WHITE PAPER

Demystifying data science

How organizations can benefit from artificial intelligence and
advanced analytics
 Contents
What is artificial intelligence and machine learning? 4

How can an organization derive business value from AI and analytics? 6

What are the requirements for adopting AI? 7

How can data science, artificial intelligence

and analytics transform business processes? 9

Common techniques and methodologies 10

Machine learning 10

Supervised learning 11

Unsupervised learning 12

Natural language processing 13

Key questions to ask and how to define high value use-cases 13

Resources 14

Demystifying data science 2/14

 Summary
By 2020, Forbes estimates that 85 percent of customer interactions will be
managed without a human.1 While many of us use AI technology, such as Alexa and
Siri, as part of our daily lives, we may not be aware of its greater uses. In fact, with
machine learning applied, AI can help teach computers, target ads and personalize
content for consumers to ensure better and more informed business decisions.

This paper will clarify some key definitions around artificial intelligence and
machine learning. It will also simplify some common techniques in machine learning,
such as supervised learning, natural language processing and classification, and
identify the types of business questions these techniques can answer.

While understanding a small number of customers may not pose a challenge,

keeping pace as organizations grow and expand their customer base can be
difficult. Data analytics can help reveal trends and metrics that would otherwise
be lost among the masses of information. Organizations are now starting to
leverage descriptive, diagnostic, predictive and prescriptive analytics to address
the growing needs and demands of their customer base.

The promise of artificial intelligence is exciting but before jumping in organizations

need the right data literacy, infrastructure and expertise. This paper will also cover
key competencies organizations need to get started with AI and how to progress
from data collection, exploration and analytics to artificial intelligence.

Finally, this paper will help define meaningful and high value use-cases with
a structured framework to gather and align business, technology and data
requirements for a successful artificial intelligence implementation.

Demystifying data science 3/14

 What is artificial intelligence and machine learning?
According to Gartner, artificial intelligence will generate $2.9 trillion USD in
Just as a human goes
business value and recover 6.2 billion hours of worker productivity by 2021. 2
through the process of To realize the high growth potential and costs savings from analytics and AI
technology, we must demystify some key artificial intelligence, machine learning
driver training to become
and analytics concepts.
proficient, a computer
Simply put, artificial intelligence systems automate and simplify tasks, such as
learns from experience or, recognizing objects, making sense of speech, etc. But, how does that lead to
learning how to drive a car? A key concept of AI technology is the difference
more specifically, data.
between learning and training. Just as a human goes through the process of
driver training to become proficient, a computer learns from experience or,
more specifically, data. Once the system has data on good driving practices and
the “rules of the road”, it becomes intelligent enough to make decisions in the
real world. While there are more complexities in the learning, management and
monitoring of such technology and solutions, this is the core of AI.

Deep learning

Popular and powerful set of machine learning techniques, which mimic the brain’s
neuron activities, called neural networks.

Machine learning
Deep learning Machine learning Artificial intelligence
Popular and powerful
AI
Field of set that learns from historicalField
of machine learning
dataof AI that learnsan
towards from
historical data towards an
Computing systems
end goal/outcome. For
capable of performing
example, which mimiclikely to default
the customers
techniques, on their home
end goal/outcome. For loan. tasks that humans are very
the brain’s neuron example, the customers good at, such as
activities, called likely to default on their recognizing objects,
Artificial intelligence
neural networks. home loan. recognizing and making
sense of speech,
Computing systems capable of performing tasks that humans very good
areself-driving cars. at,

such as recognizing objects, recognizing and making sense of speech, self-driving

cars.

Source: https://www.kdnuggets.com/2018/11/an-introduction-ai.html

Machine learning, a subset Machine learning, a subset of artificial intelligence, enables users to learn from
historical data to achieve a desired outcome. It powers targeted ads, personalized
of artificial intelligence,
content, song recommendations, predictive maintenance activities, virtual
enables users to learn assistants and more.

from historical data to Machine learning can be broken down into two key phases, learning and predicting.
In the learning phase, certain statistical techniques or algorithms are applied to
achieve a desired outcome.
historical data and/or previous business outcomes to generate a machine learning
It powers targeted ads, model. A model can be thought of as a set of rules or instructions, such as steps in
a recipe, that one must follow to make a business decision.
personalized content,
For example, in order to approve a loan application, a loan officer will consider
song recommendations,
income, age, net worth and many other factors before making a final decision.
predictive maintenance Each attribute of the application is a rule or factor that the officer must evaluate
to approve or reject the loan. Machine learning techniques follow a similar
activities, virtual assistants
methodology, comparing various attributes, historical decisions and the outcome
and more. of similar applicants to estimate the credit worthiness of the new applicant.

Demystifying data science 4/14

With the growth of data, In the predicting phase, patterns identified during the learning phase are applied
to new data or business processes to score or predict the likelihood of outcomes.
the invention of advanced
Scoring outcomes enables organizations to optimize resource allocation and
algorithms and cheaper decision-making activities, make more intelligent decisions and automate key
business processes at scale. Some key business questions that machine learning
commodity hardware to
techniques can help answer include:
process big data at scale,
1. Will my customer purchase product X?
deep learning, a powerful
2. Will my customer like a recommended song?
set of machine learning
3. Which of my customers are likely to switch to a competitor or cancel their contract?
techniques, has become
4.Of all recently submitted claims, which ones are likely to require an additional
prominent in the industry. fraud investigation unit review?

5. Is this applicant likely to default on their car loan in the future?

What do algorithms do?

Machine learning Recipe Analogy

An algorithm is a
Task Main objective
step-by-step instruction set Insight/ result
1. Take the chicken out
Task or formula for solving a 2. Salt and season
The algorithm
3. Bake it learns from
An algorithm is a step- Minimize
problem errors or
or completing a task
its mistakes/errors, finds
by-step instruction set some sort of “loss
the best approach and
or formula for solving a function” to attain the
generates insights and
problem or completing a best approach to solve
Main Minimize errors or some sort rules that canthebe
Minimize usedofto
number
task a “loss
of taskfunction” to attain the things/steps needed to take
objective best approach to solve a task
makeinpredictions
order to serve the dish
1. Take the chicken out Minimize the number Learn from your mistakes
of things/steps needed the next time you attempt
2. Salt and season
to take in order to the recipe
3. Bake it The algorithm
serve learns from its
the dish
Insight/ mistakes/errors, finds the
best approach and generates
Learn from your mistakes
the next time you attempt
result insights and rules that can be the recipe
used to make predictions

With the growth of data, the invention of advanced algorithms and cheaper
commodity hardware to process big data at scale, deep learning, a powerful set
of machine learning techniques, has become prominent in the industry. Deep
learning techniques mimic the brain’s neuron activities, which is why they are also
referred to as neural networks. Some common applications include natural language
processing, image recognition, realistic photo and video generation.

Demystifying data science 5/14

 How can an organization derive business value from AI
and analytics?
There are some common questions organizations consider when appealing
to their customer base: Who are the customers? What do they want? How can
the organization provide the best customer experience to gain a competitive
advantage? Data analytics help answer these business questions.

Data analytics is the science of analyzing raw data to draw conclusions from that
information. Data analytics techniques can reveal trends and metrics that would
otherwise be lost in a mass of information. This information can then be utilized
to optimize processes to increase the overall efficiency of a business or system.
Data analytics techniques can be broken down into four main types based on the
difficulty of analysis and business value.

a. D
 escriptive analytics parses raw historical data and draws conclusion that help
managers, investors and others determine why business changes occurred.

b. D
 iagnostic analytics provides an understanding of why events took place
by examining data. A type of advanced analytics, techniques include data
discovery and mining, correlation analysis and drill-down.

c. P
 redictive analytics uses statistics and modeling to predict future behavior. Using
data patterns, predictive analytics identifies when patterns are likely to reoccur
to identify and prevent potential risks, take advantage of future opportunities or
advantageously reallocate resources.

d. P
 rescriptive analytics uses machine learning to analyze raw data to help
organizations make better decision and take a proper course of action. Factoring
in possible scenarios, available resources, past performance and current
performance, prescriptive analytics help determine the best course of action in
a situation.

Data analytics techniques can be broken down

into four main types based onPredictive
the difficulty of
analytics
What will happen?
analysis and business value:

a. Descriptive analytics. What happened?

Prescriptive
b. D
 iagnostic analytics. Why did it happen? analytics
Business How can we
value
c. P
 redictive analytics. What willDiagnostic
happen?analytics make it happen?
Why did it happen?
d. P
 rescriptive analytics. How can we make it
happen?
Descriptive analytics
What happened?

Difficulty

Demystifying data science 6/14

 What are the requirements for adopting AI?
This hierarchical pyramid explains the competencies every organization requires
“Not going to the top is like
to ensure a successful AI implementation.
an insight engine working
at half capacity, not using
all its potential.” AI,
deep
learning
AI, d
 eep learning
Machine learning
Machine learning a
 nd benchmarking: A /B testing, experimentation
and benchmarking:
A/B testing,
BI or analytics: Metrics, segmentation, aggregation, data labelling
experimentation

Explore and transform: Data preparation, cleaning and exploratory data analysis
BI or analytics:
Data flow: Infrastructure, pipelines, ETL,segmentation,
Metrics, s
 tructured and unstructured data storage
aggregation, data labelling
Data collection: External data, logging, sensors, u
 ser generated content

Explore and transform:

Data preparation, cleaning
and exploratory data analysis

Data flow:
Infrastructure, pipelines, ETL,
structured and unstructured data storage

Data collection:
External data, logging, sensors,
user generated content

Data collection. At the bottom of the pyramid is data collection. At this stage, the
goal is to identify what data is needed and what is available. If it is a user-facing
product, are all relevant interactions logged? If it is a sensor, what data is coming
through and how? Without data, no machine learning or AI solution can learn or
predict outcomes.

Data flow. Identify how the data flows through the system. Is there a reliable
stream/ETL process established? Where is the data stored, and how easy is it to
access and analyze?

Explore and transform. Only when data is accessible can it be explored and
transformed for modelling. This stage is one of the most time-consuming and
underestimated of the data science project lifecycle. It is at this stage that teams
and organizations realize that they are missing data, their machine sensors are
unreliable, they are not tracking relevant information about customers and other
key issues. It forces them to return to data collection and ensure the foundation is
solid before moving forward.

Demystifying data science 7/14

 Business intelligence and analytics. When teams can reliably explore and clean
data, organizations can start building what is traditionally thought of as business
intelligence or analytics, which includes defining key metrics to track, identifying
how seasonality impacts product sales and operations, segmenting users based on
demographic factors, etc. However, as the goal is to build an artificial intelligence
solution, it is important to start thinking about the features or attributes to include
in machine learning models, what training data the machine will need to learn, what
to predict and automate and how to create the labels from which the machine will
learn. Label creation can be done automatically, such as when the machine breaks
down and it automatically registers an event in the back-end system. Or, it can
be done by introducing humans. For example, an engineer reports an issue when
a machine part seems to be faulty during a routine inspection and the result is
manually added to the data.

Machine learning and benchmarking. Although there is sample data that can be
used to make predictions, work is not complete. A/B testing or experimentation
framework needs to be in place to deploy models incrementally and avoid real
world disasters. Model validation and experimentation approaches provide a
rough estimate of the effects of changes before practical implementation. At
this stage, a very simple baseline or benchmark for performance tracking should
be established. An example fraud detection system includes monitoring high
risk credit card transactions that were proved to be fraudulent and comparing
them with the current operational performance of machine learning models to
accurately detect fraud.

Artificial intelligence. At this stage a team might be looking to make

improvements in production. This can be achieved by learning new methods
and techniques in machine learning and deep learning to improve processes,
predictions, outcomes and insights. By leveraging advanced and new techniques,
teams can gain an Information Advantage from massive amounts of data, explore
and model it faster and build solutions, such as voice assistants.

Demystifying data science 8/14

 How can data science, artificial intelligence and analytics
help transform business processes?
Data science is a multi-disciplinary field that uses scientific methods, processes,
algorithms and systems to extract knowledge and insights from structured and
unstructured data. Because there are a number of different techniques and
methodologies, it is often difficult to narrow down the scope of how data science
can impact business performance, operations, customer experience and costs. Here
are just a few ways data science can be leveraged.

Augment employee decisions with data-driven insights and intelligence

Utilize subject matter experts’ knowledge of how employees make business

decisions and transform the steps into data point, applying machine learning
techniques to identify the decision making pattern from this historical data to
predict future business outcomes. Organizations can design an intelligent system
that can handle complex requests or tasks, provide intelligent/best fit decisions
for individual scenarios and empower employees to make decisions quickly and
more effectively. Some example uses include credit risk scoring, automated
underwriting, wealth management fund assistants and customer service chatbots.

Automate and improve the efficiency of operations with intelligent,

data-driven decisions

Leverage AI and analytics techniques to drive operational efficiency. By utilizing

sensor information from machines, machine learning can help predict when a
specific machine is likely to require maintenance, allowing technicians to be
proactive rather than reactive in maintenance efforts. Some AI applications in this
context include predictive maintenance, recommender systems, robotic process
automation and airline scheduling.

Apply data driven insights to make timely and consequential tactical

and strategic decisions

Better inform management and strategic decisions by leveraging machine learning

and advanced analytics. These tend to be ad hoc projects or solutions, where
the goal is to apply statistical techniques to gain key insights around business
processes. For example, by measuring analytics related to cleanliness, customer
service, overall satisfaction, etc., an amusement park operations manager can
determine the likelihood of repeat customers, identify key gaps in operations and
better market the value of the park to the right demographics.

Personalize customer experiences

Identify and recommend personalized products at scale with recommender

systems. The likes of Google, Amazon, Facebook and Apple have made
personalization an expectation. Recommender systems are one of the more
popular examples of how machine learning can be leveraged to analyze data
across millions of users to accomplish this. By analyzing and tracking various
customer touchpoints, some retailers are now able to predict the likelihood
of users buying future products. It is important to note that machine learning
solutions need not be 100 percent accurate to realize business value and ROI.
The goal should be to conduct data-driven decision making at scale to reduce
operational costs and optimize resources and targeting efforts.

Demystifying data science 9/14

 Utilize data driven insights and intelligence to accelerate new product
development

Convert massive amounts of big data into meaningful and actionable insights. Voice
assistants, autopilot features and smart home devices have become part of day-
to-day life. This new class of AI-driven products are powered by machine learning
and advanced analytics techniques, allowing organizations and teams to better
understand consumer needs and wants, feature requests and usage patterns.

Common techniques and methodologies in machine learning

Machine learning takes what comes naturally to humans and applies it at scale.
For example, when machine learning reviews a loan application, it can review
5,000 credit transactions, three credit reports, 10 incidents, the five-year income
history of Joe Adams in seconds. This would not be possible by a domain expert.
They simply do not have the capacity to reviews with the speed of a machine and
provide a decision on his loan as soon as it is submitted. Even if the expert is highly
experienced and efficient, it takes considerable time to review application details
and there is still room for human error. Machine learning uses past experience and
trends in historical data related to customers in both good standing and those that
defaulted on loans to make a decision. With the combination of machine learning
and good quality data, organizations can quickly make unbiased, data-driven
decisions at scale in seconds.

Machine learning offers various approaches to solve business problems. The first
approach is based on whether there is data related to the outcome of a process.
Did the machine stop working? Did the customer leave? Did the employee quit?
It is important to understand and model how behavior and fluctuations in data
lead to a certain business outcome. This type of machine learning is known as
supervised learning.

Machine learning Unsupervised learning

Unsupervised learning Supervised learning
• Data does not contain a category/response to predict
• Data does not contain • We know the outcome/
a category/response to response
• The goal here is to learn about the underlying structure
predict or distribution of the data, and discover inherent
• Takes known
patterns set ofit input
from within in order to draw inferences
• The goal here is to learn data and responses to it, and 
about the underlying trains a model to generate
structure or distribution reasonable predictions for 
Machine
of the data, and discover the response tolearning
Supervised new data
learning
inherent patterns from
within it in order to draw
inferences • We know the outcome/response

• Takes known set of input data and responses to it, and

trains a model to generate reasonable predictions for
the response to new data

Demystifying data science 10/14

 If there is no response or category to predict, the goal is to learn the underlying
structure of the data and discover patterns to draw real world inferences. For
example, unsupervised learning approaches are commonly used to segment
customers based on demographic, behavior and past product purchase history.
This allows an organization to learn more about their customers, which products
are frequently bought together and how different groups prefer certain services
and products over others. It may not immediately understand that Emily from
Philadelphia falls within customer segment X, but the organization can learn how
many of its customers are similar to Emily based on behavior and consumption
characteristics. Are they active on mobile? Do they use social media? Do they
visit retail stores for purchases? Are they affluent? These insights can allow
organizations to make data-driven decisions for future marketing campaigns,
product development, etc.

Supervised learning

Supervised learning can be broken down into two categories based on what it is
trying to predict.

Use if data can be tagged,

Classification categorized or separated
Supervised learning (categories) into specific groups
or classes
Classification (categories) Regression(numbers)
Supervised
Use learning
if data can be tagged, Use if the response being
categorized or separated predicted is a real number
into specific groups or Regression Use if the response being
classes (numbers) predicted is a real number

Classification algorithms or approaches are used when asking questions regarding

categories. Examples include:

• Will this customer switch to another competitor in the next month?

• Will this customer default in the next month, six months or year?

• Is an email spam or genuine?

• Is this document for compliance, legal or customer support?

Regression algorithms or approaches are used when asking questions with

numerical outcomes:

• What will the temperature be at 6:00 pm today?

• In how many days will this machine stop working?

• What should be the price of a property based on size, number of rooms and location?

• How many orders am I likely to receive in the next three months for my product?

Demystifying data science 11/14

 Unsupervised learning

There are multiple unsupervised learning approaches and techniques that can
be utilized to gain meaningful insights. One of the more popular techniques is
clustering, which groups things that are similar or have features in common.
Organizations use clustering techniques to answer business questions, such as:

• How many distinct customer groups exist for my products? Who belongs to
which group?

• To which customer subgroups should I market my product and how should I target
them? What are the key characteristics of each group?

• How can I group my documents into distinct categories?

If a business is looking to answer questions around the identification of anomalies

or rare behavior and occurrences, anomaly detection techniques are utilized to
identify unusual patterns that do not conform to expected behavior, called outliers.
It has many applications in business, from intrusion detection, such as identifying
strange patterns in network traffic, to system health monitoring, including spotting
a malignant tumor in an MRI scan. Some additional questions that can be answered
using these techniques include:

• Given a massive database of financial data, which transactions are suspicious

and likely to be fraudulent?

• Given the huge number of container shipments arriving at a country’s ports every
day, which should be opened by customs to prevent smuggling, terrorism, etc.?

•G
 iven a log of all the traffic on a computer network, which sessions represent
attempted intrusions?

Association mining, another set of techniques, can help find correlations

between different products or factors in an organization’s data. For example, if a
customer purchases baby diapers, he or she has a 60 percent chance of purchasing
baby lotion within the next month. By identifying such insights using association
mining, retailers can predict the need for new products and target customers with
coupons and offers before the customer even realizes they are running out of baby
lotion. The most common application of association mining algorithms is in market
basket analysis.

Unsupervised learning
Partition the data set into X groups so that records
Clustering Anomaly
Clustering detection Association mining
in the same group are similar to each other, and
Unsupervised learning

records in different groups are dissimilar

Partition the data Identification Find frequent
set into X groups of rare items, patterns,
so that records in events or correlations
Identification ofand
rare items, events or observations
the same group Anomaly detection which
observations which raise suspicion by differing significantly from
associations
the majority of the data
are similar to each raise suspicion among a set of
other, and records by differing items in relational
in different groups significantly from and
Findtransaction
frequent patterns, correlations and
Association mining associations among a set of items in relational
are dissimilar the majority of the databases
and transaction databases
data

Demystifying data science 12/14

 Natural language processing (NLP)

Natural language processing is a set of systematic processes for intelligently

and efficiently analyzing, understanding and deriving information from text data.
It can organize massive amounts of text data and perform numerous automated
tasks, such as automatic summarization, machine translation, named entity
recognition, relationship extraction, sentiment analysis, speech recognition and
topic segmentation.

Key questions to ask and how to define high value use-cases

To identify meaningful and high value use-cases for teams and organizations, it is
important to gather relevant information and requirements on three key pillars.

1. Business knowledge
• What is the current business process? How are things done currently? Does
someone manually identify which products to recommend to each customer?
Does someone manually review each loan application for fraud or risk? Does an
engineer manually inspect all machinery each week for failure? Be as specific and
detailed as possible in defining the current process.

• Define measurable goals and objectives. Is it to replace or enhance an existing

process? Is it to increase revenue and conversions from product upsell and cross sell
opportunities or to increase software subscriptions by three percent this quarter?

• What areas can be improved? What is the business question to be answered

with analytics? Leverage subject matter experts to identify key pain points and
gaps in the current business process. Determine what part of the process can
be enhanced. Where can data-driven insights be used? Is the objective to speed
up loan application processing? Identify high risk transactions on credit cards?
Understand customers better? Specify key challenges and areas for improvement.

2. Solution vision
• Why is it important to solve the current business problem/use-case? Define
what success would look like. Specifically, in order to execute a successful
project, what are the minimum requirements and success criteria?

• Define what decision or business process will be affected by the analytical

solution. Who will be affected by this tool? Who are the users of this tool? Will
this impact the marketing department and analysts? Will it impact planning and
maintenance personnel? Will it impact the claims processing unit of an insurance
company that is responsible for mitigating fraud?

• How is the ROI of AI and analytics measured? Is there any current method to
track/benchmark the performance of current business processes and outcomes?

3. Data adequacy
• What data is available? Is it structured or unstructured? Is there data relevant to
answering the business problem? Example: Operational data is required to predict
when a machine will fail.

• How much data does the organization have? Where is the data stored?

• Is data readily accessible for analysis and modelling?

Demystifying data science 13/14

 Use-case evaluation worksheet
Learn more Section 1: Business knowledge

OpenText™ Magellan™

OpenText™ Magellan™
product overview

OpenText AI white paper

OpenText™ Magellan™ infographic
Section 2: Solution vision
Join the conversation

Keep up to date

Watch the videos
Section 3: Data adequacy

Tips and tricks

• Framing the right business question is key to success.

• Identify what success means and what the end solution will look like at the start.

• Remember that AI applications have a very different lifecycle—training, testing,

modelling, experimenting and creating.

• Start small, start early!

• Iterate, iterate, iterate!

About OpenText
OpenText, The Information Company, enables organizations to gain insight through
market leading information management solutions, on-premises or in the cloud. For
more information about OpenText (NASDAQ: OTEX, TSX: OTEX) visit: opentext.com.

Connect with us:

• OpenText CEO Mark Barrenechea’s blog
• Twitter | LinkedIn

1 Forbes, 5 Ways AI Is Transforming The Customer Experience, April 16, 2019.

2 Gartner, Gartner Says AI Augmentation Will Create $2.9 Trillion of Business Value in 2021, August 5, 2019.

opentext.com/contact 14/14

Copyright © 2021 Open Text. All Rights Reserved. Trademarks owned by Open Text.
For more information, visit: https://www.opentext.com/about/copyright-information • (23.04.21)17629.EN#