Passive feedback loop for refining daily identification of fund-needy customers Qorus-Infosys Finacle Banking Innovation Awards 2024

Our project optimizes cross-selling strategies within our banking ecosystem by refining the daily trigger model for Personal Loan (PL) product recommendations. The existing model's effectiveness was constrained by its reliance solely on recent customer signals. To sharpen this further, we leveraged comprehensive datasets including i. Historical PL recommendations, ii. PL purchase history, and iii. Aggregated transaction data spanning six months This approach improved the model's ability to discriminate and rank-order potential leads effectively.

Further we implemented – 1. Active Learning - Facilitating efficient model learning from pertinent data for refining trigger models 2. Reinforcement learning mechanisms - The model iteratively improved its identification capabilities by reinforcing actions that led to accurate triggers and penalizing those that did not.

By synthesizing these techniques, we achieved the following – i. Mitigated both false positives and false negatives ii. Optimized conversion rates through a deeper understanding of customer behaviour dynamics

This initiative underscores our proactive approach to driving tangible business outcomes through cutting-edge methodologies such as Active Learning, Reinforcement Learning and Surrogate LSTM modelling. Driven by our commitment to be more customer-centric, our innovative experiment successfully uncovered genuine hidden intent signals within historical data, beyond immediate signals. Our efforts significantly enhanced business performance, resulting in incremental annual PL disbursals of over $28 mn and a 15% rise in productivity across customer-facing channels.