Customer Relationship Management: A Database Approach

1
Customer Relationship ManagementA Database
Approach
MARK 7397 Spring 2007
Class 7
James D. Hess C.T. Bauer Professor of Marketing
Science 375H Melcher Hall jhess_at_uh.edu 713
743-4175
2
Review of CLV approximation before calculating
customer assets
m r(1g)
CLV St1? m(1g)t rt/(1i)t
1i-r(1g)
where m is contribution margin g is growth rate
of margin r is retention rate i is interest rate
Note if the world stopped after T5 periods then
this might exaggerate the total by about 20.
3
CDNow Customer Acquisition
Customer acquisition costs 30-55 Gross margin
m 10-20 Growth in margins g 0 Retention rate
r .51-.68 Interest rate i .12
20 .68(10)
30.90
Best case CLV
1.12-.68(10)
10 .51(10)
8.36
Worst case CLV
1.12-.51(10)
It does not pay to acquire new customers!
4
Bertelsmann purchase of CDNow
Gross margin m 15 Growth in margins g
0 Retention rate r .70 Interest rate i .12
15 .70(10)
25
Typical CLV
1.12-.70(10)
Size of CDNow house file 3,290,000 customers
Value of house file 25?3,290,000 82 million
In July 2000 Bertelsmann bought CDNow for 117
million!
5
Capital One Customer Retention
Gross margin m 144 Growth in margins g
0.02 Retention rate r .85 or .90 Interest rate
i .12
144 .85(1.02)
192
Low retention CLV
1.12-.85(1.02)
144 .90(1.02)
255
High retention CLV
1.12-.90(1.02)
Capital One could afford to pay 63 to increase
retention rate by 5.
6
ATTs Acquisition of TCI and Media One
Customer acquisition costs per customer
4,200 Gross margin m ? Growth in margins g
0 Retention rate r .90 Interest rate i .12
m .90(10)
4,200
Breakeven CLV
1.12-.90(10)
m 1,027
Profit margin 0.45
Breakeven Sales 1,027/.45 2,280 or
190/month
To justify the acquisition, ATT needs ALL
customers to sign up for cable, high speed
internet, and cable telephone. This is
unrealistic.
7
Can we predict CLV using the past?
Prediction is very difficult, especially about
the future.
Neils Bohr
Malthouse and Blattberg took data and pretended
to move back to the future
They ran a regression to explain the future CLV
using historic variables on the calibration
sample. They then used the regression to predict
CLV for the validation sample.
Actual
False negative
Bottom 80
Top 20
85
55
Bottom 80
Predicted
15
45
Top 20
False positive