How Data Analytics Professionals Can Sleep Better
LEAN MANUFACTURING SECRETS THAT YOU CAN APPLY TO DATA ANALYTICS
What could data analytics professionals possibly learn from car manufacturers? It turns
out, a lot. Automotive giant Toyota pioneered a set of methods, later folded into a disci-
pline called lean manufacturing, in which employees focus relentlessly on improving quality
and reducing non-value-add activities. This culture enabled Toyota to grow into the one of
the world’s leading car companies. The Agile and DevOps methods that have led to stellar
improvements in coding velocity are really just an example of lean manufacturing principles
applied to software development.

Conceptually, manufacturing is a pipeline process. Raw materials enter the manufacturing

floor through the stock room, flow to different work stations as work-in-progress and exit
as finished goods. In data-analytics, data progresses through a series of steps and exits in
the form of reports, models and visualizations. Each step takes an input from the previous
step, executes a complex procedure or set of instructions and creates output for the subse-
quent step. At an abstract level, the data-analytics pipeline is analogous to a manufacturing
process. Like manufacturing, data analytics executes a set of operations and attempts to
produce a consistent output at a high level of quality. In addition to lean-manufacturing-in-
spired methods like Agile and DevOps, there is one more useful tool that can be taken from
manufacturing and applied to data-analytics process improvement.

W. Edwards Deming championed statistical process control (SPC) as a method to improve

manufacturing quality. SPC uses real-time product or process measurements to monitor and
control quality during manufacturing processes. If the process measurements are maintained
within specific limits, then the manufacturing process is deemed to be functioning properly.
When SPC is applied to the data-analytics pipeline, it leads to remarkable improvements in
efficiency and quality. For example, Google executes over one hundred million automated
test scripts per day to validate any new code released by software developers. In the Google
consumer surveys group, code is deployed to customers eight minutes after a software engi-
neer finishes writing and testing it.

100 • The DataOps Cookbook

 Figure 54: Tests verify the results for each intermediate step in the analytics pipeline.

In data analytics, tests should verify that the results of each intermediate step in the
production of analytics matches expectations. Even very simple tests can be useful. For
example, a simple row-count test could catch an error in a join that inadvertently produces a
Cartesian product. Tests can also detect unexpected trends in data, which might be flagged
as warnings. Imagine that the number of customer transactions exceeds its historical average
by 50%. Perhaps that is an anomaly that upon investigation would lead to insight about
business seasonality.

Tests in data analytics can be applied to data or models either at the input or output of a
phase in the analytics pipeline. Tests can also verify business logic.

Figure 55: Tests are applied to inputs, outputs or business logic.

The data analytics pipeline is a complex process with steps often too numerous to be moni-
tored manually. SPC allows the data analytics team to monitor the pipeline end-to-end from
a big-picture perspective, ensuring that everything is operating as expected. As an automat-
ed test suite grows and matures, the quality of the analytics is assured without adding cost.
This makes it possible for the data analytics team to move quickly — enhancing analytics to
address new challenges and queries — without sacrificing quality.

The DataOps Cookbook • 101

102 • The DataOps Cookbook
The DataOps Cookbook • 103
 DataOps Trail Mix Cookies
by Mike Beaverson

INGREDIENTS AND TOOLS

• 1 cup butter, room temperature
• 1 1/4 cup brown sugar
• 1/4 cup honey
• 1 tbsp vanilla extract
• 2 large eggs
• 2 1/4 cups all purpose flour
• 1/2 tsp baking soda
• 1 tsp salt
• Dash of cinnamon
• 1 bag (8-12oz) trail mix (dark chocolate, dried fruit, candy bits, nuts are all
good choices)

Instructions
1. Preheat oven to 325F. Line a baking sheet with parchment paper.
2. In a large bowl, cream together butter and sugar until light and fluffy.
3. Beat in honey, vanilla, and both eggs, adding the eggs in one at a time.
4. In a medium bowl, whisk together flour, baking soda, cinnamon, and salt.
5. Working by hand or at a low speed, gradually incorporate flour mixture into
honey mixture.
6. Stir in trail mix.
7. Shape cookie dough into 1-inch balls and place onto prepared baking sheet,
leaving about 2 inches between each cookie to allow for the dough to
spread.
8. Bake for 12-15 minutes, until cookies are golden brown.
9. Cool for 3-4 minutes on the baking sheet, then transfer to a wire rack to cool
completely.

Makes about 4-dozen cookies

104 • The DataOps Cookbook

 DataOps and
Your Career

DataOps Engineer Will Be the Sexiest Job in Analytics

Years ago, prior to the advent of Agile Development, a friend of mine worked as a release en-
gineer. His job was to ensure a seamless build and release process for the software develop-
ment team. He designed and developed builds, scripts, installation procedures and managed
the version control and issue tracking systems. He played a mean mandolin at company
parties too.

The role of release engineer was (and still is) critical to completing a successful software
release and deployment, but as these things go, my friend was valued less than the software
developers who worked beside him. The thinking went something like this — developers
could make or break schedules and that directly contributed to the bottom line. Release
engineers, on the other hand, were never noticed, unless something went wrong. As you
might guess, in those days the job of release engineer was compensated less generously
than development engineer. Often, the best people vied for positions in development where
compensation was better.

The DataOps Cookbook • 105

 RISING FORTUNES
Today, the fortunes of release engineers have risen sharply. In companies that are imple-
menting DevOps there is no more important person than the release engineer. The job title
has been renamed DevOps engineer and it is one of the most highly compensated positions
in the field of software engineering. According to salary surveys, experienced DevOps
engineers make six figure salaries. DevOps specialists are so hard to find that firms are hiring
people without college degrees, if they have the right experience.

Whereas a release engineer used to work off in a corner tying up loose ends, the DevOps
engineer is a high-visibility role coordinating the development, test, IT and operations
functions. If a DevOps engineer is successful, the wall between development and operations
melts away and the dev team becomes more agile, efficient and responsive to the market.
This has a huge impact on the organization’s culture and ability to innovate. With so much
at stake, it makes sense to get the best person possible to fulfill the DevOps engineer role
and compensate them accordingly. When DevOps came along, the release engineer went
from fulfilling a secondary supporting role to occupying the most sought-after position in
the department. Many release engineers have successfully rebranded themselves as DevOps
engineers and significantly upgraded their careers.

DATAOPS FOR DATA ANALYTICS

A similar change, called DataOps, is transforming the roles on the data analytics team. Data-
Ops is a better way to develop and deliver analytics. It applies Agile development, DevOps
and lean manufacturing principles to data analytics producing a transformation in data-driv-
en decision making.

Data engineers, data analysts, data scientists — these are all important roles, but they will be
valued even more under DataOps. Too often, data analytics professionals are trapped into
relying upon non-scalable methods: heroism, hope or caution. DataOps offers a way out of
this no-win situation.

The capabilities unlocked by DataOps impacts everyone that uses data analytics — all the
way to the top levels of the organization. DataOps breaks down the barriers between data
analytics and operations. It makes data more easily accessible to users by redesigning the
data analytics pipeline to be more flexible and responsive. It will completely change what
people think of as possible in data analytics.

In many organizations, the DataOps engineer will be a separate role. In others, it will be a
shared function. In any case, the opportunity to have a high-visibility impact on the organi-
zation will make DataOps engineering one of the most desirable and highly compensated
functions. Like the release engineer whose career was transformed by DevOps, DataOps will
boost the fortunes of data analytics professionals. DataOps will offer select members of the
analytics team a chance to reposition their roles in a way that significantly advances their
career. If you are looking for an opportunity for growth as a DBA, ETL Engineer, BI Analyst,
or another role look into DataOps as the next step.

And watch out Data Scientist, the real sexiest job of the 21st century is DataOps Engineer.

106 • The DataOps Cookbook

Building a DataOps Team
Picture what you could accomplish if your organization had accurate and detailed informa-
tion about products, processes, customers and the market. If your company does not have
a data analytics function, you need to start one. Better yet, if data analytics is not serving as
a competitive advantage in your organization, you need to step up your game and establish a
DataOps team.

Data analytics analyzes internal and external data to create value and actionable insights.
Analytics is a positive force that is transforming organizations around the globe. It helps cure
diseases, grow businesses, serve customers better and improve operational efficiency.

In analytics there is mediocre and there is better. A typical data analytics team works slowly,
all the while living in fear of a high-visibility data quality issue. A high-performance data
analytics team rapidly produces new analytics and flexibly responds to marketplace demands
while maintaining impeccable quality. We call this a DataOps team. A DataOps team can
Work Without Fear or Heroism because they have automated controls in place to enforce
a high level of quality even as they shorten the cycle time of new analytics by an order of
magnitude. Want to upgrade your data analytics team to a DataOps team? It comes down to
roles, tools and processes.

MEET THE DATAOPS TEAM

There are four key roles in any DataOps team. Note that larger organizations will tend to
have many people in each role. Smaller companies might have one person performing multi-
ple roles. See the table down below for some key tools associated with each of the roles described
as well as alternate job titles. Most of these roles are familiar to data analytics professionals,
but DataOps adds an essential ingredient that makes the team much more productive.

DATA ENGINEER
The data engineer is a software or computer engineer that lays the groundwork for other
members of the team to perform analytics. The data engineer moves data from operation-
al systems (ERP, CRM, MRP, …) into a data lake and writes the transforms that populate
schemas in data warehouses and data marts. The data engineer also implements data tests
for quality.

The DataOps Cookbook • 107

 DATA ANALYST
The data analyst takes the data warehouses created by the data engineer and provides
analytics to stakeholders. They help summarize and synthesize massive amounts of data.
The data analyst creates visual representations of data to communicate information in a way
that leads to insights either on an ongoing basis or by responding to ad-hoc questions. Some
say that a data analyst summarizes data that reflects past performance (descriptive analytics)
while future predictions are the domain of the data scientist.

DATA SCIENTIST
Data scientists perform research and tackle open-ended questions. A data scientist has
domain expertise, which helps him or her create new algorithms and models that address
questions or solve problems.

For example, consider the inventory management system of a large retailer. The company
has a limited inventory of snow shovels, which have to be allocated among a large number of
stores. The data scientist could create an algorithm that uses weather models to predict buy-
ing patterns. When snow is forecasted for a particular region it could trigger the inventory
management system to move more snow shovels to the stores in that area.

Table 6: The DataOps Team

108 • The DataOps Cookbook

DATAOPS IS THE PROCESS AND THE TOOLS
Many data analytics teams fail because they focus on people and tools and ignore pro-
cess. This is similar to fielding a sports team with players and equipment, but no game plan
describing how everyone will work together. The game plan in data analytics is included in
something that we call DataOps.

DataOps is a combination of tools and process improvements that enable rapid-response

data analytics, at a high level of quality. Producing analytics that are responsive, flexible, con-
tinuously deployed and quality controlled requires data analytics to draw upon techniques
learned in other fields.
• Agile Development – an iterative project management methodology that completes
software projects faster and with far fewer defects.
• DevOps – a software development process that leverages on-demand IT resources and
automated test and deployment of code to eliminate the barriers between develop-
ment (Dev) and operations (Ops). DevOps reduces time to deployment, decreases time
to market, minimizes defects, and shortens the time required to resolve issues. DevOps
techniques help analytics teams break down the barriers between data and ops (DataOps).
• Lean Manufacturing – DataOps utilizes statistical process control (SPC) to monitor and
control the data analytics pipeline. When SPC is applied to data analytics, it leads to re-
markable improvements in efficiency and quality. With quality continuously monitored
and controlled, data analytics professionals can Work Without Fear or Heroism.

The process and tools enhancements described above can be implemented by anyone on
the analytics team or a new role may be created. We call this role the DataOps Engineer.

DATAOPS ENGINEER
The DataOps Engineer applies Agile Development, DevOps and statistical process controls
to data analytics. He or she orchestrates and automates the data analytics pipeline to make
it more flexible while maintaining a high level of quality. The DataOps Engineer uses tools
to break down the barriers between operations and data analytics, unlocking a high level of
productivity from the entire team.

As DataOps breaks down the barriers between data and operations, it makes data more
easily accessible to users by redesigning the data analytics pipeline to be more responsive,
efficient and robust. This new function will completely change what people think of as possi-
ble in data analytics. The opportunity to have a high-visibility impact on the organization will
make DataOps engineering one of the most desirable and highly compensated functions on
the data-analytics team.

The DataOps Cookbook • 109

110 • The DataOps Cookbook
The DataOps Cookbook • 111
 DataOps Triple Chocolate Peanut Butter Cookies
by Aarthy Kannan Adityan

INGREDIENTS AND TOOLS

• 1 cup butter
• 3/4 cup brown sugar
• 3/4 cup white sugar
• 1 1/2 teaspoon vanilla extract
• 3/4 cup chocolate peanut butter
• 2 eggs
• 2 1/3 cup flour
• 1 teaspoon baking soda
• 3/4 cup cocoa powder
• 1 cup semi-sweet chocolate chips
• 1 cup peanut butter chips

INSTRUCTIONS
1. Preheat oven to 350ºF
2. Soften butter to room temperature
3. Line a baking sheet with parchment paper
4. In a large bowl, cream together softened butter, brown sugar and white
sugar
5. Add vanilla extract, chocolate peanut butter and eggs and mix well
6. Stir in flour, baking soda and cocoa powder and combine until blended
7. Fold chocolate chips and peanut butter chips into batter
8. Scoop batter onto prepared baking sheet using a cookie or ice-cream scoop,
leaving enough space in-between for cookies to expand
9. Bake for 14-16 minutes
10. Transfer cookies to a wire rack to cool

Serving size: about 20 cookies

112 • The DataOps Cookbook

 DataOps Examples and
Case Studies

Grow Sales Using a DataOps-Powered

Customer Data Platform
Data analytics can help drive corporate growth by providing customer analytics and ulti-
mately actionable insights to the sales and marketing teams. Unfortunately, the fast-paced,
dynamic nature of sales makes it difficult for the customer-facing teams to tolerate the slow
and deliberate manner in which analytics is typically produced. In an earlier chapter, we
identified eight major challenges of data analytics:
• The Goalposts Keep Moving – Sales and marketing requirements change constantly
and the requests for new analytics never cease.
• Data Lives in Silos – Data is collected in separate operational systems and typically,
none of these systems talk to each other.
• Data Formats are not Optimized – Data in operational systems is usually not structured
in a way that lends itself to the efficient creation of analytics.
• Data Errors – Data will eventually contain errors, which can be difficult to resolve quickly.
• Bad Data Ruins Good Reports – When data errors work their way through the data
pipeline into published analytics, internal stakeholders can become dissatisfied. These
errors also harm the hard-won trust in the analytics team.
• Data Pipeline Maintenance Never Ends – Every new or updated data source, schema
enhancement, analytics improvement or other change triggers an update to the data
pipeline. These updates may be consuming 80% of your team’s time.
• Manual Process Fatigue – Manual procedures for data integration, cleansing, transfor-
mation, quality assurance and deployment of new analytics are error-prone, time-con-
suming and tedious.
• The Trap of “Hope and Heroism” – To cope with the above challenges, data profession-
als work long hours, make changes (without proper testing) and “hope” for the best or
just retreat into a posture of over-caution in which projects just execute more slowly.

The DataOps Cookbook • 113

 OVERCOMING THE EIGHT CHALLENGES
If you have managed an analytics team for any period of time, you have likely encountered
these and similar challenges. However, you don’t have to accept the status quo. It is possible
to implement processes and methodologies that address these challenges and enable your
data-analytics team to improve their productivity by an order of magnitude while achieving
a higher level of data quality. In this new approach to customer and market analytics, the
data-analytics team executes at previously unimaginable speed, efficiency and quality:

Rapid-Response Analytics – The sales and marketing team will continue to demand a
never-ending stream of new and changing requirements, but the data-analytics team will
delight your sales and marketing colleagues with rapid responses to their requests. New
analytics will inspire new questions that will, in turn, drive new requirements for analytics.
The feedback loop between analytics and sales/marketing will iterate so quickly that it will
infuse excitement and creativity throughout the organization. This will lead to breakthroughs
that vault the company to a leadership position in its markets.

Data Under Your Control – Data from all of the various internal and external sources will be
integrated into a consolidated database that is under the control of the data-analytics team.
Your team will have complete access to it at all times, and they will manage it independently
of IT, using their preferred tools. With data under its control, the data-analytics team can
modify the format and architecture of data to meet its own operational requirements.

Impeccable Data Quality – As data flows through the data-analytics pipeline, it will pass
through tests and filters that ensure that it meets quality guidelines. Data will be monitored
for anomalies 24x7, preventing bad data from ever reaching sales and marketing analytics.
You’ll have a dashboard providing visibility into your data pipeline with metrics that delineate
problematic data sources or other issues. When an issue occurs, the system alerts the appro-
priate member of your team who can then fix the problem before it ever receives visibility.
As the manager of the data-analytics team, you’ll spend far less time in uncomfortable meet-
ings discussing issues and anomalies related to analytics.

114 • The DataOps Cookbook

Automated Efficiency – Data feeds and new analytics will be deployed using automation,
freeing the data-analytics team from tedious manual processes. The analytics team will be
able to focus on its highest priorities — creating new analytics for sales and marketing that
create value for the company.

The processes, methodologies and tools required to realize these efficiencies combine two
powerful ideas: The Customer Data Platform (CDP) and a revolutionary new approach to
analytics called DataOps. Below we’ll explain how you can implement your own Data-
Ops-powered CDP that improves both your analytics cycle time and data-pipeline quality
by 10X or more.

CUSTOMER DATA PLATFORM

A Customer Data Platform (CDP) provides sales and marketing with a unified view of all cus-
tomer-related data whether internal or external, in a single integrated database. Once setup,
a CDP enables the analytics team to create and manage customer data themselves, without
reliance upon resources from IT or other departments. This helps sales and marketing better
leverage the company’s valuable data while responding to market demands quickly and
proactively. The figure below shows how a CDP consolidates data from numerous databases.
Each operational database becomes a data source that continuously feeds a copy of its data
into a centralized CDP database.

Figure 56: The Customer Data Platform consolidates data from operational
systems to provide a unified customer view for sales and marketing.

DATAOPS
A CDP is a step in the right direction, but it won’t provide much improvement in team
productivity if the team relies on cumbersome processes and procedures to create analytics.
DataOps is a set of methodologies and tools that will help you optimize the processes by

The DataOps Cookbook • 115

 which you create analytics, manage the data-analytics pipeline and automatically deploy new
analytics and data. DataOps rests on three foundational principles:

Agile Development – DataOps utilizes a methodology called Agile Development to minimize

the cycle time for new data analytics. Studies show that software development projects
using Agile complete significantly faster and with far fewer defects.

DevOps – In DataOps, new analytics production is automated and monitored. Automated

tests verify new analytics before publishing them to sales and marketing users. This allows
the analysts to focus less on the mechanics of deploying analytics and more on the creation
of new insights that address sales and marketing requests. In the software development
domain, the automated deployment of code is called DevOps. Prominent software industry
leaders use DevOps to publish software updates many times per second while assuring
quality. DataOps incorporates DevOps methods and principles to publish new analytics and
data in an automated fashion.

Statistical Process Control – DataOps employs a methodology called Statistical Process

Control (SPC) to assure data quality using end-to-end data pipeline automation and quality
controls. SPC is a lean manufacturing method that institutes continuous testing on data
flowing from sources to users, ensuring that data stays within statistical limits and remains
consistent with business logic. SPC monitors data and verifies it 24x7. If an anomaly occurs,
SPC notifies the data-analytics team via an automated alert. This reduces the operational
burden on team members while improving data quality and reliability. Also, the quantity of
data and the number of data sources can more easily scale independently of the size of the
data engineering team.

When implemented in concert, Agile, DevOps and SPC take the productivity of data-analyt-
ics professionals to a whole new level. DataOps will help you get the most out of your data,
human resources and integrated CDP database.

116 • The DataOps Cookbook

Achieving Growth Targets by Implementing
a DataOps-Powered Customer Data Platform
As the leader of a data-analytics
organization, your mission is to
utilize data from operational systems The Eight Challenges of Data Analytics
and other sources to create insights 1. The Goalposts Keep Moving
that help the organization achieve its 2. Data Lives in Silos
growth targets. Figure 57 provides 3. Data Formats are not Optimized
a conceptual view of this flow. The 4. Data Errors
dark boxes show the domain that is 5. Bad Data Ruins Good Reports
under the control of the analytics
6. Data Pipeline Maintenance Never Ends
leader. It includes people, tools,
7. Manual Process Fatigue
technologies and processes that to-
8. The Trap of Hope and Heroism
gether comprise the DataOps-pow-
ered CDP.

The DataOps Cookbook • 117

 Figure 57: The resources under data-analytics control
that leverage data to meet business objectives

DATA ANALYSTS AND THEIR TOOLS

The Data Analyst works to satisfy the needs of the sales and marketing department by
continually delivering insights. The data analyst creates visual representations of data to
communicate information in a way that leads to insights either on an ongoing basis or by
responding to ad-hoc questions. With a DataOps-powered CDP, data analysts work with
autonomy and speed, drawing analytics from CDP data. Analysts use tools like Tableau and
Alteryx to create these insights and independently promote their investigative work into
production deliverables as needed.

Every resource, technology and tool in the data-analytics organization exists to support the
data analyst’s ability to serve Sales and Marketing. This also applies to Data Scientists who
also deliver insights directly to Sales and Marketing colleagues.

DATA ENGINEERS AND THEIR TOOLS AND DATABASES

The Data Engineer works with the IT department and all data source providers to institute
automated processes that move data from various data sources into a trusted, integrated
CDP database under the complete control of the data-analytics team. The CDP database
may include a data lake, which provides revision history, easy access, control and error
recovery.

The engineer writes transforms that operate on the data lake, creating data warehouses and
data marts used by data analysts and scientists. The data engineer also implements tests that
monitor data at every point along the data-analytics pipeline assuring a high level of quality.

The data engineer lays the groundwork for other members of the team to perform analytics
without having to be operations experts. With a dedicated data engineering function,
DataOps provides a high level of service and responsiveness to the data-analytics team.

118 • The DataOps Cookbook

DATA OPERATIONS
The DataOps-powered CDP provides automated support for the creation, monitoring and
management of the end-to-end data pipeline. This includes stewardship of every aspect of
the journey from data sources to reporting. Statistical Process Control (SPC) data-quality
tests monitor each stage of the automated pipeline, alerting data engineering when data fails
to meet statistical controls or match business logic.

With tests monitoring each stage of the automated data pipeline, DataOps can produce a
dashboard showing the status of the pipeline. The DataOps dashboard provides a high-level
overview of the end-to-end data pipeline. Is any data failing quality tests? What are the error
rates? Which are the troublesome data sources? With this information at his or her finger-
tips, the Data Engineer can proactively improve the data pipeline to increase robustness. In
the event of a high-severity data anomaly, an alert is sent to the Data Engineer who can take
steps to protect production analytics and work to resolve the error. If the anomaly relates to
a data supplier, data engineering can work with the vendor to drive the issue to resolution.
Workarounds and data patches can be implemented as needed with information in release
notes for users. In many cases, errors are resolved without the users (or the organization’s
management) ever being aware of any problem.

AGILE TEAM MANAGEMENT

The Agile methodology governs the creation of new analytics, producing a steady stream of
valuable innovations and improvements to analytic insights in short increments of time. Agile
is particularly effective in environments where requirements are quickly evolving  —  a situa-
tion all too familiar to data-analytics professionals. Agile development is not only a method;
it is also a philosophy and a mindset. Developers collaborate with sales/marketing custom-
ers, respond to change, measure progress through “delivered analytics,” release frequently,
seek feedback on releases, and adjust behavior to become more effective.

DATAOPS PLATFORM
The various methodologies, processes, people (and their tools) and the CDP analytics data-
base are tied together cohesively using a technical environment called a DataOps Platform.
The DataOps Platform includes support for:
• Agile project management
• Deployment of new analytics
• Execution of the data pipeline (orchestration)
• Integration of all tools and platforms
• Management of development and production environments
• Source-code version control
• Testing and monitoring of data quality
• Data Operations reporting and dashboards

The high degree of automation offered by DataOps eliminates a great deal of work that has
traditionally been done manually. This frees up the team to create new analytics requested
by stakeholder partners.

The DataOps Cookbook • 119

 A DataOps Platform is not a one-size-fits-all tool. It is the central application that coordi-
nates the various tools that drive your orchestration, testing, deployment, model deploy-
ment, development-environment management, change management, and data integration.
You can create your own DataOps Platform from scratch, although partnering with a supplier
can reduce time to market. Working with a partner also gives you the option of treating the
entire CDP and data pipeline as a managed service, which can be initially outsourced and
then partly or entirely taken over by internal resources at a later time.

DATAKITCHEN DATAOPS-CDP SOLUTION

An enterprise can outsource data engineering, databases, data operations, Agile team
management and the DataOps Technology Platform to gain efficiencies. DataKitchen offers
a DataOps Technology Platform as well as managed services for each of the gray boxes in
Figure 58. In essence, DataKitchen offers all aspects of the DataOps-powered CDP except
for data analysis and data science, which rely upon vertical market expertise and close col-
laboration with sales and marketing.

The enterprise can also outsource the functions shown initially but insource them at a later
date. Once set-up, the DataOps Platform can be easily and seamlessly transitioned to an
internal team.

Customer Data Platforms promise to drive sales and improve the customer experience by
unifying customer data from numerous disjointed operational systems. As a leader of the
analytics team, you can take control of sales and marketing data by implementing efficient
analytics-creation and deployment processes using a DataOps-powered CDP. A DataOps
platform makes analytics responsive and robust. This enables your data analysts and scien-
tists to rise above the bits and bytes of data operations and focus on new analytics that help
the organization achieve its goals.

Figure 58: DataKitchen DataOps-Powered CDP and Managed Services

120 • The DataOps Cookbook

How a Mixed Martial Arts Fighter Would
Approach Data Analytics
By James Royster, Director, Commercial Analytics, Celgene
Mixed Martial Arts (MMA) combines striking, wrestling and other fighting techniques into a
unified sport. Every martial art and fighting technique has its strengths and strategic advan-
tages. Boxing is known for punching but also provides footwork, guard position and head
movement. Wrestling relies upon takedowns. Karate features striking techniques such as
kicking. MMA is a hybrid of all of these (and many more) drawing upon each mode of combat
as needed for a given competitive situation. If an MMA athlete competed against a boxer or
karate expert, the mixed martial artist would clearly have an unfair advantage. MMA’s real
strength is its versatility and its ability to absorb new methods.

DataOps is the mixed martial arts of data analytics. It is a hybrid of Agile Development,
DevOps and the statistical process controls drawn from lean manufacturing. Like MMA, the
strength of DataOps is its readiness to evolve and incorporate new techniques that improve
the quality, reliability, and flexibility of the data analytics pipeline. DataOps gives data ana-
lytics professionals an unfair advantage over those who are doing things the old way — using
hope, heroism or just going slowly in order to cope with the rapidly changing requirements
of the competitive marketplace.

Agile development has revolutionized the speed of software development over the past
twenty years. Before Agile, development teams spent long periods of time developing
specifications that would be obsolete long before deployment. Agile breaks down software
development into small increments, which are defined and implemented quickly. This allows
a development team to become much more responsive to customer requirements and ulti-
mately accelerates time to market.

The DataOps Cookbook • 121

 Data analytics shares much in common with software development. Conceptually, the data
analytics pipeline takes raw data, passes it through a series of steps and turns it into action-
able information. Files, such as scripts, code, algorithms, configuration files, and many others,
drive each processing stage. These files, taken as a whole, are essentially just code. As a
coding endeavor, data analytics has the opportunity to improve implementation speeds by
an order of magnitude using techniques like Agile development. DevOps offers an additional
opportunity for improvement.

The difficulty of procuring and provisioning physical IT resources has often hampered data
analytics. In the software development domain, leading-edge companies are turning to
DevOps, which utilizes cloud resources instead of on-site servers and storage. This allows
developers to procure and provision IT resources nearly instantly and with much greater
control over the run-time environment. This improves flexibility and yields another order of
magnitude improvement in the speed of deploying features to the user base.

DataOps also incorporates lean manufacturing techniques into data analytics through the
use of statistical process controls. In manufacturing, tests are used to monitor and improve
the quality of factory-floor processes. In DataOps, tests are used to verify the inputs,
business logic, and outputs at each stage of the data analytics pipeline. The data analytics
professional adds a test each time a change is made. The suite of tests grows over time
until it eventually becomes quite substantial. The tests validate the quality and integrity of
a new release when a feature set is released to the user base. Tests allow the data analytics
professional to quickly verify a release, substantially reducing the amount of time spent on
deploying updates.

Statistical process controls also monitor data, alerting the data team to an unexpected vari-
ance. This may require updates to the business logic built into the tests, or it might lead data
scientists down new paths of inquiry or experimentation. The test alerts can be a starting
point for creative discovery.

The combination of Agile development, DevOps, and statistical process controls gives Data-
Ops the strategic tools to reduce time to insight, improve the quality of analytics, promote
reuse and refactoring and lower the marginal cost of asking the next business question.
Like mixed martial arts, DataOps draws its effectiveness from an eclectic mix of tools and
techniques drawn from other fields and domains. Individually, each of these techniques is
valuable, but together they form an effective new approach, which can take your data analyt-
ics to the next level. 

122 • The DataOps Cookbook

Reinvent Marketing Automation with
the DataKitchen DataOps Platform
A global pharmaceutical giant sought to drive top-line growth by modernizing its marketing
operations. The project included a migration to Salesforce Marketing Cloud, integrations
with numerous internal and third-party data sources, and a continuous flow of data. The
plan initially required eighteen months for implementation. Using the DataKitchen DataOps
Platform, which automates deployment, controls quality and supports Agile development
of analytics, the company was able to start delivering value in six weeks and completed the
migration in about one third the time.

THE CHALLENGE OF MARKETING AUTOMATION AT SCALE

The company faced numerous difficulties when implementing marketing automation for
multiple global business units:
• Distributed Data – The company’s marketing analytics and customer data were dis-
tributed in many specialized systems that do not easily talk to each other. This made
it challenging to link customer data from one system with another. Customers engage
through emails, partner websites, advertisements, campaigns and across product lines.
Each of these touchpoints produces a continuous stream of fine-grained opt-in/opt-out
requests which all must be consolidated and synchronized. Cross-referencing customer
data with third-party databases also provides valuable segmentation information.
• Supporting Agile – The company lacked the technical infrastructure to implement Agile
development of data flows and analytics. Using a slow and inflexible development
process prevented them from keeping up with the fast-paced requirements of the sales
and marketing teams.
• Iterating on Data Quality – Analysts had trouble specifying data quality rules until they
saw the data. In a non-agile environment, this caused requirements to keep changing,
causing delays.
• Continuous Change Requests – Once a system is operational, users are inspired to
request additional data sources, segmentations and other enhancements. With long
development cycles, it was difficult for the team to keep up with the users’ continuous
demands.

AUTOMATED EFFICIENCY AND QUALITY WITH DATAKITCHEN

The company utilized the DataKitchen Platform to oversee and monitor the end-to-end data
pipeline. With the DataKitchen solution, the company is now able to:
• Automate and monitor data pipelines – Data flows from sources to user analytics in a
continuous pipeline.
• Implement continuous deployment – New analytics are tested and deployed to users
with speed and confidence using automation.

The DataOps Cookbook • 123

 • Control quality – Data is continuously monitored for anomalies with alerts and dash-
boards that provide real-time information about data quality and operations.
• Manage sandboxes – Development environments are created as needed to prevent
enhancements from disrupting operations.

Figure 59: With DataKitchen, marketing automation data flows continuously from
numerous sources through the analytics pipeline with efficiency and quality.

BUSINESS IMPACT
With the DataKitchen Platform, the company was able to break the long 18-month project
into sprints and began to deliver value in six weeks. The agility of the DataKitchen DataOps
approach enabled the analytics team to rapidly respond to changing user requirements
with a continuous series of enhancements. Users no longer waited months to add new data
sources or make other changes. The team can now deploy new data sources, update sche-
mas and produce new analytics quickly and efficiently without fear of disrupting the existing
data pipelines.

DataKitchen’s lean manufacturing control helped the team be more proactive addressing
data quality issues. With monitoring and alerts, the team is now able to provide immediate
feedback to data suppliers about issues and can prevent bad data from reaching user analytics.

All this has led to improved insight into customers and markets and higher impact marketing
campaigns that drive revenue growth.

DataKitchen’s DataOps Platform helped this pharmaceutical company achieve its strategic
goals by improving analytics quality, responsiveness, and efficiency. DataKitchen software
provides support for improved processes, automation of tools, and agile development of new
analytics. With DataKitchen, the analytics team was able to deliver value to users in 1/10th

the time, accelerating and magnifying their impact on top line growth.

124 • The DataOps Cookbook

Meeting the Product Launch Challenge with DataOps
Celgene is a $12B biopharma-
ceutical company committed to “So much of what we do involves business
delivering innovative treat-
questions that are fire drills. Executives want
ments for patients worldwide.
answers as quickly as possible. The infrastruc-
Celgene relies upon DataKitch-
ture that we’ve set-up with DataKitchen allows
en to enable rapid-response,
us to mix and match data in new ways so that
high-quality data analytics that
help the company maximize we can quickly get the answer to a question.”
product lifetime revenue. —Manager, Data Analyst, Celgene
It costs between $2-3B to
bring a new pharmaceutical
to market. When a new drug is introduced, it is already halfway through its patent life. This
makes the first 6-12 months of a pharmaceutical launch critical to a product’s lifetime rev-
enue. The vendor needs up-to-date information to allocate samples, plan marketing events,
and monitor progress vs. goals. With so much at stake, pharmaceutical companies like Celgene
make strategic investments to maximize product adoption and adherence during the initial
phase of a drug product’s life cycle.

Figure 61: The first year is critical to a pharmaceutical

product’s lifetime revenue.

THE ROLE OF DATA IN LAUNCH SUCCESS

Celgene has found that using analytics to understand customers and markets can significant-
ly improve product launch success. The analytics produced range from weekly, standardized
reports to ad-hoc analyses. In the first months of a product’s lifecycle, the sales and market-
ing teams can’t wait weeks or months for new analytics. Every new data source, question,
and innovative idea demands an immediate and accurate response.

The DataOps Cookbook • 125

 THE OBSTACLES TO RESPONSIVE, HIGH-QUALITY ANALYTICS
The Celgene data engineering and analytics teams faced many obstacles that prevented
analytics responsiveness and quality. Data was organized in silos—using a variety of tech-
nologies and isolated platforms. Without the right processes and tools in place, the data
engineering and analytics teams can spend a majority of their time on data engineering and
pipeline maintenance. This distracts them from their main mission — producing analytic
insights that help the business attain its objectives.

THE DATAKITCHEN PLATFORM

Celgene chose DataKitchen to enable data engineering that
streamlines development and data operations processes,
helping the data analysts and scientists to remain focused on
creating value — at the speed of business. With the Dat-
aKitchen Platform, Celgene has been able to:
• Automate orchestration – DataKitchen automates the
deployment of new analytics and performs data pipeline
orchestration, freeing up the team from manual process-
es and enabling them to focus on extracting value from
data.
• Monitor data quality – Statistical process control and
dashboards help monitor and control the quality of the Table 7
end-to-end data pipeline, and real-time alerts provide
high-level visibility into incidents and provide critical
information.
• Automate deployment – Once new features pass all their tests, just a push of a button
is required to deploy into production, with confidence.

ANALYTICS AT SUPER SPEED

With help from DataKitchen, Celgene was able to improve the productivity of the data
engineering function by an order of magnitude. With automation of their data pipeline, they
were able to update 30X more visualizations per week. They were able to compress the
cycle time required to produce new
analytics from weeks (or months) to
one day. This enabled the data analyt-
ics team to successfully address the
volume of questions from sales and
marketing, helping them maximize
product adoption during the critical
first phase of their product launch.
Instead of just fighting fires, the
data team felt like they had acquired
Table 8 superpowers.

126 • The DataOps Cookbook

MOVING FORWARD WITH DATAKITCHEN
Use of DataKitchen fostered
a tight collaboration between “DataKitchen has enabled us to become nimble
data analytics and the business and agile when it comes to data. We are now
unit, unlocking creativity made
a self-service data organization — from the
possible by DataOps agility
marketing department to the sales reps.”
and quality. Impressed by the
performance of the data-ana- —Director, Market Insights, Celgene
lytics team using DataKitchen,
Celgene decided to expand
their use of DataKitchen throughout the company.

The DataOps Cookbook • 127

128 • The DataOps Cookbook
Additional
Recipes

The DataOps Cookbook • 129

 DataOps Vegan Corn Chowder
by Eran Strod

INGREDIENTS AND TOOLS

Cashew Cream
• 1 cup cashews soaked in water for at least 2 hours
• 2 cups veg stock
• 4 teaspoons cornstarch (can sub tapioca starch if desired)
• Drain the cashews. In a blender, combine all the ingredients and work for 2
to 5 minutes or until smooth, scraping down the sides with a rubber spatula
several times. Set aside.

Soup
• 1 Tablespoon olive oil
• 1 large onion coarsely chopped
• 2 celery ribs, chopped
• 3 cups veg broth
• 1 large carrot chopped
• 1 red pepper diced (could sub 1 bag Frozen mixed-vegetables, thawed in a
pinch)
• 1 potato, diced
• 3 ears of fresh corn (cut the kernels off and scrape the corn cobs for corn
milk to add to the soup)
• Can of corn

INSTRUCTIONS
1. Heat the oil in 4-quart pot
2. When hot add onion and celery with a pinch of salt, cook until start to soften.
3. Add carrots and potatoes
4. Add corn and red pepper and stir-fry for 10 minutes
5. Add 3 cups veg stock and the corn milk
6. Bring to a boil, lower the head and cover — simmer 10 min or until veg tender
but not overcooked.
7. Stir in Cashew Cream and stir gently for 7 minutes until nicely thickened.
8. Blend up to half the soup to make more liquid and add it back in
9. Add salt & pepper to taste, depending on the type of veg stock you used.

My own adaptation of a vegan New England Clam Chowder recipe from the Boston
Globe from Isa-does-it by Isa Chandra Moskowitz
130 • The DataOps Cookbook
DataOps Energy Bytes
by Eric Estabrooks

INGREDIENTS AND TOOLS

• 1 cup rolled oats
• ⅓ cup coconut flakes
• ½ cup peanut butter
• ½ cup ground flax seed
• ½ cup chocolate chips
• ⅓ cup raw honey
• 1 tsp vanilla

INSTRUCTIONS
1. Add rolled oats, coconut flakes, nut butter, flax seed, honey, and vanilla to a
mixing bowl and mix.
2. Mix well so that you can form the balls easily.
3. Add chocolate chips if using or other desired mix ins.
4. Chill the mixture in the fridge for an hour so that balls will bind together.
5. Roll the balls into about a 1-inch diameter.

The DataOps Cookbook • 131

 DataOps Classic Baked Macaroni and Cheese
by Joanne Ferrari

INGREDIENTS AND TOOLS

• 2 Cups Milk
• 2 Tablespoons Butter
• 2 Tablespoons All-Purpose Flour
• ½ Teaspoon Salt
• ¼ Teaspoon Freshly Ground Black Pepper
• 1 (10 oz.) Block Extra Sharp Cheddar Cheese, Shredded
• ¼ Teaspoon Ground Red Pepper (Optional)
• ½ (16 oz.) Package Elbow Macaroni, Cooked

INSTRUCTIONS
1. Preheat oven to 400◦. Microwave milk at HIGH for 1 ½ minutes. Melt butter
in a large skillet or Dutch oven over medium-low heat; whisk in flour until
smooth. Cook, whisking constantly for 1 minute.
2. Gradually whisk in warm milk and cook, whisking constantly 5 minutes or
until thickened.
3. Whisk in salt, black pepper, 1 cup shredded cheese, and if desired, red
pepper until smooth; stir in pasta. Spoon pasta mixture into a lightly greased
2-qt. baking dish; top with remaining cheese. Bake at 400◦ for 20 minutes or
until golden and bubbly.

NOTES
For this recipe, it is recommended that you grate the block(s) of cheese. I combine
Sharp Cheddar and Swiss cheeses — my favorite. Pre-shredded varieties won’t
give you the same sharp bite or melt into creamy goodness over your macaroni
as smoothly as block cheese that you grate yourself. You can go reduced-fat (but
then it’s even more important to prep your own). Grating won’t take long, and the
rest of this recipe is super simple. Use a pasta that has plenty of nooks to capture
the cheese—like elbows, shells, or cavatappi. Try it just once, and I guarantee that
Classic Baked Macaroni and Cheese will become your go-to comfort food.

132 • The DataOps Cookbook

 DataOps
Resources

The Agile Manifesto http://agilemanifesto.org/

DatOps Blog http://bit.ly/2Ef2Hto

The DataOps Manifesto http://dataopsmanifesto.org

DataOps News http://bit.ly/2ORDlUr

DataOps SlideShare http://bit.ly/2PygnSb

DataOps Videos http://bit.ly/2UFcKO8

Scrum Guides http://www.scrumguides.org

https://en.wikipedia.org/wiki/
Statistical Process Control
Statistical_process_control

https://en.wikipedia.org/wiki/
W. Edwards Demming
%20W._Edwards_Deming

Wikipedia DataOps http://bit.ly/2DnlqR1

Wikipedia DevOps https://en.wikipedia.org/wiki/DevOps

The DataOps Cookbook • 133

 About the
Authors

Christopher Bergh is a Founder and Head Chef at DataKitchen where, among other activi-
ties, he is leading DataKitchen’s DataOps initiative. Chris has more than 25 years of research,
engineering, analytics, and executive management experience.

Previously, Chris was Regional Vice President in the Revenue Management Intelligence
group in Model N. Before Model N, Chris was COO of LeapFrogRx, a descriptive and pre-
dictive analytics software and service provider. Chris led the acquisition of LeapFrogRx by
Model N in January 2012. Prior to LeapFrogRx Chris was CTO and VP of Product Manage-
ment of MarketSoft (now part of IBM) an innovative Enterprise Marketing Management
software. Prior to that, Chris developed Microsoft Passport, the predecessor to Windows
Live ID, a distributed authentication system used by 100s of Millions of users today. He
was awarded a US Patent for his work on that project. Before joining Microsoft, he led the
technical architecture and implementation of Firefly Passport, an early leader in Internet Per-
sonalization and Privacy. Microsoft subsequently acquired Firefly. Chris led the development
of the first travel-related e-commerce web site at NetMarket. Chris began his career at the
Massachusetts Institute of Technology’s (MIT) Lincoln Laboratory and NASA Ames Research
Center. There he created software and algorithms that provided aircraft arrival optimization
assistance to Air Traffic Controllers at several major airports in the United States.

Chris served as a Peace Corps Volunteer Math Teacher in Botswana, Africa. Chris has an M.S.
from Columbia University and a B.S. from the University of Wisconsin-Madison. He is an
avid cyclist, hiker, reader, and father of two college age children.

Gil Benghiat is a Founder and VP of Products at DataKitchen where he is focusing on Dat-

aKitchen users and the Agile data practices.

Gil has held various technical and leadership roles at Solid Oak Consulting, HealthEdge,
Phreesia, LeapFrogRx (purchased by Model N), Relicore (purchased by Symantec), Phase For-
ward (IPO and then purchased by Oracle), Netcentric, Sybase (purchased by SAP), and AT&T
Bell Laboratories (now Nokia Bell Labs).

134 • The DataOps Cookbook

Gil’s career has been data oriented starting with collecting and displaying network data
at AT&T Bell Labs, managing data at Sybase, collecting and cleaning clinical trial data at
PhaseForward, integrating pharmaceutical sales data at LeapFrogRx, protecting patient and
financial data at Phreesia, processing claims data at HealthEdge, and liberating data at Solid
Oak Consulting.

Gil holds an M.S. in Computer Science from Stanford University and a Sc.B. in Applied Math-
ematics/Biology from Brown University. He completed hiking all 48 of New Hampshire’s‚
4,000 peaks and is now working on the New England 67, and is the father of one high school
and two college age boys.

Eran Strod works in marketing at DataKitchen where he writes white papers, case studies
and the DataOps blog. Eran was previously Director of Marketing for Atrenne Integrated
Solutions (now Celestica) and has held product marketing and systems engineering roles at
Curtiss-Wright, Black Duck Software (now Synopsys), Mercury Systems, Motorola Computer
Group (now Artesyn), and Freescale Semiconductor (now NXP), where he was a contributing
author to the book “Network Processor Design, Issues and Practices.”

Eran began his career as a software developer at CSPi working in the field of embedded
computing.

Eran holds a B.A. in Computer Science and Psychology from the University of California at
Santa Cruz and an M.B.A. from Northeastern University. He is father to two children and
enjoys hiking, travel and watching the New England Patriots.

The DataOps Cookbook • 135

136 • The DataOps Cookbook
The DataOps Cookbook • 137
138 • The DataOps Cookbook