Customer Retention and Modeling in the Utility / Telecom / Insurance Business
First published 9/15/01
Some businesses look on the surface like they many not be suitable
for using behavioral profiling. Take the electric power
business. The customer bill is just about the same every month,
with seasonal variations, and it's not like the customer has a choice
as to whether to buy power each month. Yet they can defect to an
alternative provider, and before you know it - it's too late. The cellular
biz has similar attributes,
particularly with the growth of annual contracts and bulk rate
minutes. How do you know when a telecommunications customer is about to defect?
Or take the insurance business. A very long cycle business
with very little transactional activity. You sign up for
insurance, get billed once or twice a year, and that's it. How
do you profile behavior in a business like this, where the customer
may be around for 5 years and then all of a sudden, just defects?
The answer is you profile activity other than the revenue related
activity, for example, calls to a phone center or visits to a web
site. When you look at "best" customers vs.
"worst," are there calling or visiting patterns which stand
out? In some service-oriented cases like this, you have to
"flip over" the behavioral approach - if the customer is Recent
and Frequent on phone calls, this may be a "bad" thing
and a pattern people engage in just before they defect.
The key is to group customers by "best" and
"worst" and look for any pattern that separates the two,
then use behavioral profiling methods to detect the likelihood of
defection. You can also do some limited testing on your customer
base to find out if behavioral profiling is useful on your particular
customers. If you haven't already, see the following article,
and substitute "calls to call center" or "visits to web
site" for purchases:
Techniques - Determining
Potential for Marketing ROI Before You Buy
If you have a website or telephone "self service"
interface, falling use of it might mean customers are getting ready to
defect, or it might mean they are satisfied and are going to stay long
For example, if a pattern of increasing Recency
and Frequency of visits to the "help" section of the web
site tends to predict customer defection, you could trip an
intervention call from the call center to "just check in."
In a long cycle, low transaction frequency business like insurance,
you may have to extend your time horizon to pick up enough meaningful
transactions. Instead of looking at behavior on an annual basis
as suggested in the above article, you might have to look over a 3
year or 5 year period.
There's no way to tell in advance what these metrics will be, but
the customer behavior will "speak" and tell you which data
points matter. Here's what you should do:
1. Make sure you understand all the internal data points
available to you - what exactly they are, where they come from, how
they are derived. Billing records, service records, installation
records, and so on. Customer source is always very important, if
you can get at that through the marketing area.
2. Isolate "best customers" - those who signed up and
stayed signed up for the longest time, with the least cost (variable
cost to you - installation, marketing etc, not in terms of total calls
to the center).
3. Run profiles over time (LifeCycle)
each piece of "action-oriented" data available to you, and
determine which provides the highest correlation to best customer
For example, high Recency and Frequency of calls to the center might
be highly positively linked (good service leads to better customer) or
might be highly negatively linked (billing problems create repeated
calls = mad customers who disconnect).
In service businesses, you generally look for
sharp changes in behavior - a drop of 30% in
usage, and increase of 50% in calls. These are
good targets for automation since they're quite clear cut.
And finally, source of customer is absolutely critical in this kind
of business, especially since your "markets" may be
geographically constrained. Good customer retention starts
with proper customer acquisition, and it should be relatively easy to
look at LifeTime Value by customer source.
Here's what I mean.
Pick a start date, say one year ago (3 or 5 years for long cycle
businesses), and take a quick look at your highest value customers
(gross billings?) over this time and see where (what campaign,
geography, etc.) they came from. Then look at lowest value (high
churn, disconnected) customers from the same start point, and see
where they came from. If there are differences, you're on
your way to finding the answer you're looking for. In addition, once
you determine there is a difference, survey a subset of each
group and try to find the commonality in the groups and differences
between the groups. This links the data to the emotions and
provides a backdrop for improving acquisition technique.
Don't try to do this starting from a "micro"
level and looking up. Start with macro ideas
(geography?) then "drill down" (couldn't resist)
a layer, then another. When you get down to
the level where their appear to be no sizable
differences between best and worst groups anymore, you're done.
Going any lower is just "noise" and is not generally
Something else that works in cycle billing environments is Latency
of payment. If a customer averages payment 20 days after
billing, and that slips to 30 or 40 days, it can be a good indicator,
especially when combined with the "seeking help" activity
For more on Latency, see:
with the Customer LifeCycle:
Trip Wire Marketing
You really have to just sit down with a customer record, look at
all the data in it, and say to yourself, "If there was a change
in this data point, could it mean something?." Then compare
best (long life) and worst (fast churn) customers and see if there is
Perhaps the Latency indicator you are looking for is outside of
billing. Look at Latency of initial service - the number of days
asking for service and getting it. Or the number of days between
getting service and the first "trouble call" or billing
dispute. There is something in there - always is. The
"it" may differ by geography, which can lead to the
discovery of other more operationally oriented problems causing
You may not be tracking any of these kind of metrics now, and you
may have to "make up" some using the raw data. My
suggestion would be to just grab a couple of these ideas, and see if
you can make a dent. Find a couple of best and a couple of worst
customers, and really take a calculator to some of these ideas. If you
get a Eureka! moment (it always happens that way), then ask IT to run
a broader cross section of customers on the same parameters to prove
it out. There is always something, a tip-off by the customer, as
to what they're thinking.
Here's a personal example. I switched long distance providers
and accepted an offer of a personal 800 number. I never used it
or gave it to anybody. When I got my first bill, I was billed
for a bunch of 800 number traffic. Since they supplied the
"origin" numbers, I called them. Every one of them was
some kind of internal engineering number, the kind that spit numbers
back at you, like "System 4820, online" and the like.
When it happened again the second month, I just switched everything
to the local phone company. If the offending company looked at
my billing record, saw I was a new residential 800 customer, profiled
all new 800 customers who defected, and compared results with those
who didn't, they might see a series of very short 800 number calls
from internal engineering numbers on a decent amount of the defected
customer bills - and none on those that did not defect. Or maybe
they could have just looked at average call length on new 800 numbers,
which was very short on my bill compared to what you would expect.
They could have seen a pattern - defection potential detected, and
taken a corrective action before the defection.
A note on win-back - it's very tough in a commodity service-oriented
business. The customer is facing substantial potential switching
costs - tangible and intangible. Once they go, they're gone, and
calling them to say "did you know we offered a cheaper rate"
which is to say "did you know we have been screwing you all this
time" is not helpful. The flip side of this situation is
the ROI can be very, very high when you can predict defection and save
the customer - and very trackable too.
What's the point of all this? Just because you don't have a
lot of customer-controlled purchase activity going on doesn't mean you
can't use behavioral profiling to predict customer defection.
Once you find your predictive data points using the process described
above (or the much more expensive data mining route), then you can use
the concepts on this site and in the book to organize and track your
customer retention program. Instead of purchases, you track
phone calls, web visits, or some other indicator using the Drilling
Down method to create your early
warning system defection indicators.
Many companies offering long purchase cycle products actively
shorten the cycle by employing an inter-purchase contact
strategy. By actively contacting the customer between purchases,
these companies try to "bridge" the purchase cycle and
maintain Recency of contact. This approach can lead to an
increase in repeat purchase rate, if handled correctly.
In fact, this approach is not new and has nothing to do with the
Internet. State Farm Insurance has for a long time pursued this
contact strategy using mail. Weber-Stephen Products Co., the manufacturer of Weber
Barbecue Grills, sends a quarterly magazine full of seasonal cooking
tips and accessories to customers who buy high-end grills.
If the above makes sense to you, then you are on your way to
designing the highest ROI customer marketing campaigns of your
career. The Drilling
Down book teaches you all of the proven LifeCycle-based marketing
techniques step-by-step, building up from simple ideas like
Latency to full-blown visual customer LifeCycle mapping techniques.
you want to start returning profits of 2 - 5 times the money you spend
on a customer marketing campaign, you need this book!