Monday, October 09, 2006

Building a Customer Experience Simulation

What’s the simplest possible Customer Experience Matrix?

You probably haven’t spent a lot of time wondering that, and I don’t generally lose sleep over it myself. But as we at Client X Client start building more sophisticated simulation models, defining the simplest possible case gives us an important reference point.

Of course, the question has an obvious answer: the simplest version of any matrix is a single cell. In a Customer Experience Matrix, this would describe one activity in one channel. So the real question is whether a one cell model is too simple to be useful.

Let’s assume the activity our cell monitors is purchasing. So our single cell matrix would predict purchases for a group of customers over time. This is certainly a useful thing to know.

Because we’re simulating results over time, our single cell model generates one data point for each time period. Each represents whether or not the customer made a purchase. To do this, our model needs the following functions: create a number of customers; calculate a purchase probability for each customer for each period; apply the probability to determine whether or not a (simulated) purchase was made; and store the result. The output can be imagined as a stream of ones and zeros.

Having a stream of numbers lets us use those numbers to calculate cumulative statistics. The obvious ones are total purchases to date and time since last purchase. Anyone trained in direct marketing will immediately recognize those as Recency and Frequency: two thirds of the classic RFM measure. If we had recorded the amount of each purchase, we could calculate M (Monetary Value) too. We can also use the same data to assign customers to the four basic behavioral segments: prospects (never purchased), new customers (first purchase last period), active customers (multiple purchases including one last period), and lapsed customers (one or more purchases, but none last period).

So it turns out that even a single cell simulation model can estimate customer lifetime purchases, RFM codes and behavior segments. This is enough to provide a useful starting point for a system design.

And there’s more. I’ll continue this thread tomorrow.

No comments: