How Does AI Fraud Scoring Work for Field Visits?
Every field visit receives a fraud risk score from 0 to 100. The score is built from the transcribed audio screened for bribery language, AI vision analysis of key frames from the video, audio quality signals that can indicate tampering, and whether the required evidence is present at all. Low scores auto-approve, elevated scores route to a supervisor, and high scores are blocked and escalated as a case.
What feeds the score
The transcript
The audio is transcribed with speech-to-text and screened for bribery language. Audio quality signals are checked too: excessive silence, muffling or obstruction, and sudden cuts, because each can indicate tampering.
The video
Key frames are extracted from the video for AI vision analysis: do serial numbers match the paperwork, are seals intact, are there signs of manipulation.
The evidence itself
Consent is recorded before capture begins, and GPS, timestamps, and the digital signature travel with the visit as one evidence chain. A work order without evidence does not close cleanly, and missing evidence raises the risk score.
What each score band triggers
The score decides the route. Below the review threshold, the visit auto-approves and flows through. Between the review threshold and the critical threshold, it routes to a supervisor for review. Above the critical threshold, the work is blocked and escalated as a case, with an urgent alert fired and the complete evidence chain attached: video, transcript, GPS, consent, and signature. The thresholds themselves are not fixed facts of the product. Scoring weights, keywords, and escalation thresholds are dashboard configuration, adjustable by the client's own managers without developers.
In the end-to-end test on the production configuration, transcription flagged the bribery keywords at 92% confidence.
See a visit scored end to end.
The fraud demo runs the same scoring pipeline described on this page.
Book a consultationStraight answers
Who can change the scoring weights and thresholds?
The client's own managers. Scoring weights, keywords, and escalation thresholds are dashboard configuration in decision tables, not code, so they are adjusted without developers.
Is the score the final decision?
No. Every consequential decision is gated by a human, and every AI action is logged. The score decides routing: what flows through, what a supervisor reviews, and what is blocked and escalated as a case.
Can the score or the evidence behind it be tampered with?
Evidence is sealed with a cryptographic fingerprint into an immutable manifest, so later alteration is detectable. Access to footage is role-gated and every access is logged, and there is a complete audit log of every upload, analysis, decision, and role change.
Does capture work without connectivity?
Yes. The mobile app is offline-first, capture continues without signal, and encrypted uploads queue and resume automatically when connectivity returns.