United KingdomFinance and insurance2023 – ongoingOphelos
Machine learning decision engine and vulnerability detection for debt collection
Ophelos saw 25% more customers reach a debt resolution after machine learning chose each message and flagged vulnerable customers
Problem
Ophelos runs debt collection for its clients and wanted a more considerate approach for people under cost-of-living pressure. It needed to tailor contact to each person and spot vulnerable customers in their messages.
Approach
Ophelos built a decision engine that uses reinforcement learning to choose which message to send each customer and when. It also built OLIVE, a language model that scans inbound messages to flag vulnerable customers and likely causes. The models are trained, tracked and retrained daily on Databricks with MLflow and a feature store.
Results
Takeaway
Tailoring contact by model led more customers to resolve their debts, and most handled it without speaking to anyone. The page does not give a baseline period for the comparisons.
Categories
- Size
- Small business
- Industry
- Finance and insurance
- Business function
- Customer service
- Innovation type
- Process innovation
- AI technology
- Predictive analytics and ML
- Adoption
- Off-the-shelf tool or platform
- Evidence
- Vendor case study
Sources
Source summaries
Databricks describes Ophelos's reinforcement-learning decision engine and its OLIVE vulnerability model. It reports a 25% increase in customers reaching a resolution, a 67% increase in satisfaction, and 88.3% of customers self-serving.
Full details are in the original sources.
