Data & Marketing Analytics Theatre; Long time no see: using predictive modelling to reactivate long lapsed customers
1. Long Time No See
Using predictive modelling to win
back long lapsed customers
Data Based Marketing presented by: Paresh Patel
2. Agenda
• Long Lapsed Customers – The Theory
– Definition of long lapsed
– The information and data you need
• Long Lapsed Customer Marketing – Real World Case
Study
– The background
– The process
– The results
• Recap, hints and tips
• Q&A
Data Based Marketing
3. Long Lapsed Definition
• Lapsed customer:
“An individual or business who is no longer considered
active and no longer purchases from your company”
• Long:
“A significant amount of time”
• Examples
— Mail Order Customers who purchased over 37 months ago
— Charity Supporters who have last donated over 5 years ago
— Online customers who last ordered products over 2 years
ago
Data Based Marketing
4. Business Objective
• Product /Service offering or Charity ask?
• Relevant now as it was it when the customer was
active?
• Response or Value? Maybe both?
• What is the planned customer journey?
What Dataneed Marketing
you Based
5. Customer Data
1 Data quality matters Data = Better Results
Clean
• PAF Validation, name
verification, telephone number
2 Get buy in fromclean owners of the database
other
Apply Standard Suppressions
Know the data or the activity that took place at the
3 • Consumer - Movers, goneaways
time when the data was active
or Deceased
• Business - Movers, Trading
Understand the history. Communications, financial
4 Status
purchases, visits…… beware of inconsistent data!
Data Based Marketing
6. Data Enhancement
• Your data has been dormant for a long time, consider….
• Appending demographics (Household composition, wealth indicators,
SIC, employees, turnover……..)
• Appending behaviours (Media consumption, purchases, event visits,
website visits……)
• Appending share of wallet or other customer transaction behaviour
(Abacus)
• What do you already know about your lapsed customer?
• Tenure, Time since last purchase, previous baskets, mailing history……
• Segment customers
• Product , Service, Communication activity
Data Based Marketing
7. Apteco FastStats
Database marketing tool -
For data mining and analysis
developed by Apteco
Faststats delivers…
Data selections and Campaign
management
Business reporting
Statistical modelling and
clustering
Data Based Marketing
8. Long Lapsed Customer
Reactivation Campaign
Case Study – Barnardo’s
Data Based Marketing
9. The Cause
• 25% of children in the UK eat their
only hot meal at school
• 31% of children in Inner London
live in poverty
• 33% of British families are
surviving on just £10 each day
• Help support vulnerable children
across the UK, by making a
donation to Barnardos, visit
http://www.barnardos.org.uk
Data Based Marketing
10. Background
• Qbase working with Royal Mail, Qbase Direct, Call
Credit & An Abundance
• Identify dormant data Qbase can reactivate for a
cash donation mailing in February 2011.
Cash Fundraising Mailing
Consists of mailing scored long lapsed supporters
Cold data
Packs are split between a letter and a Box Pack
Prompt ask is £20, £50 and £100
Data Based Marketing
Real example
11. Process
Receive Barnardo’s Data Extract
insight
Build FastStats Marketing Database
profile
Campaign Objectives
model
Know Your Audience
output
Analysis, Profile, Model Data
DataWarm Marketing
Based
Real example Results
12. The Data
Insight-The Data
• Total supplied 2.7 million
supporters (individuals only)
Receive Charity Data Extract
• 17 million payments, made
Build FastStats Marketing database
by 1.4 million Supporters
• 10 million communication
Campaign Objectives appeals to 1.1 million
supporters
Know Your Audience
• Other Tables include:
Analysis, Profile, Model Data • Forms of Support
• Letter logs
• Membership
• Communication History
DataWarm Marketing
Based
Real example Results
13. Objectives
Insight-The Data
• Primary objective - Warm cash appeal
• Audience
Receive Barnardo’s Data Extract
• reactivate long lapsed cash supporters
Build FastStats Marketing database • Identify and score supporters who are
likely to give cash gift but have not done
Campaign Objectives
previously
• By...
Know Your Audience • Using FastStats Marketing Database
containing
Analysis, Profile, Model Data • Demographic data enhancements
(Lifestyle variables)
• Financial variables
• Audience net of Barnardo’s exclusions
are scored
DataWarm Results
Based Marketing
14. Audience
Insight-The Data
• Standard suppressions (GOA, Deceased,
Mailing flags)
Receive Barnardo’s Data Extract • Challenge events
• Supporters with active relationships are
Build FastStats Marketing database
excluded ( for example MEM, SAP, CHI,
Pledges enquirers etc.)
Campaign Objectives
• Any form of communication made with
Know Your Audience
the supporter last 18 months (Appeals,
Letter logs, Me contacts)
Analysis, Profile, Model Data • Any postal donation in the last 72 months
• Any Lottery generated income in the last
96 months
• Lapsed supporter types such as CG with a
form of help
DataWarm Results
Based Marketing
15. Audience Behaviour
Insight-The Data
• Total contactable audience is 905K, however...
• 62% (561K) have no payment date (archive legacy data)
Top 90% of Forms of support
Order FOH from Payment table Supporters % %cumulative
No of supporters with Last Pay Date 1 Barnardo Catalogue Purchase 98,019 18% 18%
50,000 2 GENERAL LOTTERY INCOME 74,651 13% 31%
45,000 3 GENERAL HOUSE TO HOUSE INCOME 71,268 13% 44%
40,000 4 Limericks Prospective 42,975 8% 51%
35,000 5 Postal Appeal Annual Subscribers 36,255 6% 58%
30,000 6 General Box Individuals 33,056 6% 64%
25,000
7 Retail Value of Donated Goods 29,414 5% 69%
20,000
8 Postal Appeal Donations 25,224 5% 73%
15,000
9 Barnardo Trading Donation 23,647 4% 78%
10,000
10 Gardeners Arcade Prospective 21,742 4% 81%
5,000
11 Limericks Purchaser 13,794 2% 84%
0
12 GENERAL DONATED INCOME 13,396 2% 86%
LE 1992 1994 1996 1998 2000 2002 2004 2006 2008
1990 13 GENERAL H2H 11,580 2% 88%
14 General Box Group 9,974 2% 90%
Data Based Marketing
16. Analysis, Profile, Model
Insight-The Data
• Over half of the supporter audience
have no known financial payment
Receive Charity Data Extract
• Those that do, the majority have made
a payment over 8 years ago. Look at
Build FastStats Marketing database
• Tenure/Loyalty, Value, Frequency, First, Las
t, Average Values
Campaign Objectives
• Look at other supplied attributes:
Know Your Audience • Channel of recruitment
• No of class codes
Analysis, Profile, Model Data • No of derived relationships (has letter
log, sent appeal, has contact)
• Apply demographics such as lifestyle
attributes
DataWarm Results
Based Marketing
17. Profile Variables
Insight-The Data
• Profile needs to consider supporters with and without financial values
Profile of Cash supporters (Final attributes)
Acquisition Segment
FinancialSegment Attitudes…
FinancialSegment Savings Level 1
LifestyleSegment Houshold Age…
LifestyleSegment Lifestage…
Cameo Financial Level 1
Sex
100
150
200
250
300
350
400
0
50
DataWarm Results
Based Marketing Variable predictive weight
18. Model Curve
Insight-The Data
Propensity model created
13 variables went
into the model
Model curve shows good fit at
identifying cash givers
Percent of Cash givers
Percent of supporter base
DataWarm Results
Based Marketing
19. Model Deciles
Insight-The Data
% of Cash giver comparison by banded score
% of CG within Decile (decile size 95K) % of total CG
18 70
16
60
Highest decile 14
contains over 50
12
60% of cash 40
10
8 30
6
20
4
10
2
0 0
Highest 2 3 4 5 6 7 8 9 Lowest
Decile score
% of Cash Givers % of CG within
DataWarm Results
Based Marketing
20. Mailing Cells
Insight-The Data
• Top decile selected (Decile 1: 75K)
• Decile 1 then split into deciles again
• Then split between Box Pack and Letter
Mailed
Score
Box Pack Letter TOTAL
Seg 1 & 2 Decile 1 (Highest) 8,817 8,808 17,561
Seg 3 & 4 Decile 1 8,739 8,799 17,499
Seg 5 & 6 Decile 1 3,477 2,824 6,295
Seg 7 & 8 Decile 1 964 961 1,925
Seg 9 & 10 Decile 1 (Lowest) 952 955 1,903
TOTAL 22,949 22,347 45,183
• Volume determined by charity based on cold/warm appeal mix
• 50/50 split between type
• Cell sizes split proportionality within deciles based on cash supporters
DataWarm Results
Based Marketing
21. Model Results
Overall Warm Model Response Analysis
Insight-The Data
• Model response curve shows that best scoring supporters are more likely to respond and is
flat from segment 3 with no real decline, however...
• Looking at the curve by segment code, the curve flattens at 4, then no real change at from 6 onwards
• Sample size for segment 6 to 10 are 1K, next time we should consider having an equal proportions across cells for
validation
DataWarm Results
Based Marketing
22. ModelbyResults
Response Split Type-Warm
Insight-The Data
Box Pack Torn Letter
Resp
4.00%
3.50%
3.00%
2.50%
2.00%
1.50%
1.00%
0.50%
0.00%
01 02 03 04 05 06 07 08 09 10
Segment Mailed
• Box pack Curve inline with identifying cash givers based on model
• Letter response significantly lower
DataWarm Results
Based Marketing
23. Model Results
Warm Model Gift size by Decile and Mail Type
Insight-The Data
(Gift value GT £50)
Column % of Responders Column % of Income Average Gift Value
Decile groups Warm BPR Warm TLR TOTAL Warm BPR Warm TLR TOTAL Warm BPR Warm TLR TOTAL
Seg 1 & 2 Decile 1 (Highest) 51.06% 45.45% 50.00% 52.06% 77.42% 61.13% £60.42 £240.00 £91.38
Seg 3 & 4 Decile 1 25.53% 36.36% 27.59% 24.60% 16.13% 21.57% £57.08 £62.50 £58.44
Seg 5 & 6 Decile 1 17.02% 9.09% 15.52% 16.16% 3.23% 11.53% £56.25 £50.00 £55.56
Seg 7 & 8 Decile 1 2.13% 0.00% 1.72% 3.59% 0.00% 2.31% £100.00 £100.00
Seg 9 & 10 Decile 1 (Lowest) 4.26% 9.09% 5.17% 3.59% 3.23% 3.46% £50.00 £50.00 £50.00
TOTAL 100.00% 100.00% 100.00% 100.00% 100.00% 100.00% £59.26 £140.91 £74.74
• Half of the supporters who gave a cash gift over £50 came from best scoring segments (1&2)
• Average gift value from the top segment is nearly 4 times the size for TLR compared to Box
Pack
DataWarm Results
Based Marketing
24. Model Results
Warm vs. Cold
Insight-The Data
Long Lapsed supporters
Average gift values higher for
Warm selection than cold
Warm is performing better then
banker lists
DataWarm Results
Based Marketing
25. Model Results
Warm vs. Cold
Insight-The Data
Long Lapsed supporters
Average Gift higher overall for
Letter vs. Box Pack
Again long lapsed customers
outperform cold data
DataWarm Results
Based Marketing
26. Creative Profiles
Creative Profiles -
Gender
Social Class Household Income
60.0% 18.0%
50.0% 16.0%
14.0%
40.0% 12.0%
30.0% 10.0%
8.0%
20.0% 6.0%
10.0% 4.0%
2.0%
0.0% 0.0% Box Pack
Torn Letter
Box Pack Torn Letter
Torn Letter has greater appeal to Torn Letter responders have a higher
higher Social Classes household income
Data Based Marketing
Cold List Profiles
27. To Recap, hints and tips
Strategic
recommendations
• The right data is key to Modelling
1. Know your data or find someone who does (knowledge is…)
2. Do you have confidence in the data quality?
3. Use a data mining tool such as FastStats to understand past
behaviour or identify inconsistent data
4. Give your data a revamp- append information such as
demographics. This can also fill in the gaps where you have bad
data!
5. Remember suppressions but don’t over suppress!
Data Based Marketing
Recommendations
28. To Recap, hints and tips
Strategic
recommendations
• Modelling works!
• Through 1 model we have identified long lapsed supporters worthy of
communication and have shown to be better responders then banker
lists!
• The letter appeals to more affluent households, therefore Qbase will be
creating niche models that identify supporters who are more likely to
respond to Torn letter vs. Box Pack
• Use modelling to identify and build customer or supporter relationships
• Remember Test, test and more test…
Data Based Marketing
Recommendations
29. Q&A
Strategic
recommendations
Paresh Patel
Business Insight Director
paresh@qbase.net
Data Based Marketing
Recommendations
Notas del editor
You need someone in your organisation who knows the data or the activity that took place at the time when the data was active