Tool: A Step-By-Step Guide to Begin With Your

Analytics Initiative
Published 14 July 2020 - ID G00729782 - 13 min read
By Analyst(s): Kurt Schlegel, Melissa Davis
Initiatives: Executive Leadership: Data and Analytics; Analytics, BI and Data Science
Solutions

Executive leaders are looking for ways to get started with

analytics. This Tool provides them with a step-by-step guide to
help their organizations start and mature an analytics initiative.

Overview
Executive leaders responsible for starting an analytics initiative in their organizations
should use this Tool. It is based on four questions:

1. What is analytics?

2. Who will do the analytics work?

3. How will we deploy analytics?

4. Why will we deploy analytics?

This Tool answers these questions in a way broadly applicable to any vertical market or
horizontal domain.

Directions for Use

This research is adapted from “Toolkit: How to Get Started With Analytics,” which helps
data and analytics (D&A) leaders start and mature analytics initiatives.

When to Use
Executive leaders looking to start an analytics initiative in their organizations should use
this Tool. It is based on four key questions:

1. What Is Analytics?

Gartner, Inc. | G00729782 Page 1 of 10

This research note is restricted to the personal use of [email protected].

Gartner defines analytics as “the discipline that applies logic and mathematics to data to
provide insights for making better decisions.” We recommend describing analytics more
simply as your organization’s ability to measure, classify and decide (see Slide 1 in the
accompanying Microsoft PowerPoint attachment).

2. Who Will Do the Analytics Work?

Analytics requires a cross-functional team that blends domain expertise with data
engineering and data science skills (see Slide 2). It’s almost impossible to find a single
individual who has all three of these skills. Sometimes an individual can provide two of
the three, but usually, a team of three to five people needs to work together to provide
comprehensive skills. In practice, executive leaders should ensure their direct reports to
build a team by training existing staff, employing new workers, offering internships and
hiring consultants.

3. How Will We Deploy Analytics?

Analytics requires a prototype-first approach (see Slide 3).

Traditional analytics programs inherited the ERP “playbook,” which focused on a

production-first mentality. Nothing was delivered until it went through a rigorous
development and testing period and finally reached consensus. With decades of
hindsight, we recognize that the probabilistic nature of analytics requires a prototype-first
mindset.

We recommend starting with an analytics “sandbox” and a lot of freedom to blend

datasets and create analytic content such as dashboards and predictive models. Then
select the most useful content and promote it to a pilot or production tier by applying
more rigor to assess quality and consensus.

4. Why Will We Deploy Analytics?

Analytics initiatives can be successful in a myriad of ways. One of the most common
ways to show ROI is to automate the manual data preparation tasks that recur throughout
a large organization. When getting started with analytics, it’s important to clearly
communicate the business benefits that analytics provides. See Slide 4 for three common
ways in which analytics initiatives consistently deliver value — by allocating resources
more efficiently, optimizing conversions and detecting outliers.

Gartner, Inc. | G00729782 Page 2 of 10

This research note is restricted to the personal use of [email protected].

Obviously, there are more than just these three ways of delivering business value. For
example, another common benefit in some industries is route optimization. However,
given the aim of this Tool — to provide a simple plan for starting an analytics initiative —
we suggest starting with the three ways identified in Slide 4, given their broad applicability
to any vertical or horizontal domain.

Downloadable Attachments

Applied_Analytics_Guide_729782.pptx

This Microsoft PowerPoint file starts with a one-page vision that defines analytics in three
verbs: measure, classify and decide. The second slide defines the key roles needed to
conduct analytics initiatives. The third shows how analytics content should be created
quickly as prototypes and then promoted to a production tier. The fourth identifies three
key areas of business benefit that a successful analytics initiative typically provides. The
fifth provides a list of steps to follow. These can be applied to any aspect of your
company. Executive leaders can use this slide deck to communicate a simple plan for how
to get started with analytics, regardless of their organization’s current level of analytics
maturity.

How to Use
The slides provide answers to four key questions. Executive leaders can use these
answers as a generic framework that can be applied to virtually any analytics situation in
any vertical or horizontal domain.

Slide 1: What Is Analytics?

Answer: It’s your ability to measure, classify and decide.

“Measure, classify and decide” is not meant to replace “descriptive, diagnostic, predictive
and prescriptive.” These four styles are the foundation of the analytic continuum.
“Measure, classify and decide” is just a simpler way to communicate the essence of
analytics to an audience of beginners.

The process of building one’s organizational capabilities in terms of measuring,

classifying and deciding can also be viewed in terms of a “crawl, walk and run” model of
maturity. Most organizations’ analytics programs typically progress from “left to right.”
First, we measure. Second, we classify data to understand our business. Third, we apply
empirical systems to how we make decisions. The typical sequence in most organizations
is measure, classify, decide, but whatever the order, the ultimate purpose of analytics is to
bring logic, creativity and scrutiny to decision-making processes.

Gartner, Inc. | G00729782 Page 3 of 10

This research note is restricted to the personal use of [email protected].

Start on the upper left of the slide. Identify a measure that matters. Set a goal or expected
value. Then compare that goal or expected value to the actual value. Add more
sophistication with precise measurement and time series. Then gain more insight with
classification.

Measure

Analytics is fundamentally about measurement. This is the place to start any analytics
initiative. Too often, organizations start with the data. Executive leaders should work with
their peers and encourage their direct reports to participate in a project to create a data
dictionary. They should work with them to help create a performance metrics framework,
too. More often than not, they will be more enthusiastic about the latter project.

The ability to answer the following questions in the affirmative is fundamental to an

analytics program and incredibly useful to an organization. As our systems of
measurement mature:

■ Can we identify our key performance indicators (KPIs)?

■ In addition to KPIs, can we identify other measures that really matter to our
business?

■ Can we set goals and target values for what those measures ought to be?

■ Are we monitoring actual performance against the goal for each measure?

■ Can we discover leading indicators that influence our lagging indicators?

■ Can our organization model and estimate time series data?

■ Can we forecast what these measures are likely to be in the future?

Classify

Progressing to the middle column in Slide 1, we move from measurement to

classification. Using the verb “classify” in connection with analytics may seem less
natural than using “measure” and “decide,” but, in reality, most data analysis is a form of
classification. When we look at categorical data, we perform some type of classification
to organize categories with more granularity.

Classification can have a powerful impact.

Gartner, Inc. | G00729782 Page 4 of 10

This research note is restricted to the personal use of [email protected].

Answer the following questions to assess your organization’s ability to perform analytics’
classification:

■ How well can we dive into our data and identify the most important attributes?

■ Can we identify which attributes are most representative of customers who are
likely to churn? Which parts are likely to break? Which employees are likely to
succeed? What is the root cause?

■ Can we discover clusters (new categories) within our data?

■ In addition to finding homogeneous clusters, can we find useful associations within

our data — events that co-occur?

Performing this type of classification of data requires an ability to drill into the details,
blend disparate data sources and create custom groups. Eventually, this analysis can be
used to create taxonomies that describe customers, products and employees — all the
entities that matter to a business. This type of classification is often done with the help of
both visual data discovery tools (to identify categories visually) and machine learning (to
discover clusters, associations and perform classification algorithmically).

Decide

Progressing to the right side of Slide 1 confirms that the analytics program has matured
from measurement and classification to a more empirical and transparent system of
decision making. This is a more advanced analytics activity that few organizations have
yet achieved.

Only the most innovative organizations have created an analytics program focused on
evaluating how they make decisions empirically. These organizations put their decision-
making process under scrutiny, with an intent not to penalize but to learn from and avoid
mistakes. Organizations that have adopted this style of decision management typically
answer “yes” to the following questions:

■ Does your organization outline each decision with a creative list of available
alternatives?

■ Does your organization build transparency into each decision, making it clear who
makes each decision?

Gartner, Inc. | G00729782 Page 5 of 10

This research note is restricted to the personal use of [email protected].

■ When evaluating alternatives, does your organization use experimental design as a
means of testing the efficacy of a particular decision?

■ Are decision-making capabilities such as simulation, optimization and driver-based

planning made available to support the decision-making process?

■ Is there a step in the process for an extreme devil’s advocate to challenge decisions
in order to help avoid key biases (such as availability bias and confirmation bias)?

This may be a simple view of analytics, but improving how your organization measures,
classifies and decides empirically — with precision and granularity — will make a big
impact on your success. Now that we have defined and communicated what analytics is,
we need to define who will do the work.

Slide 2: Who Will Do the Analytics Work?

Answer: Cross-functional teams that include data engineers, data scientists and domain
specialists.

The old analytics “playbook” relies on teams of specialists to perform tasks and then
pass the associated projects on to another team of specialists, which does likewise, and
so on. This model is extremely ineffective when building analytics solutions, particularly
when you are not sure what you are aiming for.

Gartner recommends replacing or augmenting these teams of specialists with cross-

functional teams that blend all the skills of the specialists. The members of the new
teams should be jointly responsible for creating analytics content, such as dashboards
and predictive models. These cross-functional teams can have a variety of roles, but they
should consistently include roles that cover three basic areas: data integration (data
engineers), quantitative analysis (data scientists) and domain expertise. It’s almost
impossible to find an individual who has much knowledge of all three of these areas.
Assembling a cross-functional team is therefore the best way to ensure all three are
covered.

This requires a new organizational model in which the cross-functional teams operate at
both centralized and decentralized layers. This type of model has huge benefits for
knowledge sharing. It also gets around the virtual impossibility of training a data scientist
in a business domain in a short time. Moreover, the joint development fostered by cross-
functional teams leads to more rapid iterations of analytic content, which makes this one
of the best ways to reduce the risk of failing to meet requirements.

Gartner, Inc. | G00729782 Page 6 of 10

This research note is restricted to the personal use of [email protected].

Slide 3: How Will We Deploy Analytics?
Answer: By Taking a Prototype-First Approach

One problem with the old business intelligence (BI) playbook leaders is a lack of
resources. There are never enough people on the central team to empower decentralized,
domain-specific teams.

Domain-specific teams have to be created mostly from local resources. As with most
strategies and playbooks, successful outcomes depend on having the right personnel.
Executive leaders should:

■ Identify key individuals who are already performing data integration and data
science tasks locally.

■ Ensure training of these individuals to perform the semiformal roles of citizen data
integrator and citizen data scientist, and empower them with a “prototype first”
mindset.

■ Encourage these individuals to play a more collaborative role in the overall analytics
program.

These individuals may be trained by central resources, and their skills augmented by the
increasing array of self-service technologies that reduce the skill level required to use
analytics tools.

Encourage the creation of analytics content in a sandbox to foster a prototype-first

mindset among analytics leaders in individual domains. With a sandbox, there is more
freedom to integrate data, select a particular tool and perform analysis autonomously.
However, analytics content created in a sandbox needs to be treated as prototype content,
and must be vetted before being promoted to a production tier where it can be
disseminated to a wider audience. In fact, some organizations may need a three-tier
model that starts with the prototype, which may then be promoted to a limited pilot tier,
and then to a production tier. The vast majority of prototypes, however, will never progress
to the subsequent tiers, so this model alleviates the bottleneck problem that arises with a
production-first approach. Initial requests can be immediately satisfied with a prototype in
days or weeks, instead of waiting months for content built using the traditional
“production first” mentality.

Slide 4: Why Will We Deploy Analytics?

Gartner, Inc. | G00729782 Page 7 of 10

This research note is restricted to the personal use of [email protected].

Answer: Because successful analytics initiatives consistently deliver value in three
categories: resource allocation, conversion optimization and outlier detection.

Analytics ROI success stories can often be grouped into three categories. Obviously, there
are many other ways in which analytics can drive ROI (such as route optimization). But
three categories are vertically agnostic and as a result broadly applicable. It doesn’t matter
if your organization sells insurance, manufactures durable goods or provides healthcare,
oris in the private or public sector. Every organization has an opportunity to benefit from
the following three categories:

Resource Allocation

Every organization allocates resources such as head count, budget and inventory. Allocate
too little and you have a quality problem. Allocate too much and you have a cost problem.
A precise system of measurement can enable organizations to allocate just enough head
count, budget or inventory to balance the cost/quality trade-off.

Conversion Optimization

Every organization needs to segment its business, customers, employees and products.
The capability to classify with granularity can enable organizations to personalize their
business processes and, therefore, improve conversion ratios.

One can think of most business processes as trying to “convert” an undesirable situation
into a desirable one. We convert prospects into customers, opening trouble tickets into
closed trouble tickets and converting raw materials into finished goods. We convert new
recruits into fully trained employees. Our ability to collect disparate data sources and build
granular micro-segmentation models often using clustering can help us to identify
particular segments. We can personalize the business process and show ROI by
improving the conversion rate.

Outlier Detection

Gartner, Inc. | G00729782 Page 8 of 10

This research note is restricted to the personal use of [email protected].

Surprisingly, most dashboards don’t show a true comparative benchmark. Instead, they
show a goal or target value for what that metric ought to be and then compare the goal to
the actual. But they don’t show how individuals are performing compared to their peers.
Dashboards that provide true comparative benchmarks are captivating. Most viewers
can’t look away from a metric that compares their performance to their peers’
performance. True comparative benchmarking is difficult and requires a trusted data
aggregator to collect data at a low level of granularity and calculate metrics consistently
across all peers. It can also be dangerous as people don’t like to have a spotlight shone on
their failings. Executive leaders should therefore be politically savvy when rolling out
comparative benchmarks. For example, have the people who are to be measured by a
metric have a say in how the metrics are calculated. This will result in better and more
accepted metrics.

Where is the benefit in comparative benchmarking? The answer lies in outlier detection.
Through comparative benchmarking, organizations can identify the positive outliers, learn
their best practices and use them to remediate the negative outliers.

These three categories of benefit — resource allocation, conversion optimization and

outlier detection — are universal, regardless of the industry in which you compete. Focus
your analytics systems of measurement, classification and decision support on these
three categories of benefit and your organization will improve performance in every part of
the business.

Recommended by the Authors

“Practical Data and Analytics Strategy for Midsize Enterprises”

“How to Create Data and Analytics Everywhere for Everyone: Top Insights for Digital
Business”

“Build a Data-Driven Enterprise”

Disclaimer: Unless otherwise marked for external use, the items in this Gartner Tool are for
internal noncommercial use by the licensed Gartner client. The materials contained in this
Tool may not be repackaged or resold. Gartner makes no representations or warranties as
to the suitability of this Tool for any particular purpose, and disclaims all liabilities for any
damages, whether direct, consequential, incidental or special, arising out of the use of or
inability to use this material or the information provided herein.

Gartner, Inc. | G00729782 Page 9 of 10

This research note is restricted to the personal use of [email protected].

© 2021 Gartner, Inc. and/or its affiliates. All rights reserved. Gartner is a registered trademark of
Gartner, Inc. and its affiliates. This publication may not be reproduced or distributed in any form
without Gartner's prior written permission. It consists of the opinions of Gartner's research
organization, which should not be construed as statements of fact. While the information contained in
this publication has been obtained from sources believed to be reliable, Gartner disclaims all warranties
as to the accuracy, completeness or adequacy of such information. Although Gartner research may
address legal and financial issues, Gartner does not provide legal or investment advice and its research
should not be construed or used as such. Your access and use of this publication are governed by
Gartner’s Usage Policy. Gartner prides itself on its reputation for independence and objectivity. Its
research is produced independently by its research organization without input or influence from any
third party. For further information, see "Guiding Principles on Independence and Objectivity."

Gartner, Inc. | G00729782 Page 10 of 10

This research note is restricted to the personal use of [email protected].