How Do Utilities Detect Meter Tampering?
Meter tampering is confirmed in the field, by an inspection visit, which means detection is only as reliable as the visit itself. In Fluyenta's model, every inspection is captured on bodycam, AI vision checks the key frames for intact seals, serial numbers that match the paperwork, and signs of manipulation, and every visit receives a fraud risk score from 0 to 100 that routes suspicious work to a supervisor.
Why the inspection visit is the weak link
Whatever first raises suspicion about a meter, the confirmation happens at the meter, on a visit. The inspector's report becomes the official record: the seal was intact, the serial matched, nothing was manipulated. If the visit itself produces no verifiable evidence, that record rests entirely on the inspector's word, taken at a location where nobody else was watching.
That is also where detection fails. A tampered meter survives when the visit that should catch it is rushed, is closed from a parking lot, or is bought. The failure mode is not that the tampering was invisible. It is that the inspection was.
Step 1: Make the visit prove itself
The inspection should generate its own evidence. In Fluyenta's model, the customer's consent is recorded before capture begins, the visit is captured on bodycam with video, audio, GPS coordinates, and timestamps, and the work order closes with a digital signature. The evidence is sealed with a cryptographic fingerprint into an immutable manifest, so any later alteration is detectable, by anyone, including the utility itself.
Step 2: Put AI vision on the meter, every time
When the visit closes, key frames are extracted from the video for AI vision analysis: do the serial numbers match the paperwork, are the seals intact, are there signs of manipulation. The audio is transcribed and screened for bribery language, and audio quality signals such as excessive silence, muffling or obstruction, and sudden cuts are checked, because each can indicate tampering with the recording itself. Every visit gets this review, not an audited sample.
Step 3: Score the visit and route the exceptions
Every signal feeds a fraud risk score from 0 to 100. Low scores auto-approve and flow through. Visits above the review threshold route to a supervisor. Visits above the critical threshold are blocked and escalated as a case, with an urgent alert fired and the complete evidence chain attached: video, transcript, GPS, consent, and signature. Scoring weights, keywords, and thresholds are dashboard configuration the utility's own managers adjust without developers.
The rule that covers the inspector too
Meter tampering and inspection fraud are the same problem wearing two uniforms, and one mechanism covers both. If evidence were optional, the compromised visit would simply skip it. So a work order without its evidence does not close cleanly, and the missing evidence raises the visit's risk score on its own. The visit that tries to avoid scrutiny is the visit that attracts it.
You can't put a supervisor in every truck. You can put an AI witness on every visit.
This model runs in production with real field crews at a Colombian electric utility, live since July 2026.
Which inspections would you evidence first?
Deployments start with one inspection type in one zone, with value quantified before scaling.
Book a consultationStraight answers
Can AI really tell if a meter seal is broken?
AI vision analyzes key frames extracted from the visit video and checks whether seals are intact, whether serial numbers match the paperwork, and whether there are signs of manipulation. The AI does not make the final call: flagged visits route to a human supervisor who reviews the actual evidence, and every consequential decision is gated by a human.
What if the inspector is part of the fraud?
The same evidence that documents the meter documents the visit. The audio is transcribed and screened for bribery language, and the absence of evidence is itself a signal: a work order without its evidence does not close cleanly, and the missing evidence raises the visit's risk score. Access to footage is role-gated, and there is a complete audit log of every upload, analysis, decision, and role change.
What hardware do inspectors need?
Standard Android devices and bodycams. The client owns the hardware and Fluyenta licenses the software. Devices pair with a PIN and receive over-the-air updates in staged rollouts, and the app is offline-first: capture continues without signal and encrypted uploads queue and resume.