As a general rule, the more data, the more powerful the model. With enough data, you can build "segmented model" systems, which use multiple models for different customer segments with different predictive patterns. Depending on available data, business goals and other considerations, here are some options:
-
Custom models—based on your company’s customer data.
When you have sufficient historic data, custom models are built specifically to predict how customers will perform for you. They can incorporate your knowledge of customer patterns or important variables. -
Pooled-data models—based on many companies' data.
Pooled data or "consortium" models draw on a wide range of customer experiences in your industry. As a result, they often identify patterns not yet present in your customer population. Credit bureau scores and the leading fraud scores are based on pooled models. -
Expert models—built judgmentally using analyst expertise.
Expert models can be built quickly and cost-effectively when you don’t have enough data. They are a good alternative to custom models when no pooled models are available.
