About
A comprehensive product recommendation solution that runs daily, encapsulating all customer intent for the product – coupling of latest intent signal with customer profile, digital behaviour and all transactional footprints over the last 6 months.
Innovation presentation
Traditional Setup:
Conventionally, triggers are developed basis rulesets eg. A person clicking on a personal loan(PL) promotional sms/email would be immediately approached for PL both via email/sms and the Relationship Managers.
However, this does not take into account the historical behaviour and the customer profile overall. Thus, the intensity of intent is not clear - result being all customers are treated alike and bombarded with the same communications, even if the customer had clicked by mistake.
Need:
Hence, there was a dire need to not only extract latest customer intent signals, but also provide a context to it, thus differentiating actual intent from noise.
Innovative Solution:
We designed an all-encompassing AI/ML framework that runs on a daily basis, segregating the immediate need of customers from a non-urgent one. The lift in performance (PL conversions in recommended customer base) observed has been 5 times the best of the erstwhile microsegment rules in place.
Eg. Customer transferred a Travel agency a sum of $5,000 yesterday. Conventionally, this would have been a travel related microsegment for PL take up. To this, we added the relevant recent digital footprints (eg. Browsed through our PL ‘Tools and Calculator’ page 10 days ago) and his recent financial standing (maintaining a low balance since past 3 months) to sharpen the intent identification.
The development of this solution entailed the following :-
1. Creation of a dedicated, centralized datamart for all warehouse data, deficient data, channel responses, digital signals residing in silos on multiple platforms
2. Leveraging the propensity model for product cross-sell that captures all relevant signals over the past 1 year
3. 30 mn+ transactions along with 10 mn+ digital information are mined on a daily basis to identify product specific signals
4. 25+ distinct patterns of behaviour unearthed, providing small and sharp slivers of customers with unprecedented conversion rates for loans.
5. Ensemble model of advanced ML algorithms like XgBoost and LightGBM rank-ordered the identified customers and partitioned further into
i.Top priority customers to be immediately engaged and
ii. Customers where nurturing could lead to conversions over a longer period
6. The identified customer behaviour along with the context relayed to all marketing and manned channels for having a uniform communication across all touchpoints of the customer
7. Automated airflow DAGs implemented to cull out the relevant base everyday
8. End to end automation of pipeline ensured omnichannel connect within 2 hrs of identification of relevant customers everyday
9. Feedback from channels, both digital and physical, looped back in to enhance the solution continuously
Uniqueness of the project
Real time integration and mining of all the data, that too, at this scale has not been endeavoured in the Indian Banking space previously.
Moreover, the project has been smoothly running without any manual intervention and performing and improving with time for over 6 months now.