A payments platform lives or dies on onboarding. Theirs had turned into a queue: every new business waited days while a person read its documents by hand and checked its name against international watchlists. Legitimate merchants gave up and left, and the forgeries the team was straining to catch were the ones slipping through.
When demand for account verification spiked, the manual process buckled. Businesses that wanted to start clearing payments waited days for someone to approve their registration documents and identity papers, and a painful share of them abandoned the sign-up before it finished.
Behind the queue, the compliance team was spending its hours on the wrong work. Skilled analysts burned most of the day on routine document checks and name-matching against sanctions lists, the kind of clerical comparison a person should never be the fastest way to do, instead of investigating the genuinely suspicious cases that needed a human mind.
And the existing tools were losing the arms race. High-quality forgeries, identity documents edited in Photoshop, faked watermarks and security marks, were good enough to pass the old checks, which meant the firm carried both a growth problem and a real exposure to regulatory fines from a single missed match.
Automate the routine screening at machine speed, and put analysts only on the cases that carry real risk.
We wired a deep-learning engine into the sign-up flow that reads and verifies identity documents and corporate papers from more than 50 countries in real time, spots visual forgeries, and extracts the data straight into the core system.
The engine cross-references each applicant against international sanctions and risk lists in a fraction of a second. Clean, low-risk applications are approved automatically with no human touch; anything doubtful is held and routed onward.
For medium and high-risk files, the system builds the investigator a focused summary dossier that points straight at the discrepancy or the suspicious element, so the human decision is fast, informed and final.
Photoshopped certificates, tampered watermarks and altered security marks that used to pass are now flagged before they reach the core system, closing the exposure that worried the regulator most.
The combined engine-and-analyst team processes more than 100 KYC and KYB checks and investigations an hour while holding a strict 96% quality-assurance score, so speed never comes at the cost of a wrong call.
Each approval and rejection is logged with the evidence behind it, so the firm can show a regulator exactly why any account was cleared or held, and apply the same standard every time.
The legitimate customer is through in minutes; the risky one meets a human.
For legitimate customers, opening an account fell from several days to a few minutes. The abandonment that came with the old queue disappeared, and the join rate climbed as the friction that had been costing the platform new business was taken out of the flow.
The firm met its international AML obligations without exception, while the operating cost of the compliance function fell sharply. Analysts stopped clearing routine paperwork and spent their time where judgement earns its keep, on the cases that genuinely needed investigating.
Manual compliance forces a bad choice: onboard fast and miss things, or screen properly and lose customers to the queue. The value here is a green lane that clears the clean cases instantly and a hard line that sends only genuine doubt to a person, whose ruling is logged and defensible. It is the same KYC discipline behind our bank underwriting work, aimed here at AML and merchant onboarding at payments scale rather than lending.
A bank's underwriting took three weeks and lost clients before they started. A secure pipeline with AI document abstraction and a green lane cut approvals to 24 hours, dropped false alerts 70%, and kept every record inside the bank.
Read the case → Fraud detection · InsuranceBlunt fraud rules were punishing honest customers and still missing the real thing. A real-time model scored risk in the moment, catching more fraud while letting genuine claims and payments through without the friction.
Read the case →
Reef TRH