Jim's Intro: A more complex version of the RFM
model: customer behavior over time.
Ah, time. Time is the most overlooked and underutilized
tool in the Data-Driven marketing toolbox but one of the most powerful.
You can see the strength of timeís influence in the importance of Recency, the most powerful of the modeling variables.
changes over time and these changes are clues to the future. These
changes over time are known as the customer LifeCycle; and the LifeCycle
is used to calculate LifeTime Value. Without
knowing the LifeCycle, it's impossible to find LifeTime Value.
Customer profiles tend to be a "snapshot" of
the customer, their characteristics or behavior at a single place in time. But are you the
same person now as you were last year? Have you not had experiences and
thoughts and triumphs and downfalls in the past year that changed the way
you think and behave?
Even though activity-oriented human behavior can be
modeled at a point in time, it is far more powerful to look at it over a
time period. This issue is lost on many who do "profiling" in
todayís environment, and even some of the pros from the catalog and TV
shopping worlds where these techniques were developed donít grasp the
power of profiling customers over time.
I think youíd agree knowing not only the current state of the
customer, but also the path to this state, can provide critical additional
clues to a customerís potential and future behavior. Knowing the
powerful input into the marketing approach to be used to address the
current state of the customer.
How is this accomplished? You need to store customer scores like Recent
Repeaters or RFM
scores for the same variable (purchases, visits, etc.)
in your customer database, and extract trends from these scores over time.
Itís like a blend of Hurdle Rate
analysis and customer scoring techniques. Instead of having one Hurdle
Rate to watch over time, you look at the multiple customer scores
themselves over time. For an example of applying this technique to
choosing the most profitable ads to run, see the tutorial: Comparing the Potential Value
of Customer Groups.
Itís tremendously powerful stuff. Imagine this.
month (or week) you do a profile of your customers. But instead of
replacing the old score, you keep it, and add the new one. Then look at
the spreadsheet or query the database and ask, "Find everybody who
has a lower score this month than last month, and find everybody who has a
higher score this month than last month." Think of what you could do
with that information.
Anybody who had a high loyalty score and doesn't anymore
deserves your promotional attention, before they drop even lower, with
visits or purchases. Anyone with a rising score should be encouraged with
Generally, anyone who is moving lower in score is in the
process of defecting ó of leaving you as a customer. The more
dramatic the move, the more likely they are to be defecting. The mirror
image is true for those climbing in score. The more dramatic the move, the
more likely it is they are on their way to becoming a more valuable
customer and adding value to your business. Here is the dirty little
secret of interactive behavior:
The customer lifecycle for interactive businesses is
more exaggerated than for traditional businesses. The behavior
ramps up faster at the beginning of the LifeCycle, but then falls off faster
into the end of the LifeCycle.
The speed or rate of behavior change is incredibly
important to modeling interactive behavior, much more important than in
offline models. Small changes over time are to be expected; rapid and
accelerating changes are much more significant and a signal for
So once you use scores to identify customer
behavior, think about re-profiling your customers at certain intervals and
keep track of scores over time. You'll be glad you did.
Step by step instructions for creating likelihood to respond
and future value scores for each customer, and using scores to create high ROI
promotions based on the customer LifeCycle are found in the Drilling Down