Improving the Banking Onboarding Assistant

In a previous article,, we have demonstrated how to use a banking virtual assistant to assist client on boarding. Today we want to go a step further and implement product recommendation as well. Product recommendation are widely used and use called Recommender Systems algorithms. The assistant will personalize product recommendation according to the customer profile.

Recommender Systems

There are different ways of implementing recommendations like those we can see on Amazon or Netflix for example. Such algorithms are usually called recommend systems, we can find them in different flavors:

  1. User based ( collaborative filtering): the recommender system will assume that similar users have similar tastes. The system will try to find products that similar users have selected
  2. Item based ( collaborative filtering): the recommender system will try to find similar products that the client or similar clients have already selected
  3. Classification: the recommender system is based on usually a multi-class classifier. It will predict what kind of product the user will be selecting based on user characteristics

Multi-class classifier

In our case, we will use a multi-class classifier that depending on the answer provided by the user. The classifier will select the product with the highest probability. Using a classifier allows us to avoid having to store past customer behaviour to train the model. In a banking context that would make little sense anyway. The system will try to find out of around 15 products which one has the highest matching probability. The model has been trained with a few hundred examples that have been created manually. We can train and refine the model continuously to adapt to get more specific as time goes on.

The dialogue ends with a product recommendation directly to the user profile, see below

You can also see the full dialogue in the following video.

(Visited 30 times, 1 visits today)
%d bloggers like this: