Next best activity, offer of product

Let’s take a practical example: Let’s assume that you are selling 3 different products: A, B and C. You want to create an automated system that selects “the right product for the right prospect”. In other words, you want to know which of the 3 products has the highest chance of purchase for a given customer.

This system (that selects “the right product for the right prospect”) is quite simple, conceptually:

  • Create (with TIMi) 3 “propensity to buy” predictive model (one for each product: A, B and C).
  • Apply your 3 predictive models on your customer database:

For each prospect inside your customer database, you obtain:

  • From the model 1: the probability of purchase of product A.
  • From the model 2: the probability of purchase of product B.
  • From the model 3: the probability of purchase of product C.

For each customer, you now have 3 probabilities (one for each product). You select the product with the “highest probability of purchase” (amongst the 3 probabilities that you just computed for this customer) and you simply propose it to your customer.

This is a true “One-to-One” recommendation system because each customer has different “probability-of-purchases” and can potentially receive a different product.

The above system can be refined (instead of using 3 predictive models that are of the “one-against-all” type, we could have used 6 predictive models that are of the “one-against-one” type) but the principle stays the same.

To create the “individuals models”, you should, of course, use a database containing as much information as possible about your customers. You can collect all the off-line and on-line activities of your customers and save them into your database. The more data about your customers (the more columns describing your customers), the better. TIMi is the only datamining tool that is not limited in the numbers of columns that it can process. Anatella is the only ETL tool on the market that is able to process data tables containing several thousand columns.

The accuracy of such system (“Next Best Activity/Offer/Product” system) is only as good as the accuracy of the “individual predictive models” that are composing the system. Once again, TIMi is the preferred solution when building such systems because:

  1. You need very high accuracy for the individual predictive models
  2. You need a very fast analytic tool to create the many different predictive models. TIMi is, by a very large margin, the fastest analytical tool on the market.

TIMi includes a special tool (the “Model Merger” tool) that can apply thousands of different predictive models simultaneously using only one pass over your customer database. The computation time of TIMi is nearly independent of the number of applied models (and also independent of the complexity of the individual predictive models). Typically, using the “Model Merger” tool, you can process your customer-database containing 10 million customers (10.000.000) in less than 2 minutes (independently of the number and complexity of the predictive models). When using “TIMi models exported as optimized Teradata SQL code”, the scoring speed is even higher. TIMi speed is unmatched. The time required to score the same database using the fastest solution available, (other than TIMi or TIMi +Teradata) is, in average, above 3 hours (it depends on the complexity of the models and can grow to 20 hours-uninterrupted-computing-time if you used complex models). Really, compared to Timi, other analytical tools are Rube-Goldberg machines!