Business History Books » Business Consultant » Customer Lifetime Value Algorithms
Customer Lifetime Value Algorithms
Question:
> Hi all, > I’m interested if people actually calculate individual customer’s lifetime > value. How successful are you? Which algorithms and approaches do you use? > I’m currently working with the Pareto/NBD model (Schmittlein 1987 and 1994), > accidentially anyone here using that approach as well? > TIA, Daniel
Hi Daniel, I would be interested in responses to your question. For what its worth I have only ever felt the need to look at actual purchase histories, extrapolate into possible futures and look at types of customer. Seemed to give enough data to classify e.g. decide which accounts could be more attractive to target / milk. Simpler though perhaps in industrial markets than b2c though who knows what the future will bring 
. — Mark A. www.sticky-marketing.net
Response:
> Hi all, > I’m interested if people actually calculate individual customer’s lifetime > value. How successful are you? Which algorithms and approaches do you use? > I’m currently working with the Pareto/NBD model (Schmittlein 1987 and 1994), > accidentially anyone here using that approach as well?
Daniel, That’s a really interesting question and I do see how this can be charted, but in all but the most simple businesses models, e.g. gasoline sales, I don’t see how this could be used to make a prediction. There are so many factors I fail to see how somebody could apply a generic calculation. I think that this has to be an experiential thing. Furthermore, coming up with formulas to predict this kind of thing, even on a customer-by-customer basis, would be crazy. Your best bet may be to chart past performance and visually extrapolate future projections. (If you’re willing to accept a flat line, try regression analysis.) As a consultant, quite a few of my customer’s spending habits have been shaped more like a bell curve. Small projects at first, then bigger ones, and an eventual tapering off. Mike
Response:
> That’s a really interesting question and I do see how this can be charted, > but in all but the most simple businesses models, e.g. gasoline sales, I > don’t see how this could be used to make a prediction. > There are so many factors I fail to see how somebody could apply a generic > calculation. I think that this has to be an experiential thing. Furthermore, > coming up with formulas to predict this kind of thing, even on a > customer-by-customer basis, would be crazy.
I must agree that you will always have great degree of uncertainty in such models, especially if you try to get estimates on a one-to-one basis. But the necessity to calculate such a value (or ranking) is clearly there, e.g. for catalogue-optimization (who gets my expensive 4-colour-catalogue, who doesn’t) or for priority based call-center routing. > Your best bet may be to chart past performance and visually extrapolate > future projections. (If you’re willing to accept a flat line, try regression > analysis.)
What I’m trying to find out is, if there are approaches – e.g. stochastical ones – which can outperform regression analysis, cut-off heuristics and RFM-analysis in terms of prediction quality. The Pareto/NBD model I mentioned basically looks at recency and frequency, does some advanced stochastic calculus and shows up with a value. The advantage is that compared to a simple cut-off heuristic it can distinguish between different buying frequencies (high-frequency-buyer is more likely to have churned for a certain recency r than a low-frequency buyer). I guess I will simply compare this to the other approaches and try to figure out which fits my data better. > As a consultant, quite a few of my customer’s spending habits have been > shaped more like a bell curve. Small projects at first, then bigger ones, > and an eventual tapering off.
I read a dissertation regarding this topic, they come to the conclusion, that there is an optimal spending level of a customer. If the customer overspends, she’s likely to churn, if the customer underspends, she’s likely to churn as well… Daniel
Response:
Hi all, I’m interested if people actually calculate individual customer’s lifetime value. How successful are you? Which algorithms and approaches do you use? I’m currently working with the Pareto/NBD model (Schmittlein 1987 and 1994), accidentially anyone here using that approach as well? TIA, Daniel