The starting point
FNOL was a 12-minute IVR-plus-callback journey. Reserve allocation lagged intake by hours. Fraud signals - inconsistent narratives, location anomalies, repeat claimants - were being captured forensically, not at intake.
What we deployed
A voice-first FNOL agent in Hindi, English, and three regional languages. The agent collects the structured FNOL data conversationally, extracts and reconciles the unstructured narrative, runs fraud signals at intake, and allocates an initial reserve based on the loss profile. Cases above a fraud-signal threshold route to investigators with the full conversation and signal pack pre-attached.
Outcomes
Median FNOL-to-reserve dropped to 90 seconds. Fraud signal capture at intake went up 3.4x. CSAT on the FNOL journey lifted 22 points. The investigation team's caseload became higher-signal, lower-volume - and case cycle time dropped.