White Paper

Managing the Analytical Life Cycle

for Decisions at Scale:
How to Go From Data to Decisions as Quickly as Possible
Contents
Introduction.......................................................................... 1

Decisions at Scale............................................................... 2
What Are the Challenges?.......................................................... 2

The SAS® Analytical Life Cycle: Best Practices for

Improving Your Predictive Modeling Results................ 4
The Discovery Phase of the Analytical Life Cycle................... 4

The Deployment Phase of the Analytical Life Cycle.............. 5

From Concept to Action:

How to Create an Efficient Analytical Environment .... 6

How SAS® Can Help Across the Analytical

Life Cycle............................................................................... 7
Data Preparation and Exploration:
A Systematic Approach............................................................... 7

Model Development With an Innovative

Analytics Workbench................................................................... 8

Model Implementation All From One

Easy-to-Use Environment............................................................ 8

Action! Putting Your Modeling Results to Use

as Quickly as Possible................................................................... 9

Model Management and Evaluation:

An Ongoing Process.................................................................... 9

Case Studies.......................................................................10
UK Financial Institution:
Modernizing Its Analytical Life Cycle......................................10

Orlando Magic: The Magic Behind the Magic.....................10

For More Information....................................................... 10

 1

Introduction
Consider this scenario: For starters, you need powerful, easy-to-use software that can
help you wrangle your data into shape and quickly create many
An organization has hundreds of analytical models accurate predictive models. Then it takes powerful, integrated
embedded in production systems to support decision processes to manage your analytical models for optimal perfor-
making in marketing, pricing, credit risk, operational risk, mance throughout their entire life cycle. Both IT and analytics teams
fraud and finance functions. need efficient, repeatable processes and a reliable architecture
for managing data, communicating the rationale, and tracing
Analysts across different business units develop their the predictive analytics models through the deployment cycle.
models without any formalized or standardized processes
for storing, deploying and managing them. Some models And most importantly, the key to analytical success is quickly
don’t have documentation describing the model’s turning data into insights into actions, which means you must
owner, business purpose, usage guidelines or other efficiently integrate accurate predictive models into the produc-
information necessary for managing the model or tion systems that drive your automated decisions.
explaining it to regulators.

Model results are provided to decision makers with

limited controls and requirements. Because different
data sets and variables are used to create the models,
results are inconsistent. There is little validation or back-
Using Your Models for Automated
testing for accuracy. Decision Making
Operational decisions are highly repeatable
Decisions are made on the model results received – and
everyone hopes for the best. decisions, often made many times a day.
These decisions can be made by people
Such modeling confusion may look all too familiar to many (e.g., call center employees cross-sell or
organizations. In a diverse and loosely managed modeling envi- up-sell to customers on the phone), or they
ronment, it can be difficult to answer critical questions about the may be completely automated, mimicking
predictive analytical models that your organization relies on.
human decision logic (e.g., a credit card
Questions like:
purchase is declined). However, an opera-
• Who created the models and why?
tional decision doesn’t have to be about a
• What input variables are used to make predictions and,
single customer – it can be about which
ultimately, to make decisions?
group of customers gets a specific offer.
• How are the models used?
These types of decisions are pushed into
• How are models performing, and when were they last
updated? campaign management systems each week
• Where is the supporting documentation? or month. When predictive analytical models
• Why is it taking so long to put new or updated models into are embedded into production systems, and
production? a business process uses the results to
produce instantaneous answers, you have
The organization that can’t answer those questions with confidence
effectively deployed analytical models.
cannot be sure its analytical models are delivering real value.

Analytical models are at the heart of critical business decisions.

They can identify new opportunities, help you forge new or
better relationships with customers, and enable you to manage
uncertainty and risks. For these reasons and many more, they
should be created and treated as high-value organizational
assets. But how do you do that?
 2

The growing complexity and magnitude of managing poten- 4. A way to manage and monitor the analytical models to
tially hundreds or thousands of models in flux puts organiza- ensure they are performing well and continue to deliver the
tions at the cusp of an information revolution. The old and right answers.
inefficient handcrafted approach must evolve to a more effec- 5. An architecture and processes that can grow to address new
tive automated process. needs, like streaming data and building more detailed
predictive models faster than ever.

}} “Operational decisions are what make

your business strategy real and ensure
that your organization runs effectively.”
What Are the Challenges?
Unfortunately, though data abounds and everyone acknowl-
edges what is needed, the ability to turn voluminous, disjointed
data into useful insights that drive better automated decisions
isn’t easy. There are numerous challenges:

James Taylor and Neil Raden, • Delays. Due to processes that are often manual and ad hoc,
Smart (Enough) Systems it can take months to get a model implemented into produc-
tion systems. Because it takes so long to move models
through development and testing phases, they can be stale

Decisions at Scale by the time they reach production. Or they never get
deployed at all. Internal and external compliance issues can
How many operational decisions are made in your organization make the process even more challenging.
each day? Probably more than you can even imagine. For
• Difficulty defining triggers. The step of translating answers
example, take a financial institution. How many credit card trans-
from analytical models into business actions for operational
actions are processed each hour? (At Visa, the transaction rate
decisions requires clear, agreed-upon business rules. These
can reach 14,000 per second. See page 3.) Each one represents
business rules need to become part of the governed envi-
an operational decision – to allow or disallow the transaction to
ronment because they define how the results of the models
go through based on a calculated fraud risk score. And while
are used. For example, a fraud detection model might return
each operational decision or transaction may have a low indi-
a fraud risk score as a number between 100 and 1,000
vidual risk, the large number of these decisions that are made
(similar to a FICO credit score). It is up to the business to
hourly or daily greatly compounds the associated risk.
decide what level of risk requires action. If the trigger for a
fraud alert is set too high, fraud might go unnoticed. If the
That’s why the ability to produce good operational decisions
trigger for fraud is set too low, the alerts create too many
very quickly, while incorporating ever-increasing volumes of
false positives. Both outcomes will decrease the value these
data, can mean the difference between fraud and no fraud – in
models create and also reduce trust in the results.
business success or failure.
• Poor results. Too often poorly performing models remain in
production even though they are producing inaccurate
So what does it take to make a lot of good, fast operational deci-
results that lead to bad business decisions. Model results will
sions that consistently reflect overall organizational strategy and
change as the data changes to new conditions and behav-
at the same time keep your organization humming happily
iors. The main reasons for this situation are a lack of a central
along, faster and better than anyone else?
repository for models and no consistent metrics to deter-
1. Operational applications that use data to produce answers mine when a model needs to be refreshed or replaced.
for people (or systems) so that the right actions are taken.
2. Insightful and up-to-date analytical models that the business
can rely on for optimal decisions at the right time.
3. The integration of business rules and predictive analytics into
operational decision flows that provide the instructional
insight needed for vetted, trusted decisions.
 3

• Confusion. Organizations find themselves in reactive mode –

responding in a rush to deadlines from external agencies.
Each group has a different approach for handling and vali-
Visa: Billions of Operational Decisions a
dating a model. This results in unique reports with different
levels of detail for review or models that are inconsistently Year Enhance Customer Experience and
described, making IT translation difficult. No one is sure why Reduce Fraud
the champion model was selected, how a particular score A globally recognized name, Visa facilitates
was calculated or what governs the business rules that
electronic funds transfer through branded
trigger the model.
products that are issued by its thousands of
• Lack of transparency. There is little visibility into the stages of financial institution partners. The company
model development or knowledge of who touches the model
processed 64.9 billion transactions in 2014,
as it goes through its life cycle. Conflicting assumptions arise.
and $4.7 trillion in purchases were made with
Unbiased reviewers must be called in to validate models as
a Visa card in that same year.
they pass through each group – a big resource drain.
• Loss of important model knowledge. With inadequate docu- Visa has the computing capability to process
mentation for models, important intellectual property is in
56,000 transaction messages per second,
the mind of the model owner. When that person leaves, the
which is greater than four times the actual
knowledge is gone.
peak transaction rate to date. Visa doesn’t just
• Shortage of analytical skills. Even with increasing numbers
process and compute – it is continually using
of data scientists entering the marketplace, the shortage of
analytics to share strategic and operational
analytical skills needed for model creation and deployment
insights with its partner financial institutions,
is still a big challenge for many organizations.
and assist them in improving performance.

This business goal is supported by a robust

data management system. Visa also assists
its clients in improving performance and deliv-
ering deep analytical insight. “We understand
patterns of behavior by performing clustering
and segmentation at a granular level, and we
provide this insight to our financial institution
partners,” says Nathan Falkenborg, head of
Visa Performance for North Asia.

In a recent proof of concept, Visa used a high-

performance solution from SAS that relies on
in-memory computing to power statistical and
machine-learning algorithms and then present
the information visually. Falkenborg says, “The
challenge that we have, as with any company
managing and using massive data sets, is
how we use all necessary information to
solve a business challenge – whether it is
improving our fraud models, or assisting a
client to more effectively communicate with
its customers.”

Read more
 4

The SAS® Analytical Life Cycle: The Discovery Phase of the

Analytical Life Cycle
Best Practices for Improving
Your Predictive Modeling Results • Ask a question. The discovery process is driven by asking
business questions that produce innovations. So the first step
Successful organizations recognize that analytic models are is defining what the business needs to know. And then trans-
essential corporate assets that produce and deliver answers to lating the business question into a mathematical representa-
production systems for improved customer relationships, tion of the problem, which can be solved with predictive
improved operations, increased revenues and reduced risks. So analytics. Businesspeople also need to specify the need,
of course they seek to create the best models possible. scope, market conditions and goals related to a business
question they are trying to answer – which helps in the selec-
However, few fully manage all the complexities of the complete tion of the most appropriate modeling techniques.
analytical model life cycle. It’s such a multifaceted task. • Prepare the data. Technologies like Hadoop and faster,
cheaper computers have made it possible to store and use
At SAS, we’ve developed an iterative analytical life cycle to more data, and more types of data, than ever before. But this
guide you step-by-step through the process of going from data has only amplified the need to join data in different formats
to decision. We begin by realizing that there are two sides to the from different sources and transform raw data so that it can
analytical life cycle – discovery and deployment. Let’s take a look be used as input for predictive modeling. With new data
at the tasks for both sides and see how they interact to create an types from connected devices, such as machine sensor data
iterative process that you can use to produce repeatable, or web logs from online interactions, the data preparation
reliable predictive results. stage has become even more challenging. Many

Prepare Implement

Ask
lore

Discovery Deployment
Act
Exp

l Eva
M o de lu ate

Figure 1: The analytical life cycle from SAS.

 5

organizations still report that they spend an inordinate The Deployment Phase of the
amount of time, sometimes up to 80 percent, dealing with
Analytical Life Cycle
data preparation tasks. The majority of time in the discovery
phase should be spent on exploring data and creating good • Implement your models. This is where you take the insights
models instead of preparing data. learned from the discovery phase and put them into action
• Explore the data. Interactive, self-service visualization tools using repeatable, automated processes. In many organiza-
need to serve a wide range of users (from the business tions this is the point where the analytical modeling process
analyst with no statistical knowledge to the analytically savvy slows dramatically because there is no defined transition
data scientist) so they can easily search for relationships, between discovery and deployment, or collaboration
trends and patterns to gain deeper understanding of the between the model developers and IT deployment archi-
data. In this step, the question and the approach formed in tects, much less optimized automation. In most organizations
the initial “ask” phase of the project will be refined. Ideas on the deployment environment is very different from the
how to address the business problem from an analytical discovery environment, especially when the predictive
perspective are developed and tested. While examining the models are supporting operational decision making. Often,
data, you may find the need to add, delete or combine vari- IT has to apply rigorous governance policies to this environ-
ables to create more precisely focused models. Fast, interac- ment to ensure service-level agreements with the business.
tive tools help make this an iterative process, which is crucial By integrating the discovery and deployment phases, you
for identifying the best questions and answers. can create an automated, flexible and repeatable transition
• Model the data. In this stage, numerous analytical and that improves operational decisions. Additionally, a trans-
machine-learning modeling algorithms are applied to the parent, governed process is important for everyone – espe-
data to find the best representation of the relationships in the cially auditors. Once built, the model is registered, tested or
data that will help answer the business question. Analytical validated, approved and declared ready to be used with
tools search for a combination of data and modeling tech- production data (embedded into operational systems).
niques that reliably predict a desired outcome. There is no • Act on new information. There are two types of decisions
single algorithm that always performs best. The “best” algo- that can be made based on analytical results. Strategic deci-
rithm for solving the business problem depends on the data. sions are made by humans who examine results and take
Experimentation is key to finding the most reliable answer, action, usually looking to the future. Operational decisions
and automated model building can help minimize the time are automated – like credit scores or recommended best
to results and boost the productivity of analytical teams. offers. They don’t require human intervention because the
rules that humans would apply can be coded into the
In the past, data miners and data scientists were only able production systems. More and more organizations are
to create several models in a week or month using manual looking to automate operational decisions and provide real-
model-building tools. Improved software and faster time results to reduce decision latencies. Basing operational
computers have sped up the model-building process so decisions on answers from analytical models also makes
hundreds or even thousands of models can be created these decisions objective, consistent, repeatable and
today in the same time frame. But that brings to the surface measurable. Integration of models with enterprise decision
another issue – how to quickly and reliably identify the one management tools enables organizations to build compre-
model (out of many) that performs best? Model “tourna- hensive and complete operational decision flows. These
ments” provide a way to compare many different competing combine analytical models with business-rule-based triggers
algorithms with the opportunity to choose the one that to produce the best automated decisions. And because this
provides the best results for a specific set of data. With auto- is formally defined within the decision management tool,
mated tournaments of machine-learning algorithms and updates and refinements to changing conditions are easy
clearly defined metrics to identify the champion model, this – improving business agility and governance. Once
has become an easier process. Analysts and data scientists approved for production, the decision management tool
can spend their time focusing on more strategic questions applies the model to new operational data, generating the
and investigations. predictive insights necessary for better actions.
 6

• Evaluate your results. The next and perhaps most important development are usually the responsibility of business
step is to evaluate the outcome of the actions produced by analysts and data scientists. Deployment, especially when it
the analytical model. Did your models produce the correct includes integration into operational business processes, is
predictions? Were tangible results realized, such as increased managed by IT again (though this could be a different IT
revenue or decreased costs? With continuous monitoring group than the data management group providing data).
and measurement of the models’ performance based on
standardized metrics, you can evaluate the success of these The net effect is that the models, which are supposed to yield
assets for your business. That evaluation can then feed the solid business insights, lead instead to suboptimal decisions,
next iteration of the model, creating a continuous machine- missed opportunities and misguided actions. But it doesn’t have
learning loop. If you identify degradation of analytical to be that way!
models, you can define the optimal strategy to refresh them
so they continue to produce the desired results. With
increasing numbers of analytical models, automation is From Concept to Action:
necessary to quickly identify models that need the most
attention, and even deliver automated retraining.
How to Create an Efficient
• Ask again. Predictive models are not forever. The factors that Analytical Environment
drive the predictions in a model change over time, your In an effective analytical environment, data is rapidly created
customers change over time, competitors enter or leave the and accessed in the correct structure for exploration and model
market, and new data becomes available. You have to refresh development. Models are rapidly built and tested, and
even the most accurate models, and organizations will need deployed into a production environment with minimal delay.
to go through the discovery and deployment steps again. It’s Production models quickly generate trusted output. Model
a constant and evolving process. If a model degrades, it is performance is constantly monitored, and underperforming
recalibrated by changing the model coefficients or rebuilt models are quickly replaced by more up-to-date models.
with existing and new characteristics. When the model no
longer serves a business need, it is retired. In short, a successful analytics strategy means more than
creating a powerfully predictive model; it is about managing
It is easy to imagine the many ways this process can go wrong. each of these lifecycle stages holistically for a particular model
Organizations often take months, sometimes years, to move and across the entire portfolio of models. This is no easy feat.
through this end-to-end process. There are many common
complicating factors: Consider that analysts and data scientists don’t just develop one
• The needed data sources might be scattered across your model to solve a business problem. They develop a set of
organization. competing models and use different techniques to address
complex problems. They will have models at various stages of
• Data may need to be integrated and cleansed multiple times
development and models tailored for different product lines
to support a variety of analytical requirements.
and business problems. An organization can quickly find itself
• It can take a long time for models to be manually translated
managing thousands of models.
to different programming languages for integration with
critical operational systems, which can include both batch
Furthermore, the model environment is anything but static.
and real-time systems.
Models will be continually updated as they are tested and as
• Organizations might be slow to recognize when a model new results and data become available. The goal is to build the
needs to be changed, so they forge ahead with bad deci- best predictive models possible, using the best data available.
sions based on outdated model results.
• Many of the steps in the analytical life cycle are iterative in Predictive models are high-value organizational assets, and
nature and might require going back to a previous step in success requires more than relying solely on the technology
the cycle to add and/or refresh data. element. Organizations must also closely look at people and
• Different personas add complexity to the process, which processes. For example, it’s important to constantly upgrade
makes collaboration and documentation very important. In business and technical analytical skills so that the most impor-
many organizations, data preparation in the discovery phase tant business issues are identified and analytical insights can be
is handled by the IT unit, while data exploration and model applied to operational processes.
 7

The analytical life cycle is iterative and collaborative in nature. Data Preparation and Exploration:
Staff members with different backgrounds and skills are
A Systematic Approach
involved at various stages of the process. A business manager
has to clearly identify an issue or problem that requires • Data preparation. SAS® Data Management enables you to
analytics-driven insights, then make the appropriate business profile and cleanse data and create extract, load and trans-
decision and monitor the returns from the decision. A business form (ELT) routines that produce analytical data marts, using
analyst conducts data exploration and visualization and works to only the data that is required from the database. The data is
identify key variables influencing outcomes. The IT and data staged in the database for fast loading, transformed into a
management teams help facilitate data preparation and model structure fit for model building and summarized to create
deployment and monitoring. A data scientist or data miner derived fields. These processes can be automated and
performs more complex exploratory analysis, descriptive scheduled in batch or run ad hoc and in real time,
segmentation and predictive modeling. depending on the stage of the analytical life cycle. Self-
service data preparation and data wrangling tools like SAS
To get the best analytics results, organizations need to put Data Loader for Hadoop help business analysts and data
people with the right skills in place, and enable them to work scientists streamline access, blending and cleansing of data
collaboratively to perform their roles. without burdening IT. Event stream processing from SAS
leads to high-volume throughput of hundreds of millions of

How SAS® Can Help Across the events per second – with low-latency response times. It helps
you know what needs immediate attention, what can be
Analytical Life Cycle ignored and what should be stored. Finally, in-database
processing is used to reduce data movement and improve
SAS uses integrated components to reduce the time to value for
performance.
the modeling life cycle – eliminating redundant steps and
supporting cohesion across the information chain from data to • Data exploration. SAS Visual Analytics lets business analysts
decision management. Consistent processes and technologies easily discover important relationships in data and quickly
for model development and deployment reduce the risks zero in on areas of opportunity or concern, uncover unex-
involved in the modeling process while supporting collabora- pected patterns, examine data distributions, find the preva-
tion and governance among key business and IT stakeholders. lence of extreme values, and identify important variables to
incorporate in the model development process.

Figure 2: SAS Data Loader uses embedded instructions to

minimize training needs. This is an example of the “match-
merge data” directive, which lets you merge two tables into a
single table, and runs inside Hadoop for better performance.

Figure 3: Explore data from an interactive data visualization

environment.
 8

Model Development With an Innovative Model Implementation All From One

Analytics Workbench Easy-to-Use Environment
SAS Enterprise Decision Manager is key to improving the
Analysts can build predictive models using a variety of SAS tools
deployment side of the analytical life cycle. It provides a shared,
that include a rich set of algorithms to analyze structured and
web-based environment for managing the life cycle and gover-
unstructured data.
nance of your modeling assets, and is suitable for both business
• Automated machine-learning tournaments. With SAS users and IT. Analysts select data and models from a centralized
Factory Miner, you get an interactive predictive modeling repository, while being able to define business rules within
environment that makes it extremely easy to create, modify context of their models. This framework makes it easy to trace
and assess hundreds – or even thousands – of models very modeling activities, make modifications and test continually
quickly. With just a few clicks, you can access, modify and in a single environment.
transform your data, choose which machine-learning tech-
• Model registration. When model development is complete,
niques you want to apply, and run the models in an auto-
analysts register a model package that contains the model,
mated model tournament environment to quickly identify
including all of the data transformations, imputations, etc.,
the best performer. Templates using best-practice modeling
and all of its associated output and documentation. This
pipelines can be created and shared with other users of the
package ensures that the right steps have been taken, and
system without writing a single line of code.
that a suitable, powerful model is released into the produc-
• Predictive modeling and data mining. SAS® Enterprise MinerTM
tion environment. It also helps organizations standardize the
streamlines the data mining process so you can quickly
process of creating, managing, deploying and monitoring
create fine-tuned predictive and descriptive models based
analytical models.
on large volumes of data. An interactive, self-documenting
• Governance. Managing model risk includes accountability
process flow environment shortens the model development
metrics and version control status reports that track who
time for your data miners and statisticians. They can use
changes what, when control is passed from one area to
visual assessment and validation metrics and compare
another, and more. A centralized model repository, lifecycle
multiple models.
templates and version control provide visibility into analytical
• Text mining. SAS Text Analytics provides a rich suite of tools
processes and ensure that they can be audited to comply
for discovering and extracting concepts and knowledge
with internal governance and external regulations.
from text sources, including the web, call center notes,
• Model repository. A central, secure repository stores exten-
books, etc. You can add insights from text-based sources to
sive documentation about the model, its scoring code and
your analytical models for more predictive power.
associated metadata. Modelers can easily collaborate and
• In-memory analytics. SAS distributed, in-memory analytics
reuse model code, with their activities tracked via user/group
processing takes advantage of a highly scalable and reliable
authentication, version control and audit controls.
analytics infrastructure, including database appliances – like
• Scoring. Once a model has been reviewed, approved and
Pivotal Greenplum, Teradata, Oracle and SAP HANA – and
declared ready for production, it has attained champion
commodity hardware using Hadoop. Such fast processing
status. With the click of a button, the entire workflow of your
enables iterative machine-learning and statistical models to
champion model can be turned into scoring code that can
learn very quickly from extremely large data sources.
be deployed in SAS, third-party databases, Hadoop and web
services.

Figure 4: Customizable assessment techniques in SAS

Factory Miner enable you to generate champion models for
every segment in your data.
 9

• Faster and better governed scoring processes. Because the

model is scored directly in the database, the model execu-
tion job takes advantage of the scalability and processing
speed offered by the database. Jobs that used to take hours
and days can now be completed in minutes and seconds.
Data movement is minimized because the analytical scoring
can take place where the data resides. And since the data
does not have to leave the highly governed organizational
data environment, data governance is secured. Even in more
loosely managed environments like Hadoop, performance
benefits from eliminating data movement and duplication
can be significant.
Figure 5: SAS Decision Manager helps expedite the model
deployment process. It integrates model development
automation with SAS Factory Miner and accelerates common
Model Management and Evaluation:
manual tasks, like the definition of business rules and An Ongoing Process
automatic generation of vocabularies. Once a model is in a production environment and is being
executed to provide answers, the champion model is centrally
monitored through a variety of reports based on standardized
• Validation. Scoring logic is validated before models are put
key performance indicators. When the model performance
into production, using a systematic template and process to
starts to degrade below the acceptance level – based on the
record each test the scoring engine goes through, to ensure
centrally managed model training assets – the model can be
the logic embedded in the champion model is sound.
quickly recalibrated or replaced with a new model.

• Automated workflow. A web-based workflow console

Action! Putting Your Modeling Results to Use
enables the model management process to become more
as Quickly as Possible automated, repeatable, collaborative and governed. The
With SAS, you can choose from multiple deployment options to organization (and auditors) can track each step of a
get your champion models embedded into your production modeling project, from problem statement through devel-
systems, where they can produce analytic insights as quickly as opment, deployment and retirement.
possible. The integrated and automated deployment provided • Overall lifecycle management. All stages of a model’s life
by SAS Scoring Accelerators can boost your model deployment cycle are coordinated holistically, using prebuilt and
performance in several ways. customer-defined templates aligned with the organization’s
• Faster deployment. Deploy information and analytical business processes.
processes with minimum infrastructure and cost. The conver- • Automated model monitoring. Model performance jobs can
sion of complete analytical scoring code into lightweight be scheduled on a recurring basis – and the results moni-
web services or native languages for processing in data- tored on an ongoing basis – for one model or the entire
bases happens automatically. This eliminates the time- model inventory. These results can be posted on dashboards
consuming and error-prone manual process of translating and thresholds defined for performance. And if these thresh-
and validating the model score code for different olds are violated, analysts receive alerts about the models
environments. that need attention, saving time and money.
• Flexible deployment: One size does not fit all when referring
to the various operational systems. Without requiring With a formal model management framework, the best models
change, the same models can be deployed to both batch get into production faster to start serving the business sooner.
and real-time systems. This frees up analysts from the The organization can generate more and better models, with
drudgery of having to accommodate each environment with a large variety of analytics methods – with fewer resources.
different parameters, score code, and custom wrapper code. Analytical models are continually monitored and refined,
so they are up-to-date and accurate. The entire process of
discovery and deployment becomes more transparent and
better governed, so it is easy to explain analytics-based deci-
sions to regulators and business leaders.
 10

Case Studies The Magic don’t have a crystal ball, but they do have SAS Enterprise
Miner, which allowed them to better understand their data
With a predictive analytics lifecycle approach, the “after” scenario and develop analytic models that predict season ticket holder
looks quite different from the usual modus operandi – and renewals. The data mining tools allowed the team to accomplish
creates a serious competitive advantage. more accurate scoring that led to a difference – and market
improvement – in the way it approached customer retention
UK Financial Institution: and marketing.

Modernizing Its Analytical Life Cycle

Read more
A major financial institution in the UK determined that its cycle
time from model initiation to model deployment wouldn’t meet
21st-century expectations. The process was manual, error-prone For More Information
and resource-intensive – and had little or no monitoring to To learn more about the discovery phase of the analytical life
identify model degradation. cycle, read Data Mining From A to Z: How to Discover Insights
and Drive Better Opportunities.
Working with SAS and data warehousing vendor Teradata, the
organization built a flexible predictive analytics platform by inte- To learn more about the deployment phase, read From Data
grating data management, model development and model to Decision: How SAS Decision Manager Automates
deployment using in-database technology. The platform Operational Decisions.
harnesses the scalability of the Teradata environment for model
scoring and uses the power of SAS Analytics to build models.

With the new platform, more than 55 million records can be

scored within Teradata many times during the day – something
that could never have been achieved with the former process. The
time required to promote a model to a production environment
dropped from three months to days. Data preparation time was
trimmed by 40 percent. Analyst productivity jumped 50 percent.

Orlando Magic: The Magic Behind the Magic

Professional sports teams in smaller markets often struggle
to build a big enough revenue base to compete against their
larger market rivals. By using SAS Analytics and SAS Data
Management, the Orlando Magic are among the top revenue
earners in the NBA, despite being in the 20th-largest market.

The Magic accomplish this feat by studying the resale ticket

market to price tickets better, to predict season ticket holders
at risk of defection (and lure them back), and to analyze
concession and product merchandise sales to make sure the
organization has what the fans want every time they enter the
stadium. The club has even used SAS to help coaches put
together the best lineup.
 11

How SAS® Is Different

• Data preparation for analytics, interactive data explora- • A modeling lineage from data source to analytic result
tion and discovery, modern model development, model provides essential governance, critical for meeting regu-
deployment, monitoring and management – all in one lations or strict reporting requirements.
integrated platform. • Intuitive, graphical performance monitoring dashboards
• Integration of multiple analytical disciplines (such as help track model performance across multiple projects
data mining, machine learning, forecasting, text analytics so teams can focus on projects that need the most
and optimization) for greater model development agility immediate attention and avoid model decay.
and solving more complex business problems. • Interoperability with third-party modeling tools enables
• Smart automation and customizable predictive modeling organizations to import, manage and monitor modeling
and machine-learning templates with embedded best assets created by SAS and other tools (e.g., PMML
practices provide highly productive and collaborative models, R) all together in a central repository.
machine-learning tools to a diverse set of users. • In-database scoring functions can be achieved with
• SAS can manage large numbers of complex models with widely used databases such as Teradata, Aster Data,
virtually unlimited variables and extremely large data EMC Greenplum, IBM Netezza, IBM DB2, Oracle and
volumes. Teradata.
• A web-based, automated model-tournament environ- • SAS metadata spans data management and analytics so
ment boosts productivity by enabling modelers to that the data transformations used in the discovery
quickly and easily test many approaches simultaneously phase can be reused during deployment.
using machine learning and statistical algorithms. • SAS provides extensive technical and business expertise
• With web-based workflow capabilities, users can easily for pre- and post-sales support, helping you expedite
define custom processes, manage them through to time to value and improve return on investment.
completion, foster collaboration with notifications and
establish enterprise standards.
To contact your local SAS office, please visit: sas.com/offices

SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of
SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product
names are trademarks of their respective companies. Copyright © 2016, SAS Institute Inc. All rights reserved.
106179_S150641.0216