Closed-loop reconciliation
The receipt is not the source of truth. The authorization is.
Every claim ties back to an authorization registered in our system before the dispensation. Forgery stops being a detection problem and becomes a structural impossibility.
Receipt-image forensics is a losing race. Forgers get better faster than detectors do. The durable fix is to make forgery impossible at the schema level: if the authorization is the source of truth, no fake receipt can exist without a matching authorization that was never issued. Ecuador proved the model at national scale with its government prescription registry. Inspector AI runs it for private insurers across Latin America.
Why receipt-image forensics cannot win
Forgers iterate in hours. Detectors iterate in quarters. Every new model that can identify a fake receipt is also training data for the next generation of fakes — adversarial learning is asymmetric, and it is asymmetric in favor of the forger. A fraud detection platform that treats forged receipts as an image-classification problem has already accepted that the best outcome is statistical: we catch most of them, we pay for the rest. That is not a durable position.
How closed-loop reconciliation works
Authorizations are registered in Inspector AI before dispensation. Every incoming claim reconciles against those authorizations. A claim with no matching authorization cannot be paid. A claim that drifts from the authorization on drug, quantity, or date cannot be paid silently — it is flagged and routed to review. The audit trail is the authorization record, not the receipt image. The receipt becomes a form, not a source of truth.
The Ecuadorian precedent
Ecuador runs a national prescription registry. Physicians register prescriptions centrally. Pharmacies validate against the registry before dispensing. The model works — it is a public health system operating at national scale, and it proves that closed-loop pharmacy reconciliation is not experimental. Inspector AI brings the same architecture to private insurers in Latin American markets that do not have a government registry. The technology is not novel. Making it work inside private claims workflows is.
What the audit trail looks like to a regulator
ANS RN 659 in Brazil asks health operators to prove they have adequate controls on pharmacy claims. Traditional proof is a report: here are the claims we audited after the fact, here is what we recovered. Closed-loop proof is structural: here is every authorization that existed before a claim was paid, here is the reconciliation record, here is the reject queue for claims that did not match. The first is a narrative. The second is evidence. Regulators prefer evidence. See the WAFL framework for how this audit trail maps to each loss category.
of Inspector AI claims reconcile back to a pre-registered authorization
See closed-loop in practice
A three-week proof of concept on your real pharmacy data. We register authorizations, you send claims, we show you the reconciliation trail.