Challenge: Security versus customer satisfaction
Our FinTech client faced a common industry dilemma—fighting fraud without turning away legitimate customers. Their internal machine learning models were too aggressive, mistakenly flagging valid transactions as fraudulent while letting actual fraud slip through.
Diagnostic: Narrow data scope
The root of the problem was the limited scope of the company's fraud detection models, which were trained solely on internal data. This limited focus failed to capture the diverse patterns necessary for accurate fraud identification.
Transformation: Three-step fraud detection upgrade
Here's what we did over the next three months:
Data enrichment: We worked with the client to pick out the best third-party data sources and oversaw the integration. This step enriched their system with detailed, real-time transaction data, improving fraud detection accuracy.
Model optimisation: Under our guidance, the client's internal models were refined using the enriched data. This allowed for more accurate distinctions between potential fraud and legitimate transactions, reducing false positives and accelerating transaction verification.
Selective scoring: Due to the high cost, we advised against third-party scoring for all transactions. Instead, we established a protocol where only transactions flagged as ambiguous undergo third-party evaluation, optimising costs while maintaining security.
Results
Thanks to these strategic changes, the client saw:
50% reduction in false positives
10% faster transaction speed
Over 20% reduction in chargeback rates.
These improvements reduced fraud losses and minimised disruptions in customer transactions, thereby improving overall customer satisfaction.