Cobalt Pay knotted a fraud ring in real time.
A static rules engine kept missing coordinated fraud while flagging good customers. Knot detection wove transaction, device and support threads — and caught the ring as it formed.
Rules that fought yesterday's fraud.
Cobalt Pay's fraud defence was a thicket of static rules. It was simultaneously too tight — declining real customers and generating a flood of false positives — and too slow, because new fraud patterns outran the team's ability to write rules for them. Reconciliation across payment providers took days, so losses were often confirmed long after the money was gone.
"We were tuning a smoke detector by hand while the building changed shape," said Inès Lefèvre, Head of Risk.
Six threads, watched continuously.
Rather than write more rules, Cobalt mounted their signals as live threads and let the loom watch the whole weave for anomalies — tying a knot the instant a pattern frayed, with the likely cause already worked out.
One knot caught the ring.
“It caught a ring we hadn't even named yet — and stopped declining the customers we actually wanted.”
Less fraud, fewer false alarms, faster books.
Within a quarter, fraud loss fell 58% while false positives dropped 42% — fewer good customers turned away, less money out the door. Reconciliation that once took days now closes 3.5× faster, and the loom ties most knots in under two minutes, long before a loss is realised.
"The rules didn't disappear," Lefèvre said. "But now the loom finds the pattern, and we write the rule once we understand it — not the other way around."
Read your own cloth.
Weave a free sample and see your first pattern in minutes.