Transcript Ideal Direct Marketing Execution

Data Analytics:
Lessons From Competitive Electric
Markets
Smart Grid Research Consortium Conference
October 20-21, 2011 Orlando, Florida
Tom Atkinson
Ideal Direct Marketing Execution
Consultants, Inc.
The Futures Company 2011 Research on the
State of Smart Meters
• Key Motivators to leading a more environmentally conscious life:
▫ I want to preserve the world for future generations
▫ I feel it’s the right thing to do
▫ I am concerned about my own health and my family’s health
▫ It enables me to save money
• Key Barriers to leading a more environmentally conscious life:
▫ It costs too much
▫ I don’t believe that products that claim to be “environmentally friendly”
are really that helpful for the environment.
▫ Even if I do my part, other people will never do theirs
▫ I have too many other things to be concerned about
IBM 2011 Global Utility Consumer Survey Provides
Insights on Customer Characteristics & Diversity
• Influences
▫ Saving money
▫ National economic considerations
▫ Environmental
• Sources of information
▫ Bills/Inserts
▫ Traditional media/ internet-social media/ friends & family
• Knowledge
▫ Most consumers lack knowledge of electric terms
 Unit of cost
 Smart grid/smart meter
 Time of Use
We should invest
more in the 42%,
but how do we
identify them?
• Expectations
▫ 42% committed to engaging with providers to meet personal goals
▫ 33% not likely to take added responsibility for these decisions
Consumer Resistance
• Consumers are fed up with marketing saturation
▫ National Do-Not-Call Registry, DVR’s, TV remotes, etc.
• Better marketing is not better promotion of a brand
• Better marketing is a better brand
▫ i.e. Customer centricity in both word and deed
• Marketing must be more relevant and precise
The Smart Grid Customer Data Analytics
Challenge
• Customer program success depends on
▫ Extent of information on customers
▫ Information on how customer characteristics relate to
program objectives
▫ Execution of programs utilizing customer information
• Costs and benefits of customer programs vary with the
extent of effective customer focus
The Unfocused Approach
• We want our customers to allow us to control their
thermostat during peak demand
• Let’s send them all a letter explaining why that is good
for them and the cooperative
• Cost:
▫
▫
▫
▫
▫
▫
100,000 customers
Cost per letter = $2.00
Total cost = $200,000
Positive response rate = 2%
Number of customers accepting offer = 2,000
Cost per customer accepting offer = $100.00
The Unfocused Approach:
AKA Spray and Pray
&
The Focused Approach
• Let’s mail a small sample and build a model to predict
response
120%
• The model captures 90% of the
responses in 60% of the
contacts
100%
• Cost:
Percen of Positive Responses
Cumulative Gains Chart
80%
• 60,000 letters @ $2.00 =
60%
• $120,000.00
Random
40%
Model
• 1,800 customers
• $66.66 per customer
20%
0%
Percent of Customers C ontacted
This looks
interesting!
Data Quality
• Coverage
▫ Is there a entry for every possible customer?
• Completeness
▫ Are essential facts known?
• Accuracy
▫ How accurate is the information?
• Timeliness
▫ When was the information last verified?
• Predictive
▫ Is the information predictive of behavior?
• Value
▫ Does the benefit justify the cost of the information?
What Customer Data is Available?
•
•
•
•
•
•
Billing data
Property tax records
Census demographics
Demographics from compiled databases
Credit bureaus
Other behavioral/attitudinal data
Billing Data
• Historical Usage
• Program Usage
▫ Balanced Billing
▫ Product History
• Payment History
▫ Disconnect Notices
▫ Late Payments
• Contact History
▫ Calls
 High bill complaints
 Payment problems
 Other concerns
Property Tax Records
•
•
•
•
•
•
•
•
Owner occupied status
Square Footage
Year built
Number of stories
Pool
Available from County Assessors Office
Track ownership changes through County Recorder
Consider using a third party source like CoreLogic
Census Demographics
• Census Demographics
▫
▫
▫
▫
▫
Population
Housing
Employment
Education
Income
• Addresses are linked to census geography in an address
matching process known as geo-coding
• Demographics are published by U.S. Census Bureau
• Also available from multiple commercial providers
Compiled Databases
• Many providers
▫ Acxiom, Epsilon, Experian, InfoUSA
• Limited facts
▫
▫
▫
▫
Name/address/telephone
Age of adults
Presence of children
Income estimates
Aggregated Credit Data
 Credit data aggregated to a neighborhood level – ZIP+4
 Neighborhood attributes within the following categories:





All Trades
Bankcard
Dept. Store
Mortgage
Personal Finance
Automotive
Credit Union
Bankruptcy
Collection
Sales Finance
 Available from Epsilon or Equifax
MindBase® Customer Segmentation
•
Built off the longest running, most in-depth tracking of consumer values
and lifestyle – the Yankelovich MONITOR study.
•
MindBase® is an innovative marketing tool that enables businesses to
answer the questions “Who will buy?” and “Why?”.
•
MindBase® identifies eight consumer segments that span all four
generations: Matures, Boomers, GenX and Millennials.
•
Segment drivers include perspectives on:
•
•
Level of materialism
•
Orientation to technology
•
Involvement and interest in family
•
Conservatism
•
Social interaction
•
Optimism about the future
MindBase® is optimized for direct marketers - By tying researched,
fundamental attitudes to specific names and addresses in a client’s
database, clients can make more powerful and pinpointed connections with
their customers and prospects
Source: the futures company
MindBase® is the
only marketing tool
that links fundamental
consumer attitudes
and motivations to
specific names and
addresses in
databases.”
Who are the MindBase Segments?
15%
Expressive
“Carpe
Diem”
15%
I live life
to the fullest
and I’m not
afraid to
express my
personality
8%
Down to Earth
“Ease on
Down the
Road”
I’m cruising
down life’s
path in my own
way, seeking
satisfaction
where I can
Driven
“Nothing
Ventured,
Nothing
Gained”
13%
Sophisticated
“Sense and
Sensibility”
9%
My life is
very busy
and I’m
looking for
control and
simplification
I’m ambitious
with a drive to
succeed
personally
& professionally
20%
At Capacity
“Time is of
the Essence”
8%
I am
intelligent,
upstanding
and I have an
affinity for
the finer
things in life
Source: the futures company
Measure Twice
“An Ounce
of Prevention”
I’m a
thoughtful
planner and I
seek both
actualization
and fulfillment
Rock Steady
“Do the
Right Thing”
I think of
myself as
dependable
and I try to lead
a positive life
11%
Devoted
“Home is
Where the
Heart Is”
I have
traditional
values and I
enjoy the
comfort and
familiarity of
my home
Possible Database Design
How is Data Used in Developing and Executing
Programs?
• Use data for education
▫ Explain why electric costs vary by time of day
▫ Show the benefits of peak reduction
 Savings (for the customer and for the cooperative)
▫ Personalized customer usage reports
• Probability of program participation
• Personalize message to be more relevant
Peer Usage Communication
• Email/web report showing customer usage compared to peers
• Latitude-Longitude/Tax Data/Demographics used to identify closest
customers to subject with similar housing and family size
characteristics
• Natural response is to modify behavior to lower usage
Segmentation and Modeling
• Segmentation and predictive modeling produce the
greatest return on investment of any of the direct
marketing strategies
• Knowing what data may correlate to the dependent
variable and where to source the data is critical to
success
• Demographics alone are poor predictors of behavior
• Consider adding behavioral and attitudinal data to your
analysis
Customer Program Model
• Goal: customers switch to a time of use pricing plan
• Situation:
▫ You have developed three creative messages
 Savings, national economic concerns, the environment
▫ You have developed your marketing database to include MindBase
segmentation and Neighborhood Credit Attributes
• Experiment:
▫ Which message elicits the greatest response for each MindBase segment?
▫ Conduct direct mail experiments
 3 messages X 8 MindBase segements = 24 combinations
▫ Refine your audience selection using multiple regression to forecast
response and further reduce the size of the audience
▫ Conduct direct mail campaigns and continue to test additional
treatments
Competitive Market Example: Deposit
Decision
• Permissible Purpose
• Correct initial decision is critical
▫ Require deposit from credit worthy
 In competitive market, likely to keep shopping
 In regulated market, unhappy customer
 Utility pays unnecessary interest on deposit
▫ Fail to require deposit from credit un-worthy
 In competitive market, eagerly accepts offer
 Future bad debt
• Segmented Credit Policy
• Retrospective Analysis
• Pre-approved Vs. “Invitation to Apply”
Segmented Credit Policy
No
Deposit
More
likely
to buy
Bad Payers
Good Payers
720
85% of bad debt
Customers with unacceptable risk acquired
650
Deposit
Less
likely
to buy
Customers with acceptable risk not acquired
590
85% of bad debt
Retrospective Credit Analysis
• Conduct analysis with multiple credit bureaus
• Sample of good and bad payment customers
▫ Include as much known data as possible for future modeling
▫ Include bad debt amount
• Append multiple model scores from bureaus
▫ Bureaus return an analytic file with no Personally Identifiable Information
• Analysis
▫ Kolmogorov-Smirnov Test used to select best models
▫ Segment by payment behavior (possibly: renters vs. owners)
▫ Rank on model score within segment and determine where you identify the
target percent of bad debt you need to protect with a deposit
• Consider using two bureaus whereby those that narrowly fail
at the primary bureau are scored by a second bureau
Pre-screen Vs. Invitation to Apply
• Pre-screen
▫ Advantages
 Pre-approved may have positive impact on response
 Low cost per pre-approved prospect
▫ Disadvantages




Low response rates result in higher cost per acquired customer
Onerous FCRA disclosure language may discourage response
Must make a firm offer to all that qualify – limits campaign flexibility
FCRA opt/outs are eliminated (20-25% of the population)
• Invitation to Apply
▫ Using only company data as predictors, model probability of passing real-time credit
▫ Advantages
 No required FCRA disclosure: possibly resulting in higher response
 Larger prospect universe because FCRA opt/outs not eliminated
 No requirement of a firm offer – greater campaign flexibility
▫ Disadvantages
 Payment to bureau is on a per/inquiry basis and is a higher unit cost than pre-screen
 However, cost per acquired customer may in-fact be lower due to low response rates
 Cannot claim prospect is pre-approved
Competitive Market Application:
Acquisition Model
• Three characteristics of a “best” electric customer
▫
▫
▫
The prospect is “in the market” and more likely to accept your offer
The prospect’s electric usage is above average
The prospect is not likely to exhibit bad payment behavior
• A past campaign forms the basis of the analysis
▫
▫
Ideally, a random population with no selection bias
Response is the dependent variable
• Is the prospect at the anniversary of a term agreement?
▫
▫
▫
Home purchase date from tax records
Former customer defection date
Other possible sources of contract end dates
• Is the small savings enough to motivate a switch?
▫
▫
Prospects that struggle with cash flow are more likely to switch
Epsilon Target Neighborhood Credit Attributes – Geographic aggregation of credit variables
• Consumer attitudes
▫
▫
Promotion History - Frequent past promotion has negative correlation to response
the futures company Mindbase segmentation system
Competitive Market Application:
Usage Model
• Normalize historical usage using historical weather data
• Key predictors
▫
Property tax records



▫
Census demographics


▫
Fuel used for heating
Age/Income/Family size
Normal Weather


▫
Year built
Square footage
Market value
Heating degree days
Cooling degree days
Household demographics

Age/Income/Family size
• Develop/validate models
• Expectations
▫
▫
R2 in the range of 0.60-0.70
Compare forecast usage decile to actual decile
Access and Campaign Execution
• Marketing should be equipped with Campaign
Management software to easily plan and execute multichannel campaigns
▫ Companies that avoid these tools often find themselves slow to market
and overspending on highly technical staff
• Among the leaders identified by Forrester are SAS,
Unica, Alterian, Aprimo and Siebel
Use analytics to Test – Test - Test
•
•
•
•
•
Offer
Incentive
Creative
Message
Design of Experiments
▫ Two alternative design
 Simple null hypothesis: treatment A Vs. treatment B
▫ Factorial Design
 Tests all possible combinations
 Offer, Incentive, Creative, Message, Population Segment
Application to Electric Coop and Public
Utility Markets
• Optimize the ROI of customer communications using data and
analytics to target those most likely to respond to your offers for:
• Cost containment program development
▫ Direct load control
▫ Pricing
▫ Programmable communicating thermostats
• Competitive considerations
▫ Provision of services offered by others
 Security, appliance purchase and maintenance, etc.
▫ Future service offerings by others
 Solar installations
 MicroCHP (Honda 1kW unit)
Conclusions:
•
•
•
•
•
Build a marketing database
Focus on data quality
Use Analytics to maximize Return on Investment
Use data to educate customers
Test, Test and Test some more
It won’t be easy, but
data and analytics will
get us where we need
to go.
Contact Information
• Phone: (713) 828-6396
• Email: [email protected]
• Linkedin Profile:
http://www.linkedin.com/in/tomatkinsontx
About Us
Tom Atkinson
Tom is a direct marketing operations professional. Tom worked at Reliant Energy in
Houston Texas for eleven years, most recently as Director of Database Marketing
Development. He selected the providers of marketing information and managed those
relationships for Reliant as well as providing subject matter expertise in the
development of their marketing database. Tom organized a retrospective analysis to
improve the credit policy for residential and small business. He selected sources of
specialized data by demonstrating, along with the statistical analysts, how the data
could improve the power of response models. Tom demonstrated the power timing of
solicitations to the likely “last decision” anniversary has to lift response. Earlier in his
career at Reliant he managed the campaign list development team.
Prior to joining Reliant, Tom was the Director of Data Acquisition for Donnelley
Marketing, now part of the InfoUSA group of companies. In addition to negotiating
significant agreements with data partners, he provided thought leadership and
business rules for how data was to be incorporated in the Donnelley consumer
database. Early in his twenty-two year career at Donnelley, he directed the team of
software developers responsible for maintenance of the Donnelley consumer database.
He earned a Master of Arts degree in mathematics from the University of South
Dakota with an emphasis on statistics, operations research and computer science.