INDUSTRY DEVELOPMENTS AND MODELS

www.idc.com

Predictive Analytics and ROI:

Lessons from IDC's Financial Impact Study
F.508.935.4015

Henry D. Morris

IDC OPINION
A variety of technologies form the basis for business intelligence tools and analytic
P.508.872.8200

applications. Predictive analytics utilize mathematically oriented techniques (such as

neural networks, rule induction, and clustering) to discover relationships in data and
make predictions. The category of business intelligence tools includes predictive
analytics (in the form of data mining tools) as well as tools for query, reporting, and
multidimensional analysis. Analytic applications can incorporate either or both sets of
technologies — predictive and nonpredictive. In The Financial Impact of Business
Global Headquarters: 5 Speen Street Framingham, MA 01701 USA

Analytics: Key Findings (IDC #28689, January 2003), IDC examined analytics
projects at more than 40 sites across North America and Europe. There were
significant differences in the pattern of costs and benefits for projects that
incorporated predictive analytics versus those that did not. IDC's findings include:

! Both predictive and nonpredictive projects yielded high median ROI, 145% and
89%, respectively.

! The major benefits of business analytics projects that employed predictive

analytics centered on business process enhancement, especially improving the
quality of operational decisions.

! Predictive analytics projects required higher investment levels and yielded

significantly higher overall returns over five years, implying that these projects
tackled problems of greater scope and complexity.

Filing Information: September 2003, IDC #30080, Volume: 1, Tab: Others

Analytics and Data Warehousing Software: Industry Developments and Models
IN THIS STUDY
This study extends the analysis of IDC's major study The Financial Impact of
Business Analytics: Key Findings (IDC #28689, January 2003). It seeks to establish
whether there is a comparative business advantage to using predictive analytics tools
and technologies within a business analytics project.

M E T H O D O L O G Y

For the overall study The Financial Impact of Business Analytics: Key Findings (IDC
#28689, January 2003), 43 North American and European companies were chosen to
represent a fair selection of business analytics customer sites. Senior IDC consultants
conducted interviews with IS managers, business managers, department managers,
and system users. IDC made efforts to select companies based on a balanced
sample of geographic location, industry, and company size. IDC also made efforts to
represent various types of applications within the sample.

In addition, the companies surveyed were required to meet the following criteria:

! Be in production for at least six months with an analytic application that meets
the defining criteria

! Have pre- and postimplementation information to aid a cost/benefit analysis

! Be willing to participate in the study and reveal confidential cost and benefit
information to IDC

For this study, 40 projects from The Financial Impact of Business Analytics: Key
Findings (IDC #28689, January 2003) were classified as either predictive or
nonpredictive. Of this group, 15 projects were classified as predictive and 25 projects
were classified as nonpredictive. (The study included three other cases for which
there was insufficient information to classify the project or analyze it along the
dimensions examined in this study.) Predictive and nonpredictive business analytics
projects are defined as follows:

! Predictive business analytics projects. These projects utilized business

intelligence tools that IDC classifies as "data mining" (i.e., tools that incorporate
technologies that apply mathematically oriented techniques such as neural
networks, rule induction, and clustering to discover relationships in the data and
make predictions). Alternatively, predictive projects implemented packaged
analytic applications that incorporated predictive technologies, such as a fraud
detection application.

! Nonpredictive business analytics projects. These projects utilized business

intelligence tools that IDC classifies as "end-user query, reporting, and analysis."
Alternatively, nonpredictive projects implemented packaged data warehouses or
analytic applications that incorporated these technologies.

Projects that incorporated both predictive and nonpredictive technologies were

classified as "predictive." In fact, business intelligence tools for query, reporting, and
analysis were frequently used in conjunction with predictive analytics technologies.

©2003 IDC #30080 1

SITUATION OVERVIEW
Companies are scrutinizing IT investments with greater rigor, ensuring that the cost is
justified over competing alternatives for scarce funds. Analytics projects must be
justified by measuring the contribution they make to the business in the form of
reduced costs or added revenue generation.

The Financial Impact of Business Analytics: Key Findings (IDC #28689, January
2003) demonstrated that analytics projects can make such a difference. The median
ROI for all projects was 112% across organizations of a wide of range of sizes,
industries, geographies, and applications.

The choice of technology is only one element in ensuring a project's success. Best
practices were identified from an organizational perspective including senior
executive sponsorship, end users, and IT working together to define requirements
and a commitment to train knowledge workers on new skills needed in order to
change decision-making processes to make use of the information and analytical
models. Without these practices impacting people and processes, a project's chance
of success are slim. Assuming these practices are in place, this study explores the
differences in the nature of goals and achievements of projects that employed
predictive analytics versus those that did not.

D O E S P R E D I C T I V E A N A L Y T I C S M A K E A D I F F E R E N C E
O N A P R O J E C T ?

The median ROI for the projects that incorporated predictive technologies was 145%,
compared with a median ROI of 89% for those projects that did not. Both figures are
impressive, so this difference should not determine a decision on the choice of
technology.

What is more meaningful are the types of benefits that the projects yielded. Overall,
IDC measured three different types of benefits: technology, productivity, and business
process enhancement, which are defined as follows:

! Technology. The amount of money saved on technology or technology costs

avoided by introducing the analytic solution, such as diverting data processing to
a more cost-effective system

! Productivity. Efficiency savings due to the reduced amount of time and effort
required for particular tasks

! Business process enhancement. All identifiable annual savings that were

realized due to changes in business process supported by the analytic
application

Across all of the projects, the benefits attributable to technology accounted for only an
average of 4% of the total return. In other words, 96% of the benefits overall were in
the productivity and business process enhancement categories. Figure 1 shows the
comparative benefits for the predictive and nonpredictive cases according to these
categories.

2 #30080 ©2003 IDC

FIGURE 1

COMPARATIVE BENEFITS: PREDICTIVE PROJECTS VERSUS NONPREDICTIVE PROJECTS

80
70
60
50
(%)

40
30
20
10
0
Technology Productivity Business process
enhancement

Predictive
Nonpredictive

Source: IDC, 2003

The main benefits in predictive projects were attributable to business process

enhancement. For example, better analytical models for fraud detection resulted in
improved operational decisions that resulted in bottom-line savings. Or precise
segmentation by customer profitability led to sales strategies on pricing and product
mix.

On the other hand, the main benefits in the nonpredictive cases were due to
productivity improvements. For example, more organized means of gathering
financial data resulted in less time needed by financial professionals for producing
required reports for budgeting, revenue recognition, or financial consolidation.

Both types of benefits are important. Yet there is a limit to the efficiency gains due to
productivity in bringing information together. The data-gathering process can be
streamlined only so far before diminishing returns set in.

But this is not the case with business process enhancement. The virtuous cycle of
feedback and correction can be continually improved. Models can be made more
accurate in predicting the impact of policy changes. Knowledge workers can improve
their judgment as they learn to incorporate relevant feedback in making better
decisions.

The following characteristics were especially prominent in the group of cases in which
predictive analytics was employed:

! Repeatable, operational decisions. A key to business process enhancement is

to focus on a specific process in which the same type of decision is made over
and over again. Paying attention to improving these repeatable decisions pays
dividends. Examples from the study include:

# Deciding whether or not to extend credit to a customer

©2003 IDC #30080 3

# Finding leading indicators or predictors of fraud and taking action at several
banks in the study

# Identifying early warning of quality issues and correcting for them on an

ongoing basis in a semiconductor manufacturing process

! Precise customer segmentation. This was a factor in many of the cases,

including the fraud and credit cases noted. But it is also a key in shifts from mass
marketing of products to determining a product or group of products to
recommend to individual customers, based on an affinity model. Companies went
from a limited number of large-scale campaigns to more frequent campaigns at
particular customers or segments. This practice was specialized to companies
representing diverse vertical industries, from an online retailer that used the
information to make real-time purchase recommendations to a major
entertainment and gaming company that related specific customer and slot
machine usage to drive targeted activities.

! Responding to events. Several cases featured alerting or notification when

there were changes to key trends that were being monitored. This triggered
corrective actions. One of the best examples was a computer manufacturer
seeking advance warning of product quality and customer satisfaction issues.
The manufacturer was able to communicate with customers much more quickly
and take action to improve the production issues that were the basis for the
dissatisfaction.

The payback in these cases was due to a combination of the above factors. The
companies identified a type of repeatable decision that could be optimized, developed
very precise analysis and segmentation to predict likely outcomes, and took action on
an ongoing and event-based fashion. Adaptive organizations were able to continually
translate new learning into operational improvements.

P R E D I C T I V E A N A L Y T I C S A C R O S S T H E M A J O R T Y P E S
O F A N A L Y T I C A P P L I C A T I O N S

Each of the 40 projects involved the implementation of one or more analytic

applications. Some were built from the underlying business intelligence tools, while
some were bought (i.e., packaged analytic applications) and then customized to fit
specific needs. Whether built or bought, the applications were classified in one of the
following categories according to the business process supported:

! Financial/business performance management (BPM). These projects sought

to increase efficiency in financial processes, such as budgeting.

! Customer relationship management (CRM). These projects sought to enhance

customer support and marketing initiatives, decreasing costs and increasing
revenue.

! Operations/production. These projects sought to optimize the production and

delivery of a business' products and services.

Each project was classified in one of the three application groups. Figure 2 shows the
distribution of applications for the predictive and nonpredictive projects.

4 #30080 ©2003 IDC

FIGURE 2

ANALYTIC APPLICATIONS: PREDICTIVE PROJECTS VERSUS NONPREDICTIVE PROJECTS

100
90
80
70
60
(%)

50
40
30
20
10
0
Predictive Nonpredictive

Financial/BPM
CRM
Operations/production

Source: IDC, 2003

Following is an examination of the distribution of cases in greater detail:

! Operations/production. Both the predictive and nonpredictive projects had

significant representation for operations/production. The most important type of
project in this category that employed predictive analytics was fraud detection at
a bank or credit card company. For nonpredictive projects, examples featured
demand planning and manufacturing quality analysis.

! Customer relationship management. Most of the CRM projects in the study

featured predictive analytics. Examples included marketing analysis and real-
time recommendations in the Web-based or store-based retail environments.
Data mining is well suited to customer analytical applications. It's not just that
there are many customer records (whether point of sale for retail or call detail for
telco) — but the records are very "wide." In other words, there are many
attributes available on each customer, and mathematical techniques help to
identify which attributes are the best predictors of customer behavior.

! Financial/BPM. Most of the financial/BPM projects in the study emphasized

reporting and multidimensional analysis. This is traditional for financial planning
applications, a major category in this group. Other projects centered around
greater efficiency in producing and distributing reports. There is room for greater
deployment of predictive analytics in this category, especially for BPM. The
search for the leading indicators of financial performance could lead to an
intensive examination of large and diverse data sets. With the requirement for
early warning of material changes to a company's financial position (à la
Sarbanes-Oxley), the use of predictive analytics in this domain is likely to
increase.

©2003 IDC #30080 5

O V E R A L L I N V E S T M E N T A N D O V E R A L L R E T U R N / N E T
P R E S E N T V A L U E

The most striking difference between projects that did or did not employ predictive
analytics was the difference in scope. Figure 3 compares the two groups of projects
on the following three dimensions:

! Initial investment. The incremental costs to the organization incurred due to the
decision to implement a business analytics solution, up to the point of
implementation

! Total investment. The total five-year costs, including internal and external
services, software license, and maintenance

! Total return (net present value [NPV]). Similar to the income statement for the
project, calculated by subtracting the costs (including taxes and depreciation) from
the revenue to generate a net financial return or loss for the project (The calculation
assumes a 15% annual discount rate to account for the time value of money.)

Figure 3 shows the difference when the median results were compared for overall
investment and overall return between the predictive and nonpredictive projects.

There was a significantly higher cost to projects that employed predictive analytics
due to greater scope and complexity, but there was also a greater overall return. The
scope of major CRM projects is the most prominent example. From an IT perspective,
these projects require the integration of data from diverse customer-facing systems.
From a business perspective, these projects can require the linkage of processes
across business functions — a significant organizational challenge.

FIGURE 3

INVESTMENT AND OVERALL RETURN: PREDICTIVE PROJECTS VERSUS

NONPREDICTIVE PROJECTS

3,500
3,000
2,500
2,000
($000)

1,500
1,000
500
0
Initial investment Total investment NPV

Predictive
Nonpredictive

Source: IDC, 2003

6 #30080 ©2003 IDC

FUTURE OUTLOOK
Predictive analytics projects represented three-eighths of the cases in the study, while
projects that did not employ predictive technologies represented five-eighths of the
cases. But in the overall marketplace, the ratio is far more skewed in favor of query,
reporting, and analysis tools versus data mining tools. The share of data mining tools
within the BI market is increasing, but in 2002 this share stood at only 13.2%. IDC
forecasts the share of such predictive analytics tools in 2007 to grow to 15.5% of the
BI tools market. There are several reasons for this low penetration, including the
following:

! The use of predictive analytics in projects requires specialists who understand

both the specific mathematical techniques and the business problem to be able
to apply the appropriate technique. As packaged analytic applications incorporate
the experience of custom projects in specialized domains, this type of project will
become accessible to more organizations.

The number of users of predictive analytics tools within a project is small —

especially the number of users who build the analytical models. A greater
number of users apply the models and may use reports to direct attention to
particular results. But the results of the modeling can be incorporated as rules
within a transactional application. This is the way in which predictive analytics
can reach a large number of users. The call center operator or a telemarketer
would make a new offer, unaware that the recommendation came as a result of
the "embedded analytics." They would only observe that their system is
responding differently. Such applications ultimately would not even be sold as
"analytic applications" but as "enhanced operational systems with analytics
inside." The potential is to reach workers across all types of business operations
in all industries.

ESSENTIAL GUIDANCE
The results of this examination of business analytics projects suggest the following:

! The difference in median ROI across the two groups of cases should not imply
that predictive analytics is appropriate for all projects. The choice of technology
depends on the scope and complexity of the problem to be addressed and the
availability of skilled personnel to apply the technology. The scarcity of such
personnel implies an opportunity for packaging this expertise in analytic
applications.

! The benefits of predictive analytics projects centered on business process

enhancement, using information to drive more optimal decisions. It's best to
tackle processes where there are repeatable, operational decisions that over
time have a major impact on bottom-line results.

! Predictive analytics projects required higher investment levels but yielded

significantly higher five-year returns. In other words, organizations were tackling
problems of broader scope and complexity. Such projects are more challenging
but have been demonstrated to be highly rewarding.

©2003 IDC #30080 7

LE ARN MORE

R E L A T E D R E S E A R C H

! Worldwide Total Analytic Applications Software Competitive Analysis, 2003: 2002

Shares and Current Outlook (IDC #29886, August 2003)

! Worldwide Business Intelligence Forecast and Analysis, 2003–2007 (IDC

#29519, June 2003)

! The Financial Impact of Business Analytics: Key Findings (IDC #28689, January
2003)

C O P Y R I G H T N O T I C E

This IDC research document was published as part of an IDC continuous intelligence
service, providing written research, analyst interactions, telebriefings, and
conferences. Visit www.idc.com to learn more about IDC subscription and consulting
services. To view a list of IDC offices worldwide, visit www.idc.com/offices. Please
contact the IDC Hotline at 800.343.4952, ext. 7988 (or +1.508.988.7988) or
[email protected] for information on applying the price of this document toward the
purchase of an IDC service or for information on additional copies or Web rights.

Copyright 2003 IDC. Reproduction is forbidden unless authorized. All rights reserved.

Published Under Services: Analytics and Data Warehousing Software; Business

Intelligence and Data Warehousing Strategies

8 #30080 ©2003 IDC