Next Best Action: Dynamic AI Engine to Deliver Personalized User Experience At Scale Qorus Banking Innovation Awards 2021

Marketing primarily has been product driven within banking industry, given the nature and complexity of products involved. For a bank catering to 50 million+ customers across 50 products and services, there were 80 million product communications being sent out monthly. Also, if a customer has a high propensity for multiple products, it was difficult for the frontline staff to decide which one to talk about first. Therefore, the objective was to have more of a customer focused framework, so that the Relationship Manager would limit the discussion to the most suitable 2/3 products and have more meaningful conversations. And so, we orchestrated the distilled intent from all customer touchpoints and the product propensities to arrive at the next best products at the customer level. To develop a recommendations system, we adopted an innovative framework built on individual product propensities, customer intent captured through campaign responses, digital footprints and added frontline verbatim on top of it. We followed a 4-step approach to determine the most relevant products at a customer level - 1. Product propensity models: Advanced ML algorithms like XgBoost, LightGBM, Random Forest etc. were utilized to develop best-in-class propensity models. 2. Integration of intent captured through feedbacks: Feedback from various channels, captured on a daily / weekly basis, was leveraged to identify the intent shown by the customer. a. Service Interactions wherein the sales team record their interactions with customer as free flowing text. NLP was done to identify sentiment of customer with regards to the product. b. Customer’s footprint on the bank’s website. c. Banners / ads showcased to customer while (s)he was browsing through the app or website. d. Channel response data that includes Email / SMS sent out to customer. 3. Product Prioritization at a customer level: The intent was blended with the propensities and a rank-ordering algorithm was used to arbitrate between multiple products at the customer level. Finally, up to 3 products were assigned to the relevant customers. 4. Hyper-personalised Narratives basis the customer’s behaviour and digital footprints: A BCI (Behaviour, Causation and Intervention) framework was devised to provide context to the product being recommended. By answering basic key questions like i. What is the customer doing? (Behaviour) ii. Why is the customer looking for the product? (Causation) iii. How to pitch? (Intervention) Relationship Managers are armed with rich insights before they even meet the customer. The solution has not only improved employee productivity, but enhanced customer engagements as well. As per feedback from ground, customer interactions are now more contextual, and the narrative backed recommendations serve as good conversation starters.