The challenge
Meridian Bank’s fraud operations team was reviewing over 4,000 flagged transactions per day. Each review required an analyst to manually cross-reference the transaction against account history, merchant data, and behavioural patterns — a process taking 8–12 minutes per case.
The team was overwhelmed. False-positive rates exceeded 70%, meaning analysts were spending the majority of their time clearing legitimate transactions. High-confidence fraud cases were being delayed by the noise.
The solution
Meridian deployed a BeeBlast triage agent integrated with their existing fraud detection system. The agent receives each flagged transaction, retrieves relevant account history via the core banking API, queries a merchant risk database, and applies a structured reasoning process to produce a triage recommendation: escalate, auto-clear, or hold for analyst review.
The agent does not make final fraud decisions — that authority remains with human analysts. Its role is to prioritise the queue and surface the evidence an analyst needs before they even open the case.
The results
Within 60 days of deployment, average case review time dropped from 10 minutes to 2 minutes. The agent correctly auto-cleared 61% of flagged transactions that previously required analyst time, and escalated high-confidence fraud cases with 94% precision.
Total analyst hours spent on fraud review fell by 80%. The team now focuses its capacity on complex cases that genuinely require human judgment, and fraud detection rates have improved because analysts are no longer fatigued by a backlog of false positives.
