Transcript Operational Systems Vs. Analytical Systems

Operational Systems vs.
Analytical Systems
1. IT Today: The Missed Opportunity
2. Evolution of Computer Usage
3. Business Intelligence: Old Wine in a New Bottle
4. Hard Data vs Soft Data
5. The “Satisficing Concept”
6. Requirements of Actionable Information
7. Different Methodology for Analytical Systems
8. Analytical Tools: OLAP, Data Mining & Text Mining
9. The New Power of BI: Competing on Analytics
10. Capital One: The Quintessential Analytic Competitor
Dr. Lakshmi Mohan
1
An Opportunity To Be Seized . . . .
- Computers used in Business for Nearly 50 Years
- Dazzling Progress in Technology
- Significant Investments in IT Infrastructure
Hardware, Software and Peopleware
YET . .
Focus on OPERATIONAL SYSTEMS has blurred
the potential of using IT for
MANAGING the Business
Dr. Lakshmi Mohan
2
The Information Age Paradox
Lots of DATA
. . . but no INFORMATION
to manage the business
The 1980s launched the Information Age.
Still, most managers are less than satisfied
with their information systems.
Dr. Lakshmi Mohan
3
Computer Usage Evolution
ERA I:
Accounting (1950s to early 1960s)
e.g. Payroll, Invoicing
Benefit: “Manumation”
ERA II:
Operations (mid 1960s to early 1970s)
e.g. Airline Reservations, Inventory Control
Benefit: Improved Customer Service
Better Utilization of Capital
ERA III:
Information Support (early 1970s - )
e.g. Market Analysis, Succession Planning
Benefit: Better Decisions
Increased “People Productivity”
Dr. Lakshmi Mohan
4
Evolution of Analytical Systems
Decision Support Systems (DSS)
Support, not replace, mangers in making
decisions
Executive Information Systems (EIS)
Support top management
Group Decision Support systems (GDSS)
Support a group of managers
Expert Systems
To Manage by Wire
To extract news from the data
Dr. Lakshmi Mohan
5
One Idea, So Many Names!
 “Business Intelligence”: a term coined by Gartner in 1989
– Simply defined as using information effectively to make better
decisions
 Gartner’s Emphasis Today: Corporate Performance Management
– CPM means getting a better finger on the pulse of an organization
to make a better, more accurate, and more timely assessment of
how an organization is doing. Enterprises need to move away from
asking, “How did we do last month or last quarter” to “How are we
doing right now” as well as “How will we do next week”
 Meta Group: Business Performance Management
- Companies realize they have six different tools, but they do not
have a consistent approach to results reporting, management
reporting, planning and budgeting, etc.
Dr. Lakshmi Mohan
6
The Computer Made It
to the Executive Suite In Late 80s!
The computer has little to offer executives
since their work is unstructured.
- Fortune, Nov 1983
Executives are finally getting fast, clear
information about what's happening in the
bowels of their business. The new systems
can change the way top managers work.
- Fortune, Mar. 1989
Dr. Lakshmi Mohan
7
But Not Quite !
"Every business manager I know shares
one frustration:
the difficulty of obtaining fast, accurate
and comprehensive market information."
President, Frito-Lay
Wall St. Journal, June 11, 1990
Dr. Lakshmi Mohan
8
What Happened
in the Mid-1980s?
1. INCREASING GLOBAL COMPETITION
"In today's environment, a businessman without access
to good information is playing with one hand behind his
back“.
2. TIMELY "INFORMATION " BECAME CRITICAL
"We need enough advance warning to steer around the
iceberg. What we have had so far is the world's best
damage report“.
3. CORPORATE DOWNSIZING
"Fewer staff analysts available to sift through the
mountain of ‘data’ and cull out relevant ‘information’ ”.
4. TECHNOLOGICAL ADVANCES
Made Executive Information Systems systems a reality.
Dr. Lakshmi Mohan
9
Major Growth Drivers for BI in 2005
1. Need for organizations to make a sense of the “data
tsunami” that is hitting them from their enterprise
applications
2. Focus on performance management and the need to
develop and measure the associated key performance
indicators
3. Ensuring accuracy, timeliness and consistency of data
for regulatory reporting
Market for BI tools is estimated to grow from $ 3.7B in 2002 to
$4.5B in 2007, according to industry analyst, IDC.
Source: DM Review, April 2003
Source: Gartner Management Update, Nov. 2004
Dr. Lakshmi Mohan
10
Case Example: 7-Eleven Japan
- BI for Implementing a Customer-Centric Strategy
 1974: Ito-Yokado acquired franchise rights to 7-Eleven in Japan
-
By 1998, expanded to over 5,000 stores
Company’s profitability reached 40% of sales
In contrast, 7-Eleven USA filed for bankruptcy
Sold 70% of its stores to Ito-Yokado
 IT systems captured information about customers and their needs
- Every clerk recorded customer features (gender, approx age, etc) at the
time of purchase
- Also, products requested that were not available in the store’s inventory
 Inventory management systems based on customer information
- Which products to stock in each store?
- How much shelf space for each product?
- Which are the most sellable items at different hours of the day and hence
should be displayed?
Dr. Lakshmi Mohan
11
The Urgency for BI
Gartner’s March 2005 BI Summit in Chicago and London
drew over 750 attendees at each event!
 Mountains of data; growing at 30% to 50% a year
 Pressure of regulations such as the US SarbanesOxley Act and in financial services, the Basel II capital
adequacy rules forcing companies to “fix-up” their
practices
 Competition and customer demands requiring timely –
and often real-time – information, and in plenty of detail
- to “drill down” from high-level summarized reports,
with the intervals between updates coming down,
especially in financial services (“from 4 hours to 15-25
minutes”)
Dr. Lakshmi Mohan
12
One Version of the Truth?
 Data resides in disparate systems cobbled together over
the years
 Duplication of data in different systems, frequently
conflicting – Which data is accurate?
MIS reports causing confusion –
Which report has the correct data?
 Spreadsheet Hell!
- Multiple spreadsheet databases created by users
- Slow process; Prone to Error
- Multiple Versions of the Truth
The goal of BI systems is to pull data from all internal systems
AND external sources to present a SINGLE version of the Truth.
Dr. Lakshmi Mohan
13
BI for the Masses
- Not just for Statisticians and Corporate Analysts
 Instead of a small number of analysts spending 100% of their time
analyzing data, all managers and professionals should spend 10% of their
time using BI software
 Smart companies are democratizing data access with dashboards and
other BI tools to empower everyone in the organization, at all levels, with
analytics, alerts and feedback mechanisms
- Transforms every employee into an “organization of one” who can make
the right decisions at the right time in step with company objectives.
- Everyone can work smarter!
 Smarter Companies will ensure the payoff of investments in BI systems by
making the masses accountable for data-driven action and results.
- Accountability could be in the form of rewards, penalties
Or simply, a mandated workflow
Dr. Lakshmi Mohan
14
GE’s Concept of “Span”
- Measures the operational reliability for meeting a customer request
… the time window around the Customer Requested Delivery Date
in which the delivery will happen
- High Span  Poor capability to meet customer need
Objective  Zero span
- Squeeze the two sides of the delivery span - days early and days
late - ever closer to the center - the exact day the customer desired
RESULTS :
Plastics
Aircraft Engines
Mortgage Insurance
Dr. Lakshmi Mohan
: 50 days span to 5
: 80 days span to 5
: 54 days span to 1
15
The GE Process
- In the CEO’s Annual Letter (Feb 2001)
When the order is taken, that date becomes known to everyone,
from the first person in the process receiving the castings, circuit
boards or any other components from the supplier, all the way
through to the service reps who stand next to the customer as the
process is started up for the first time.
Every single delivery to every single customer is measured and in
the line of sight of everyone; and, everyone in the process knows he
or she is affecting the business-wide measurement of span with
every action taken.
WHAT GETS MEASURED
AND REWARDED
GETS DONE !
Dr. Lakshmi Mohan
16
A Manufacturing Analogy
Raw Data = Raw Material
Conversion Process
–DSS/EIS/BI Systems
–Expert Systems
–Data Mining, etc.
Actionable Information = Finished Product
Quality of the Conversion Process
– As Important as the Quality of the Data
Dr. Lakshmi Mohan
17
Mere Access to Quality Data
- Not Enough !
. . . Will create a data overload that can affect
managerial productivity.
Investments on market research,
telecommunications, etc. to deliver better quality
data should be complemented by investments in
systems to convert the data into useful information.
Quality of the data conversion process is
equally important.
Dr. Lakshmi Mohan
18
Managers Ask for Analysis,
NOT just Retrieval of Data
Sometimes retrieval questions come up of course, but most
often the answers to important questions require non-trivial
manipulation of stored data. Knowing this tells us much
about the kind of software required. For example, a
database management system is not enough.
- John Little (1979)
• “Data” has to be converted into “Information” that
triggers managerial action.
• The conversion process is critical to get value from the
data warehouse.
Dr. Lakshmi Mohan
19
Managerial Data Encompasses
Hard and Soft Data
 Subjective judgments are an important source for
quantities that are difficult to measure, or which
cannot be measured in the time available before a
decision is made
 Soft data is more essential for certain functions...
... Marketing: Competitive Intelligence
... Human Resources: Succession Planning
... Corporate Planning: Forecasts
And, for upper levels of management
... External data about the environment
Dr. Lakshmi Mohan
20
HARD vs. SOFT DATA
• HARD DATA ... Fairly Accurate, Easy To Get
– Historical Data
e.g., Revenue, Direct Costs
– Measured Data
e.g., Bill of Materials for a Product
• SOFT DATA ... Fairly Inaccurate, Difficult to Get
– Future Data
e.g., Sales Forecasts
– Judgemental Data
e.g., Allocation of Overhead Costs
– Qualitative Data
e.g., High Potential of an Employee
Dr. Lakshmi Mohan
21
The Data Isn’t Where We Need It !
Senior Managers
-Strategic Planning
Middle Managers
- Management Control
Internal, Hard Data
External, Soft Data
Front-Lines
- Operational Control
Corporate Data Warehouse
The greatest challenge of the computer industry is to learn how to
build information bases, not databases. The really important
information cannot be easily quantified and exists outside the
organization.
- Peter Drucker (1993)
Dr. Lakshmi Mohan
22
The Problem : How to Get Soft Data
 Develop Standard Format for Data Collection
– Minimize Text Data because:
• time-consuming
• inconsistent
• not easy to analyze
– Use Categories, Statements, Rating Scales
 Explicate Mental Model of Data Provider
– Decompose the entity being estimated in one
judgemental swoop into a set of smaller elements that
are less difficult to estimate.
Dr. Lakshmi Mohan
23
Success Dimensions for
High-Potential Evaluation
1. Adaptability – maintaining effectiveness in varying
environments and with varying tasks, responsibilities, or people.
2. Decision Making – utilizing appropriate problem solving skills to
develop alternative courses of action and to subsequently direct
implementation of the most advantageous method of resolution.
3. Judgement – developing alternative courses of action and
making decisions which are based on logical assumptions and
which reflect factual information.
4. Planning – establishing a course of action for self and/or others
to accomplish a specific goal; planning proper assignments of
personnel and appropriate allocation of resources.
Dr. Lakshmi Mohan
24
Success Dimensions for
High-Potential Evaluation
5. Persuasiveness – utilizing appropriate interpersonal styles
and methods of communication to gain agreement or
acceptance of an idea, plan or activity.
6. Communications – combining necessary elements of
listening, oral and written communication skills resulting in
effective understanding and expression of information.
7. Initiative – active attempts to influence events to achieve
goals; self-starting rather than passively accepting. Taking
action to achieve goals beyond what is necessarily called
for; originating action.
Dr. Lakshmi Mohan
25
Success Dimensions for
High-Potential Evaluation
8. Leadership – utilization of appropriate interpersonal styles
and methods in guiding individuals (subordinates, peers,
superiors) or groups toward task accomplishments.
9. Problem Solving – ability to gather relevant data, recognize
and assess potential areas of concern, evaluate alternative
courses of action, anticipate problem situations and develop
contingent plans to resolve situations.
10. Teamwork – skill in coordinating activities of own
personnel with those of others to achieve complex,
interrelated goals.
Dr. Lakshmi Mohan
26
Succession Planning System:
All the Required Data is Soft !
• How to identify “high potential” employees and
match them to positions
• Measure employees among
10 “success dimensions”
relative to everyone else
• Use a 4-point rating scale
• Measure positions across
the same dimensions
• Match employees
to positions
Dr. Lakshmi Mohan
Communication
Analytics
Judgement
Teamwork
Jose
4
4
2
2
Irma Luis
3
2
2
3
3
4
3
4
Gen. Sales Senior
Manager
Accountant
Communication
4
2
Analytics
2
4
Judgement
3
3
Teamwork
4
1
27
Information Must be Tailored to
Management Level
Management Level
Lower
1. Function
2. Scope of
Responsibility
Operational
Control
Narrow
Middle
Management
Control
Top
Strategic Planning
Wide
Narrow
Wide
4. Sources of
Data
Internal
External
5. Time Horizon
Historical
6. Level of Detail
Micro
3. Scope of
Information
Dr. Lakshmi Mohan
Future
Macro ?
28
Type of Data in the BI System
• Not just Hard, Internal Data
• Not limited to Financial Data
• Must include Soft, External Data
• Key Areas to be Considered:
– Measurement of Customer Service
– Market Information on Customers & Competition
– High-Potential Evaluation, Succession Planning &
Career Development of Employees
Dr. Lakshmi Mohan
29
A Different Perspective on Data Quality
... Depending on Use
Operational Systems
(e.g., Invoicing, Airline Reservations, Electronic Commerce, etc.)
• Emphasis on complete, accurate and timely data
• But limited to internal, hard data
• Cost of data quality justifiable because
systems will be used
Analytical Systems
(e.g., Performance Evaluation, Market Analysis, etc.)
• Scope of Data is Wider - External and Soft data
• But ... Is “Better” Data Worthwhile?
• Value is zero if system is not used
Dr. Lakshmi Mohan
30
COST versus VALUE OF DATA
- “Satisficing” Concept
Better data
Higher cost
Value
Impact on the decision
Aim: Get a Satisficing Solution for Decision-Making
- Select a satisfactory decision with limited information in a
limited time instead of searching for the best solution
entailing more time and information
"We are subjecting every activity, every function to the most rigorous
review, distinguishing between those things which we absolutely need
to do and know versus those which would be merely nice to do and
know."
GE CEO
Dr. Lakshmi Mohan
31
Actionable Information
… Information that becomes the basis for action
•
Must be Timely
•
“Satisficing” Accuracy is Enough
•
Must Help in ...
Problem-Finding and Problem-Solving
Dr. Lakshmi Mohan
32
Attributes of
“Actionable” Information
• Timeliness
– If it is late, managers will make decisions without it
• Complete and Accurate? How much?
– Just good enough for decision-making
What is absolutely needed in relation to
What is at stake
– Reason: $$$$$$
100% Complete and Accurate takes time and is
expensive
– The key concept in information accuracy and
completeness is “Satisficing.”
Dr. Lakshmi Mohan
33
Timeliness vs. Accuracy Problem
• Precise Financial Data Has a Price: Time
– Accruals, adjustment entries and allocations lengthen
Monthly Closing Cycles
– Is the Precision worth the Time Lag in the Data?
• “Just-In-Time” Monthly Closing
– Timely Data with Satisficing Accuracy
– Frees up time of financial staff for value-added analysis
Dr. Lakshmi Mohan
34
The DURACELL EIS:
How It Provides “Information”
 The CEO, Robert Kidder, manipulated a mouse attached
to his workstation.
 To compare the performance of work forces in the U.S.
and overseas.
 Computer displayed a crisp table in colors showing
higher sales per employee in the U.S.
He asked the computer to drill down for more data to
explain the difference.
 At the end of the data-browsing session, the real
problem was found:
.... TOO MANY SALESPEOPLE IN GERMANY WERE WASTING
TIME CALLING ON SMALL CUSTOMERS.
Dr. Lakshmi Mohan
35
Micro-Level Data in the Duracell BI
- To Trace Problems to Root Causes
• Customer Sales Data
– Able to Segment Customers by Size
… Small, Medium, Large
• Salespeople Data
– Which Salesperson Calls on Which Customer
• Most Important: Time Spent With Each Customer
– From the Salesperson’s Call Reports
– BIG Problem: Fear of “Policing”
… Is the Time Data Usable?
– Data Feeders Must Benefit from Data!
Dr. Lakshmi Mohan
36
Why Analytical Systems are a Different Breed
 Operational Systems will be used because they run
the “bread-and-butter” business processes of the
organization - they are mission-critical
 Analytical Systems depend on managers’ desire and
ability to use them in their decision-making
processes to manage the business
Prerequisite:
The management process must be driven by the
information provided by the system. Only then will
the Analytical system be used.
Dr. Lakshmi Mohan
37
Payoff from the Analytical System
Depends on the Management Process
If a magic fairy instantly gave you all the information...the company
would ever need, do you think people would instantly know what to
do with it and use it well.
Peter Keen (1998)
• Easier to upgrade quality of the data than the
management process for utilizing the high
quality information.
• Improving the quality of data will be all costs
and no benefits if the data is not used.
• Need to upgrade the management process to
effectively use better quality data.
Dr. Lakshmi Mohan
38
Efficiency vs Effectiveness
There’s nothing so useless as doing efficiently that
which should not be done at all. Companies wrench
their guts to downsize a business they shouldn’t be in
at all.
…… Peter Drucker says it well
Effectiveness:
Efficiency:
Doing the Right Thing
Doing IT Right
Which is more important in an Analytical system ?
Dr. Lakshmi Mohan
39
A Home Truth:
A System That Is Not Used Is a Waste
• Operational Systems Will be Used
Because they are mission-critical for running
the organization
• DSS / EIS / BI Systems ???
Will not be used unless the management
process is driven by these systems
Dr. Lakshmi Mohan
40
To Get Payoff From Analytical Systems …
Raw Data
Analytical System
Actionable Information
ACT !!
Dr. Lakshmi Mohan
41
Two “Big” Factors
Affect Use of Analytical System
1. Organization Culture
“Business as Usual” - Complacent Culture
versus
“ How Can We Improve”
2. Management Style
“Left Brain” - Analytical
versus
“Right Brain” - Intuitive
Dr. Lakshmi Mohan
42
Methodology for System Development
Is Different!
DATA
PROCESSING
OUTPUT
(1)
(2)
(3)
TPS
HENCE , BOX (1)
EASY TO SPECIFY
MOST EMPHASIS
ON BOX (2)
BOX (3) IS
WELL-DEFINED
PRECISE
LEAST
IMPORTANT
ILL-DEFINED
FUZZY
DSS/EIS/BI
?
Dr. Lakshmi Mohan
43
Standard Method for System Development
Sequential Approach
Analysis
Design
Development
Implementation
Project
Start
Project
End
• Suitable for structured systems
• Because outputs of the system are easy to specify
Not Effective For DSS / EIS / BI
• Because information needs cannot be completely
defined at the outset during the Analysis Phase
Dr. Lakshmi Mohan
44
The Systems Development Life Cycle (SDLC)
Project
Identification
& Selection
Project
Initiation
& Planning
Analysis
Logical
Design
Physical
Design
Implementation
Maintenance
Dr. Lakshmi Mohan
45
An Axiom For BI Systems
Distinguish Between
Things Which Management
MUST Do AND Know
versus
Those Which Would be Merely
Nice to Do and Know
Dr. Lakshmi Mohan
46
Design of the BI System
• A Common Approach:
System Has Everything.
– Too Many Options will Overwhelm the User
– High “Intellectual Cost” to Use the System
• What is Needed:
- Not Over-Designed!
- Must Enable News in the Data to be
Quickly Gleaned
Dr. Lakshmi Mohan
47
A User-friendly System
 No Training, No Manual
 "Bomb-proof"
 Invites Usage
 Intuitive Paths to Navigate
Dr. Lakshmi Mohan
48
Problems with the Standard Approach
for Systems Development
 Interview users to define requirements
 Danger: Long "Wish-list"
 Build system to specifications
What users say they want
is not
What they actually need
Dr. Lakshmi Mohan
49
Interviewing Executives
Is Difficult Because ...
•
No time or patience to think through
•
Unable to articulate requirements
"Use a lot of soft information ...
Hard to know what to tell you"
•
Vague about their needs
"I want instant access to all relevant data"
Dr. Lakshmi Mohan
50
Prototyping - "A Must"
WHY. . .
 A live system with real data
 Users can "test-drive" it
 Constructive feedback on system design
A cost-effective means of ensuring value
of system before making the investment
on its development and implementation.
Dr. Lakshmi Mohan
51
The “Q & D” Prototype
- A "Quick and Dirty" System
- For the “GO” or “NO GO” decision
- To determine user needs
- To ensure value of system
- Low-cost System to Reduce Project Risk
- Yet should spark user interest
- Must use Real Data
- To stimulate users
- To check out potential data problem
- Modify on basis of User Feedback
Dr. Lakshmi Mohan
52
Value of Prototyping
Only Means of Ensuring that System
Design Meets User Needs
• Produces a "live" system rather than a voluminous
"paper“ system usually written from a technical
viewpoint
• Allows users to test-drive the system and see how it
works rather than imagine its operation
• Facilitates constructive feedback from users about
features they like in the system and modifications to
make it more useful
• Enables system to evolve nearer and nearer to users'
needs after three to five iterations
Dr. Lakshmi Mohan
53
Evolutionary System
Development Methodology
PROTOTYPE
INITIAL SYSTEM
USER
REACTION?
“No Go”
END
“Go”
USER FEEDBACK
FOR EXPANDING
SYSTEM
CONVERT INTO WORKING SYSTEM
<VERSION 1.0>
PILOT TEST <VERSION 1.0>
IMPLEMENT <VERSION 1.0>
Dr. Lakshmi Mohan
54
The Iterative Approach for DSS/EIS/BI
• Compresses all the four phases into
a short cycle
• System evolves through a series of iterations
• Enables users to specify information needs
in concrete terms
• Because they see actual outputs with live data
from the initial versions of the system
Dr. Lakshmi Mohan
55
Benefits of the Evolutionary Approach
 The system evolves through a series of
iterations in short cycles, each of which
results in usable versions of the system.
 New features, new data and new users are
added from user feedback.
 The best way to build a big system is not to
build one.
Start Small and Let System Evolve
Dr. Lakshmi Mohan
56
Features of Analytical Systems
• Access & Reporting
– Standard Reports: What Happened?
– Query / Drill-Down: Where Exactly is the Problem?
• Analysis Capabilities
– Statistical Analysis: Why is This Happening?
– Predictive Modeling: What Will Happen Next?
– Optimization: What is the Best We Can Do?
Analytics integrate…
… Data, Statistical Tools & Models
… With Supporting Hardware & Software
… To Drive Problem-Finding, Problem Solving
… AND Decision Making
Dr. Lakshmi Mohan
57
Tools to Get Value from Data Warehouses
Business Intelligence Tools
To enable users without programming skills to
analyze the raw data in the data warehouse.
Ad Hoc Query / Reporting
OLAP Tools to “slice” and “dice” data.
Data Mining Tools
Automate the detection of patterns in the
data warehouse
Build models to predict behavior through
statistical and machine-learning techniques.
Dr. Lakshmi Mohan
58
Drilldown Example
Washington Metro
Level 1
Level 2
Level 3
Dr. Lakshmi Mohan
510
Virginia
D.C.
Maryland
3-Region Total
160
170
170
510
Circles
20
Wheat
40
Sugared
50
Spheres
50
Circles
10
Circles
60
Circles
90
Wheat
20
Wheat
40
Wheat
100
Sugared
90
Sugared
10
Sugared
150
Spheres
40
Spheres
80
Spheres
170
59
Slicing & Dicing a Data Cube
- Sales by Location
Dr. Lakshmi Mohan
Location
Sales
New York
$440
Chicago
$380
Dallas
$325
San Francisco
$245
60
Slicing & Dicing a Data Cube
- Analysis by Type of Socks
Sweat Socks
New York
Chicago
Dallas
San Francisco
Dr. Lakshmi Mohan
Blue Socks
New York
Chicago
Dallas Sales
$50
San Francisco
$100
$25
$40
Brown Socks
New York
Chicago
Dallas Sales
$120
San Francisco
$40
$160
$95
Black Socks
New York
Chicago
Dallas Sales
$100
San Francisco
$130
$15
$40
Sales
$170
$110
$125
$70
61
Slicing & Dicing a Data Cube
- Analysis by Location
Dallas
San Francisco
Black
Brown
Blue
Sweat
Dr. Lakshmi Mohan
Black
Brown
Sales
Blue
$70
Sweat
$40
$95
$40
Chicago
Black
Brown
Blue Sales
Sweat $125
$15
$160
$25
New York
Black
Sales
$170
Brown
Blue Sales
Sweat $110
$130
$40
$100
$100
$120
$50
62
Why Data Mining ?
“Now that we have gathered so much data,
what do we do with it?”
“The datasets are of little direct value themselves. What is
of value is the knowledge that can be inferred from the
data and put to use.”
 Data volumes are TOO BIG for traditional DSS Query/
Reporting and OLAP tools.
 Organizations have to get value from the huge
investments of time and money made in building data
warehouses.
Dr. Lakshmi Mohan
63
Why is Data Mining a “Hot” Topic Today?
1. Implementation of ERP, CRM & SCM systems have resulted in
vast stores of operational data.
2. Emergence of global competition has put the pressure on
companies to be “data- driven” – i.e., make informed decisions
based on facts and not hunches.
3. The speed of change in the marketplace demands that the pearls
of actionable information have to be found faster in the ocean of
data, for companies to be one step ahead of competition.
4. The hardware needed to store and process a “ton of data” was
prohibitively expensive until recently – “You would have had
to have NASA at your disposal”.
Today, the technology makes it feasible to apply complex
models to ferret out patterns left to rot in “data jails”.
Dr. Lakshmi Mohan
64
The Payoff from Data Mining
- An Example: Farmer’s Insurance
Based on traditional data analysis, drivers of
sports cars were determined to be at higher risk
for collisions than drivers of “safe” cars such as
Volvos.
Hence charged them more for car insurance.
Data mining discovered a pattern that changed
the pricing policy…
… As long as the sports car was not the only
car in the household, the driver fit the profile of
the “safe” family car driver, not the risky sports
car driver
Dr. Lakshmi Mohan
65
Data Mining Techniques
- Decision Trees
 Derives rules from patterns in data to create a hierarchy of IF-THEN
statements, called a Decision Tree, to classify the data.
 Segments the original data set:
 Each segment is one of the leaves of the tree
 Records in each segment are similar with regard to the variable of
interest
Example: Classification of Credit Risks
Dr. Lakshmi Mohan
66
Decision Tree for Segmenting Customers
- Who Responded to a Marketing Campaign
Overall : 7% of Customers Responded
Segment of Customers Who Rent with High Family Income
and No Savings A/c : 45% response
Target this Segment for Future Direct Marketing Campaign
Dr. Lakshmi Mohan
67
Text Mining: An Imperative Today
“We are drowning in information,
but are starving for knowledge”
Unstructured data, most of it in the
form of text files, typically accounts for
85% of an organization's knowledge
stores, but it’s not always easy to find,
access, analyze or use.
Dr. Lakshmi Mohan
68
Case Example: Honda
Instituted An Early Warning Program
•

−
−

To Identify Major Potential Quality Issues …
From Warranty Service Records
Records Sent to Honda by Dealers
Included Categorized Quality Problems AND Free Text
Transcripts of Calls by Mechanics to Experts in Various
Domain at Headquarters
 Transcripts of Customer Calls to Call Centers
• Mined the Text Data from the Different Sources
− e.g., Words appearing for the first time, particularly
those suggesting major problems, such as fire
− Flagged for human analysts to look at
Source:
Davenport & Harris, Competing an Analytics, page 70
Dr. Lakshmi Mohan
69
Another Example: HP
- Adopted SAS Text Miner Software
 Textual Analysis of Comments by Customers in Call
Center Records
− “Customers who were really loyal were talking to the
Call Center about different things than Customers who
weren’t so loyal, or Customers who did not buy as
frequently or in as high a volume”
 Lead Classification Based on Textual Notes Collected
from an Initial Call Center Contact
− Divide New Leads into Cold, Medium and Hot rankings
− 80% success rate
i.e., the leads performed as predicted,
when the leads were passed to the sales staff
Source:
“7 Strategies for Profiting from Customer Data”, Destination crm.com, July 1, 2004
Dr. Lakshmi Mohan
70
Who is An Analytic Competitor ?
• Companies who have built their businesses on their ability to
… Collect Data,
… Analyze It, AND
… ACT On It.
• Sign on Desk of CEO
“In God we trust; all others bring data”
• Analytic Competitors:
–
–
–
–
–
–
Consumer Products: Frito-Lay, P&G
Financial Services: Capital One, Royal Bank of Canada
Retail: Wal-Mart, Tesco, Amazon
Transport: FedEx, UPS, Schneider National
Industrial Products: CEMEX, John Deere
Hospitality & Entertainment: Marriott, Harrah’s Entertainment
Dr. Lakshmi Mohan
71
Why Compete on Analytics?
• Geographical Advantage
… Does not matter in global competition
• Protective Regulation: Largely Gone
• Proprietary Technologies: Rapidly Copied
• High-Performance Business Processes
… Last remaining points of differentiation
… Execute your business with maximum efficiencies
… Make the smartest business decisions possible
Dr. Lakshmi Mohan
72
The Battle for Credit Card Customers
… Capital One Case Example
Winning Big in the Cut-throat World of Credit Cards

IPO in 1994

Revenue: Exploded from $95 M in 1995 to $4.97 B in 2000

2001 Year End:
−
Posted its 18th consecutive quarter of record earnings
−
Annual earnings rose by over 20%; Yearly ROE: Over 20%
−
43.8 M customers worldwide, Over 20,000 employees

Consistently Outperformed First USA, twice its size
−
Earning 40% more interest income
−
Enjoying double the profit margin
Secret To Its Success: “Information-Based Strategy”
Dr. Lakshmi Mohan
73
“Credit Cards Are Not Banking
- They are Information”
It’s all about collecting information on millions of people
that you’ve never met, and, on the basis of that
information, making a series of critical decisions about
lending money to them, hoping that they pay you back.
Each customer carries a specific and unique credit risk
and potential revenue profile, based mainly on their
previous credit history (or lack thereof). The better the
company can understand and assess a customer’s
specific risk, the better it can manage it.
AND, the better it understands the customer, the more it
can tailor its products to his or her needs.
Dr. Lakshmi Mohan
74
Business Model of Capital One
MISSION: Deliver the Right Product, At the Right Price,
To the Right Customer, At the Right Time
UNIQUE INFORMATION-BASED-STRATEGY:
When we started this company, we saw two revolutionary
opportunities: We could use scientific methodology to help us make
decisions, and IT to help us provide mass customization.
Foundation of Capital One: TEST AND LEARN
We test every product offering, every procedural change, every job
applicant. We record every customer interaction, every card purchase;
and then, with the patience of a good scientist, we run experiment
after experiment. For every action taken, we know what the reaction
has been. If we have sent you a blue envelope or a pink one, we know
which one you received, and how you reacted to that.
- Ran 45,000 Tests in 2001, Average of 120 per day
Dr. Lakshmi Mohan
75
“Information-Based Strategy”
- A Three-Step Approach
• Step 1:
– Create an idea for a new product offering
– Find a target population and a business case
– Test the idea with this group to see how they react
• Step 2:
– Gather data on the test and analyze results
• Step 3:
– Use test results to identify which people are most receptive to the
product offering
– Conduct the marketing campaign based on micro-segmentation
Example: Used IBS to track visitor’s activities and offer customized
promotions on its Web site – studied which online visitors it has
successfully converted into customers – used that information to buy
banner ads on other web sites whose visitor demographics match
those of its ideal customers
…Doubled its goal of opening 1 million new accounts online
Dr. Lakshmi Mohan
76
Mass Customization of Credit Cards
- Cap One’s “Invention Machine”



−

−
−
−


U.S. Credit Card Market in the 1980s - “One size fits all”!
Capital One Changed the Rules!
“Tailor the product to meet the customer’s needs.”
2001: More than 6,000 Products
Variations of Credit Cards
… Annual Percentage Rate, Credit Limits, Fees, Designs, etc
Examples:
No-fee Mercedes-Benz affinity card with a credit line of $20K
$29 a year fee for a card with just $200 worth of credit
A credit card with a Canadian moose on it. Or, a card with a map
of Japan and an image of Mt. Fuji on it
Other Related Products
… e.g., Card Protection Plans, Payment Protection Insurance
Other Financial Services … e.g., Travel Insurance
Dr. Lakshmi Mohan
77
Intelligent Call Routing
- Implemented in 1998

# of Calls from Customers: Over 200,000 a day

The Moment the Last Digit is Punched:
−
−
−
Caller is Identified; About two dozen items of data Analysed
Predict the Reason for the Call
ALSO, What the Caller Might Want to Buy
… even though he or she isn’t calling to buy anything
Select Best of 50 Call Routing Options for This Caller
−
−

−

−
−
Display the Relevant Info, including the Script for the CrossSell Recommendation, on Rep’s Screen
ALL BEFORE the Call Arrives in the Head-Set
Just 100 milliseconds, one-tenth of a second
… one-eights of the time between human heart-beats
How Good Is It?
Right 40% of the time initially; 1999: 60% to 70%
And, System just keeps getting SMARTER!
Dr. Lakshmi Mohan
78
How It Works – An Example
When a customer calls, system channels poor prospects
to a voice-response unit and even allows them to close
their accounts…
Others are routed to a service rep along with
information about the card holder and the
likely reason for the call with a script to deal with it.
If customer wants to close the account, the system will
display three interest rate counter-offers.
Service rep has the freedom to negotiate, and gets a
bonus if customer is persuaded to stay on at the highest
of the new rates
Dr. Lakshmi Mohan
79
The Routing Software
- Cisco’s Global Service Logistics System
One of the few shrunk-wrapped application used by Cap One
… Most software custom-built in-house
Everyone will say they use GSL the same way we do, but I think we
use it more intelligently than they do.
We use many more attributes in judging where the call goes.
And we gather more data about that call than anyone else does
AND, we use that data as a basis for creating decision rules in our
applications. - VP of Customer Relations
Examples:
1. Do you routinely call from your boyfriend’s phone – the number
for which is not on file at Capital One? Eventually, the computer
will figure out that his number should be in your CIF.
2. What language do you prefer to do business in? System will learn
that and route calls accordingly.
Dr. Lakshmi Mohan
80
Impetus for the Project
- The High Phone Bill



−
−


We pay for in-coming Customer Calls.
Calls were taking too long to handle.
Analysis showed that Customers were not to blame.
Calls simply were NOT getting to the right place soon.
Caller with a lost card or fraud problem ended up reaching an
ordinary Rep. People who just wanted to know their balance
stayed on hold to talk to a live Rep.
People unhappy with their interest rate called the “lost card”
number on the back of their card and had to be transferred to
customer service.
All that time – to take a call, to bridge the call to the right person –
that annoys the customer, we are paying for the call. You wait for
an agent, you wait for a transfer, you wait again for an agent.”
Even one extra second per call adds up to real money with over a
million calls a week.
Dr. Lakshmi Mohan
81
How to Lower the Phone Bill ?


Tried lots of options - But nothing worked?
Example:
Some people called much more often than the average of 5
times a year …
We sent out a letter at one point that said, in effect:
“Please don’t call so much” …
But it did not work !
If you want people to call you, send them a letter telling
them not to!
Ultimate Solution Suggested by IT
Why not predict the reason for each call
AND
then send that call to the agent
who is best able to handle it.
Dr. Lakshmi Mohan
82
Implementation of the Intelligent Call
Routing Infrastructure
1. Analysis of Why People Call:
90% of all calls fell into one of 10 categories
− Raise your customer’s interest rates, and they call
− Send out a new card that has to be activated; they
call.
− Same people call once a month to find out their
credit balances; some others call three times a
month
2. Decision-tree Software had to be written
3. Computers, phone switches and telecom networks
had to be taught to talk to one another
Dr. Lakshmi Mohan
83
Examples of Automated Voice Response
Example 1: Customers who call each month to check
their balance are routed to an automated system that
answers the phone this way: “The amount now due on
your account is $364.27. If you have a billing question,
press 1 …”
Example 2: Customers who call to check if their
payment has arrived could be identified and the phone
message would then be: “Your last payment was
received on February 9. If you need to speak with a
service Rep, press 1 …
Dr. Lakshmi Mohan
84
Everyone Wins !

“We can answer your question BEFORE you ask it!”

“A phone call that might have taken 20 or 30 seconds, or
even a minute, now lasts 10 seconds.”

Customers get where they are going immediately.
And, they get the information they need quickly.

Customer Service Reps handle those calls that need to be
handled by people, and they don’t waste any time passing
calls to colleagues.

Customers are automatically routed to the RIGHT Reps –
best skilled to not only deal with the problem about which
the customer is calling but also to cross-sell the product
that the system predicts the customer might want to buy.
Dr. Lakshmi Mohan
85
The Pay-Off for Capital One
- Calling System Has Become A Competitive Advantage


−
−
−
−
Lower Costs AND Better Service
Call Centers: NOT A COST CENTRE
Generate Revenues from Cross-Selling
Exceeds Cost of Operations
Actually MAKE MONEY!
“In 1998, for the first time, half of all new Cap One
customers bought another product from the company
within 12 months of signing up for their credit card. That’s
amazing penetration and it leads to high profitability.”
A simple, routine problem in search of a quick solution led
to a whole new way of doing business. It enabled us to go
back to the business side with a solution that went beyond
that problem. What makes our “T” work has nothing to do
with “T” – it has to do with our culture.”
Dr. Lakshmi Mohan
86
Every Interaction Is
A Selling Opportunity !
 Most credit-card companies, including Capital One,
have long tried to “Cross-Sell” their customers
– often by using inserts in monthly statements to
tout everything from calculators to cruises
 Data analysis of outbound telemarketing calls (made
usually at dinner time) showed it was NOT working.
 New Idea: Sell things to customers when they call
− “It seemed like a natural. If you call me and I’m
trying to sell you something, then I’m going to treat
you very nicely. That will promote better service.”
Dr. Lakshmi Mohan
87
Implementation Issues
− Service people are not sales people.
− Systems cannot service people and sell them
at the same time.
− Even with systems in place, sales will not be
enough to make longer phone calls
worthwhile.
“It can’t be done.”
That’s all the motivation we needed.
Dr. Lakshmi Mohan
88
Made the New Idea Work !
 First Test Product: Easy & Sweet
“When new customers call an automated line to
activate their card, we thought: That’s the perfect
time to sell them something
The first thing we sold was a balance transfer: “Now
that you’ve got our card, is there any debt that you
want to transfer to us.”
Customers just bought it!
 Brainstorming Session: Manager of Cross-Sell
Marketing, IT Head and Call-Center Manager
“How to extend quickly to other products?”
Dr. Lakshmi Mohan
89
Human Hurdle
Service Reps Reluctant to Sell Products
 The Solution: “Elevate this beyond the
immediate transaction level. If I am a phone
associate, my mission is to meet my customer’s
needs. If I’ve got this great product, it might
save a customer some money, or it might create
convenience. If I’m committed to service, I
should consider offering that product.”
 In 3 months, Reps started to both service and
sell during field calls.
Dr. Lakshmi Mohan
90
Management Process Ties It All Together
at Capital One
•
Organization Structure of Capital One

By market segments based on…
… Customers’ credit quality
… Activity with the card, etc.
•
•
Each Segment is a Profit Center

Head has the autonomy and team to run the operation for
that segment like a small business

Enables opportunities to be sensed from the bottom-up
and pursued quickly by motivated employees
Reward System Fosters Shared Values and Collaboration

Low turnover rate: 5% per year for customer-contact
employees vs. industry average of 15% - 20%

Improves service and helps keep costs down
Dr. Lakshmi Mohan
91