Predictive Modelling

All sounds a bit boring really, but models are out there in our everyday life, busily working away behind the scenes. The daily weather forecast, even though it generally does rain every day in the UK, is one such model. All a model is, is using information that you do know (eg atmospheric pressure, temperatures across the past few days, etc) to predict something that you don’t know (i.e. yes, it will rain).

Apply that to a marketing context, and you can build models for who is likely to purchase, who is likely to convert, or who is likely to leave. Or, sometimes more pertinently, who is *not* likely to purchase, who is *not* likely to convert, or who is *not* likely to leave.

A model is simply a formula, that scores everyone on their likelihood to do something. Sort your customers on that score, and you have all the people or businesses who are most likely to do that thing at the start of the file, and those least likely at the end of the file.

We have built models to identify look-alike customers (people who have not yet purchased, but who share similar characteristics to existing customers), legacy givers (supporters who are likely to leave a legacy in their will), product cross-sell and switchers.

All the model does is help you to target your activity. You still need wonderful creative, and an appropriate campaign plan, to make the activity work. But the model will deliver significantly higher response from a more targeted audience, and our clients have seen some fabulous uplifts through using our models.

Much of our modelling is done in IBM SPSS, but we also use software such as R, PSPP, Si-CHAID, Excel and FastStats. In reality, the choice of software is only of limited importance – the real art of modelling comes in the thinking, the design, the engineering and the build itself.