Introduction

A Temenos Analytics Next Best Product Predictive Model is a market basket analysis model implemented by using the Association Rules algorithm. It predicts the products a customer is likely to purchase in the future based on the purchase patterns of other customers in the past.

Such predictions are of great business value for creating targeted and efficient marketing campaigns as well as to entice customers to purchase at point-of-sale locations (e.g. bank branch or e-banking portal) by offering products that are very likely to be of interest to the particular customer.

Let’s illustrate this with an example. A report on the products popularity among the customers of a bank contains the data shown in below.

Product	Popularity
Product01	19%
Product02	13%
Product03	12%
Product04	12%
Product05	11%
Product06	10%
Product07	9%
Product08	4%
Product09	4%
Product10	3%

Based on the above data, if the bank is to reach out to 1000 random customers that do not have “Product04” it can be expected that about 120 of those customers (12%) will decide to buy it.

If on the other hand, the Next Best Product Predictive Model is used, after aggregating the model validation data we will get something similar to below.

Product	Predicted Count	Correct Count	Correct %	Popularity (from previous table)
Product06	80	44	54.5%	10%
Product01	13480	5840	43.3%	19%
Product07	560	160	28.6%	9%
Product02	3680	760	20.7%	13%
Product04	1000	660	66.2%	12%
Product03	1200	500	41.7%	12%
Total	20000	7964	39.8%

Now, if the bank is to target 1000 customers for whom the model predicted that are likely to buy “Product04”, with about 66% of those predictions being accurate this is going to yield sales to around 660 customers. Therefore, by putting in the same amount of resources we get 5.5 times the results compared to randomly targeting the customers. To turn this into money, assuming that the net profit from a customer is $100 per year, the model will contribute to generating $66000 profit per year, compared to only $12000 when the model is not being used.

Since no two financial institutions are the same, there is not a single predictive model that can fit all. The next table contains the default model names and descriptions of the models for predicting Next Best Product.

Default Model Name1 or other unique name specified at installation time	Description
NextBestProduct	Predicts at the lowest level of granularity, i.e. at the Product Description level. Suitable when there is low to moderate diversity of products, e.g. between 8 and 15. As the diversity goes up the accuracy of this model tends to go down. In such cases filters might be used to exclude very common or vary rare products or a model based on Product Classification might be chosen.
NextBestProdClass	Predicts at the Product Classification level of granularity. Suitable when there is a high diversity at the Product Description level, e.g. above 15. In this case the prediction accuracy tends to be much higher compared to the previous model. Yet the predictions are not as specific and should be considered a starting point. It is up to the organizations sales process to account for that and yield additional information from the customers (e.g. through qualifying questions) in order to offer specific products.
NextBestPGPLAN	Predicts at the Temenos Transact Product Group level of granularity. Developed specifically for the Temenos Transact PLAN feature. Functionally similar to “NextBestProdClass” model but uses the Temenos Transact AA Product Groups. Products from other Temenos Transact modules or other source systems will be shown using their Classification, e.g. “NonAA_Leasing”. The model name should not be changed at installation time.