A government body paying out grants and subsidies was losing public money to false and duplicate claims — while honest citizens and businesses waited months for money they were owed. We caught the fraud and freed the queue.
A public body handing out grants, subsidies and support was losing a large amount of public money to false claims, duplicate requests and bad actors exploiting bureaucratic gaps.
And the manual defence punished everyone else. Reviewers went through thousands of forms, payslips and scanned documents by hand every day; the Sisyphean workload created backlogs of months, holding up money for citizens and businesses who were genuinely entitled to it. The old rules-based system threw a false alarm at every small anomaly, blocking legitimate requests and piling impossible load onto the service and appeals desks. Fraud got through; the honest waited.
Cross-reference in real time, auto-pay the clean, and hand a human the doubtful ones with the evidence.
We built a secured data infrastructure that cross-references an applicant's details in real time against government databases, reporting history and existing records — the full picture, at the moment of the request.
An engine scans every incoming document and form in seconds, spotting complex fraud patterns — documents edited in an image editor, or a ring of applicants filing from the same IP address — and screens them out.
Valid requests above a 95% certainty bar are approved and passed to payment automatically, with no human touch — the entitled citizen stops waiting behind everyone else's suspicion.
Doubtful cases go to a human investigator with a "red summary" that points precisely at the contradiction in the data — a lead to work, not a blank form to re-read.
Scoring on genuine signals rather than blanket rules dried up the flood of false positives that used to block legitimate applicants and swamp the appeals desk.
Because each flag carries its evidence and each auto-approval its score, decisions on public money can be inspected and defended — essential in the public sector.
Catching more fraud and paying honest citizens faster stopped being a trade-off.
Losses paid out on fraudulent claims fell by around 50% in the first quarter, and the time to approve support for a legitimate citizen dropped from several weeks to a few minutes — a real lift in public trust.
Four in five false-alarm alerts were eliminated, letting a small control team focus entirely on investigating genuine attempts to defraud the public purse instead of clearing innocent applicants.
It holds because it only auto-pays where it's near-certain and cheap to be wrong, and every escalation carries the evidence a human needs to decide and to justify — which is exactly what accountability for public money demands. Automation on the clear cases, a person and a paper trail on the rest — the same discipline behind our fraud-detection work for a private insurer.
Reef TRH