Event
Summary
CRM these days is about what a customer is likely to do. Most solutions
focus on sales opportunities - what up-sells or cross-sells might be good
candidates for a particular customer. SLP InfoWare does address those
problems, but its early roots at solving churn problems in the wireless
industry, particularly in Europe but also with Cell One of Puerto Rico,
give it a handle on predicting which customers are likely to leave. This
is a well-known problem in the cellular phone industry, where consumers
are always being lured by competing offers. It is also a problem for ISPs,
especially as free ISP's come on the scene.
SLP InfoWare's ISP/CPS product uses statistical and machine-learning techniques
to analyze past customer behaviors and predict which current customers
behave like customers that, in the past, have decided to accept another
offer. Such customers can then be targeted for special promotional offers.
Market
Impact
Predicting churn is somewhat different from finding cross-sell opportunities
- although only the statistical wizards know how different - because it
requires discovering ill-defined patterns in long-term behavior. According
to SLP InfoWare, an ISP's acquisition costs for a new customer can be
as high as $400 each.
A recent report from the Strategis Group, says that almost one in four
users of ISPs in France and the UK are likely to move between providers
over the next year. Although the rate in the United States has been traditionally
much lower than in Europe, with millions of dollars riding on small reductions
in churn a successful product should be snapped up by profit focused ISPs.
User
Recommendations
An ISP should certainly give SLP InfoWare a call. However, we caution
that almost any intervention is likely to cause some improvement in whatever
statistic you measure. We therefore believe that for any product that
proposes to improve performance, testing the efficacy of the model is
necessary.
In
essence, to test a product objectively you could randomly split your user
population into two groups; new users should be assigned randomly to one
of the groups. In the case of a churn-prediction product, record the product's
predictions for members of both groups, but only attempt to intervene
with one of the groups. At the end of a four to six month period you will
be able to evaluate the success of the product (and of your interventions)
and decide whether to extend it across your customer base.
SLP
Infoware offers a version of this test as a part of its sales effort,
through a program they call Proof of Value (POV). According to Jerome
Nagel, VP for Marketing Worldwide at SLP Infoware, "The POV goes through
the cycle of preparing data, creating an initial set of predictive models,
designing a set of targeted campaigns and executing these campaigns on
a subset of the identified campaign candidates. From this we assess the
'lift' of our modeling and/or the performance of the overall system. This
initiates the detailed discussion on pricing."
We think that almost all vendors will have to offer a similar pre-sale
tryout. Our only caution is that you should continue testing after the
sale, perhaps with a test of longer duration than is reasonable in a pre-sales
environment, and periodically thereafter.
We believe that with this or any tool that promises to reduce losses or
improve sales there should be an option to make your payment at least
partially dependent on success. The model that would be fair for both
sides would be one in which your losses are minimized if the product doesn't
perform well, and the vendor stands to share if the product is very successful.
SLP Infoware does offer such a pricing model, as an alternative to a straight
license, although given that the customer has a chance to test the product
before the sale there will be less concern about risk.
Mr. Nagel says that after the POV essentially all customers prefer the
license price over a plan where both partners share the risk, "quickly
acknowledging that the business case makes the license fee the way to
go in terms of TCO (total cost of ownership)."