Data and Text Mining

1

• Data and Text Mining
• 2006 Digital Now
• Orlando, FL
• Kevin Whorton, Principal
• Whorton Marketing Research
• Columbia, Maryland
• info_at_kwhorton.com

2
Overview

• Case Studies SHRM, TMS, CRS
• General Principles

3
SHRM
4
SHRM Business Intelligence Implementation

• Ongoing, 4-year, iterative process
• Required extensive up-front buy-in and definition
  building
• Forward-thinking core issues driving the BI
  Environment
• What problems are we trying to solve?
• What questions are we trying to answer?

5
SHRM User Audiences for BI

• Different layers of interaction
• Executive/C-Suite
• Business leaders/Program Managers
• End consumer/Operational-level users
• These groups will have different perspectives on
  what the BI Environment is and what it does, and
  will create different motivations for its growth

6
SHRM Interfaces/Tools

• Microsoft DTS as the ETL tool
• Cognos BI Toolset for reporting and analysis
• Embedded in MS Sharepoint
• Designed as a one stop shop for business
  intelligence

7
SHRM Status and Lessons Learned

• Outlay has been substantial (over 2MM)
• ROI has been good
• Learning experiences
• Starting with "empowered users"
• Leaving with somewhat defined queries to ensure
  consistency of information
• Information "depth of use" much greater

8
TMA
9
Texas Medical Association Data
Warehouse/Mart/Mining

• Data warehouse
• Organize the information scattered among
  different sources and store it in a data
  warehouse.
• Extract scattered/incompatible data from
  different sources, transform by cleaning, making
  consistent
• Data mart
• More specialized cut from data warehouse
• Source files of transactional records
• Contain all data needed for related group of
  analyses
• Data is summarized in tables at appropriate
  levels
• Data Mining
• Enhance decision support by adding tools to
  access/analyze contents
• Process of selecting, exploring, and modeling
  large amounts of data to uncover previously
  unknown patterns

10
Comparison Traditional, New Data Management

• Current Report Environment
• Are your reports the same as what you used 5
  years ago?
• Do they lead you to ask new questions?
• If you think of new questions, do you ever get
  the answers?
• Decision Cubes
• Ask questions and receive immediate answers
• New information views, prompting new questions
• Compare historical information

11
TMA Decision Cubes

• Multi-dimensional data representation
• Can be viewed from different perspectives.
• Enables manipulation of parameters
• Derive metrics about your operations/association
• Displays results immediately
• Includes graphs and charts directly from the
  data

• .

A cube aggregates the facts in each level of each
dimension
12
TMA Implementation

• System Requirements
• Microsoft SQL Server includes analysis services
  as part of license
• If unfamiliar, technology staff probably hasnt
  installed it!
• Up-front costs
• Direct cost of one week of decision cube
  consultants time - 8,000 to 10,000
• Indirect cost for IT staff 2 weeks effort up
  front, ongoing future enhancements

13
TMA Analytic applications

• Applications enable users to access and
  manipulate warehouse data for better-informed
  decisions
• Demand forecasting, pricing, competitor analysis
  and customer segmentation
• Tools ProClarity, Excel, Cognos
• Costs
• 700 to 35,000 ProClarity single user and
  browser based, multiple-user systems

14
ProClarity Illustration
15
TMA Analysis Excel Illustration
16
TMA Status and Lessons Learned

• Outlay also substantial
• Return has been even greater
• TMA's choice but associations can achieve same
  impact with just the data visualization tools
  described
• Sophisticated target marketing ensures greater
  acquisition
• Market penetration steadily increasing
• Value in drilling down 45,000 total physicians
  in TX
• Ability to tailor, promote targeted CME and other
  services
• Odd that no other associations choose to use
  these tools
• It really is easy to implement in far smaller
  associations

17
CRS
18
Need Driven by Market Size/Access
Total US - 273 Million
Total Catholic - 65 Million
Typically Attend Mass - 20 Million
CRS Aware 14 Million
Donors - 400,000
19
CRS Large-Charity Illustration

• Large repository of data to analyze
• Contacts 12MM acquisition, 8MM house annual
  contacts
• Categorized by content, media, vehicle
• Response 750k gifts
• Data available vehicle (mail, phone, online),
  method of payment, one-time and monthly gifts
• Donations, not memberships varying amounts,
  upgrade/downgrade behavior
• Weaknesses/impediments
• Lack of strategic technical advisors
• Poor service donor management system vendor,
  staff DBMS
• Limited executive understanding of issues
• Hostile environment for direct marketing in
  fundraising/marketing mix
• Hard to hire experienced analysts in
  Baltimore/non-profit market
• Limited discretionary budgets for technology

20
CRS Assets Available to Us

• Solid report writing
• All campaign level results available by RFM
  segments (recency, frequency, monetary value)
• Existing staff very tactical, great memories
• Periodic investments in "snapshot"
  analyses/decision tools
• Target Analysis Group individual program
  assessment and benchmarking
• Amergent list life cycle analysis (LTV of
  acquired donors)
• Great latitude for action
• My charge "take 50 million program and double
  it in two years"
• Free to create new positions, retrain and retitle
  existing staff
• Strong research and branding support
• Disaster relief allowed us to make periodic
  major changes

21
"Oops" areas Focusing on the Doable

• Established staff of "data kids"
• Self-funding moved merge purge in-house to save
  250k per year in expenses
• Able to refine/diagnose archaic methods of
  campaign level data extraction over time
• Well-trained able to master, embed SQL queries
• Used research to provide behavioral insights
• Link giving to attitudes, motivations, other
  behaviors
• Necessary to generate new ideas, guide new
  campaigns
• Outsourced predictive models
• Worked with Genalytics to score new acquisition
  files, based on 40 million past acquisition
  contacts

22
Analysis Program Applications

• Ad hoc capabilities make new programs possible
• Upgrade/downgrade analysis
• Creation of custom gift arrays thorny issue with
  emergency giving
• All "asks" (gift arrays) built off historical
  giving
• Tsunami yielded average 280 temporary upgrade
• Record selection for mid-level programs
• Special, expensive multi-step direct mail
  campaigns intentionally asking for 3-5x highest
  prior contribution
• Originally based on seasonal and lifetime giving
• Over time, added overlays to measure capacity to
  give

23
Analysis Donor Management

• Strong tradition of test vs. control in direct
  mail
• Difficult to draw comparisons with house file
  mailings
• Examining mail frequency
• We mailed best house file names 24 times per
  year
• Attrition on an individual level vs. maximizing
  net revenue
• Applying "sweet spot" analysis to donors
• Mining complaint/comment data
• Donor research
• Focus groups to determine why people open read
• Laddering interviews to determine positioning
• Online panel survey of universe to segment
  acquisition market
• Surveying donors to define expectations,
  satisfaction

24
Improved Donor/Member Relations Jury Rigged
Use of "Interest Codes"

• Software allowed us to code member by
  "interest"
• Because we could export data, we categorized
  donors by many different variables
• After manipulating data, imported back into
  system
• Allowed us to be more flexible/responsive -
  extending our RFM selects- develop profiles of
  respondents to specific mailings- cross-tabbing
  donor types to learn more- proactively
  identify otherwise apparently good donors for
  de-selection

25
Analysis Areas Never Analyzed

• Integration of online data
• Over time 50,000 completely new tsunami givers
• Little link between online visits, e-newsletter
  reading
• Understanding true payback analysis
• Our technology investment paid itself back at
  least tenfold in first two years however
• Modeling the cost/effort/results of contacts with
  first-time donors
• Never understood what truly drives attrition of
  one-time givers
• Linking CRS behavior to world at large
• Rented names from Target cross-organization
  "model"
• Never understood "share of wallet" or reasons for
  defection

26
Traditional DM Analysis

• Graphing linear relationships finding sweet
  spots
• Very limiting if relationships are non-linear,
  changing over time

• And knowing when the relationships really are
  linear/predictive.

27
Other Analytical Tools Common Techniques
Answers

• Cross-tabulations Visual representation showing
  simple relationships between variables/segments
• Complex "grids" allow analysis, audience
  selection
• Correlations measures relationships between two
  variables
• Regressions most powerful tool
• Xf(x,y,z) or Membershipfunction of dues level,
  presence of competition, penetration, service mix
• R2 is a measure that explains relationship
  between one variable and everything driving it
• Helpful for projections and forecast models
• Logistic regressions allow prediction of yes/no
  outcomes
• Logarithms yield coefficients explaining
  percentage contributions
• Dummy variables measure seasonality, time
  trends, one-time shifts

28
Illustration Regression Analysis

• Example renewal program model avg response rate
  of 4.25, avg gift 36.25, revenue/name mailed of
  1.54
• Final equation is PRaRbFcMdO
• Or predicted revenue is a function of donors
  recency of giving, frequency, aggregate value,
  and other stuff"
• Final results for a sample donor who is exactly
  average is 1.54-0.068(6.5) 0.215(2.4)
  0.00465(156) 0.0087(85)
• Confusing, but once the formula is derived
• Real output scored file prioritizes every
  prospect, helps control your spend/return