The 95 Percent Problem: Why Compliance Breaks After the Alert
An interview with Kartik Dabbiru, by Sean Murphy
Picture a financial crime analyst at the moment an alert lands. Five, perhaps ten systems are open across as many browser tabs: the alert itself, the customer record, the documents submitted at onboarding, the log of every conversation the firm has had with them. The analyst moves between them, following one branch and then another, until resolving the case has become an impressively convoluted decision tree. Somewhere along the way, as Kartik Dabbiru puts it, the analyst can often become lost in working out how to resolve the alert that they forget why they were there in the first place.
Dabbiru is the co-founder and chief executive of ComplyStream, a London RegTech that came out of stealth at the start of 2025. His argument, refined over nearly twenty years in financial services that began with almost a decade at JPMorgan, is that the industry has been solving the wrong half of the problem. Detection has been good for a long time, screening and transaction monitoring tools raise the flags they are supposed to raise. What happens next is where the mess begins, and it accounts for the great majority of the real operational cost.
The numbers explain why. In most financial crime operations, between 90 and 95 percent of alerts are false positives, and experienced analysts discount the bulk of them almost on sight. AI triage has made that first pass faster still. It is the residual five to ten percent, the alerts that cannot be waved through, where the pain concentrates. Each one sends the analyst into the tangle of systems, and each alert is working against the clock. The sales team wants to know why a lucrative customer has not been onboarded. The customer, having submitted everything asked of them, is already talking to three other providers.
At one of the fintechs where Dabbiru worked before founding ComplyStream, he recalls, the largest customer by revenue was close to walking, with five to ten percent of its payments stopped each day by suspected sanctions alerts. The platform he built to rescue the relationship did one simple thing: it let the compliance team talk directly to the end customer, gather what was needed to reach a decision quickly, and explain that decision afterwards. By his account the customer not only stayed but went on to push roughly three times the volume it had before.
That episode, he says, held the insight ComplyStream was later founded on, that what the analyst was really missing was context, scattered across silos and impossible to assemble at speed.
Dabbiru distinguishes what he means by AI native from the summarisation features bolted onto so many platforms. The system, he says, learns from the decisions made on it.
Approve a complex customer, and when a similar case next appears the model prompts the analyst toward the adverse media check that swallowed the most time before. It is the judgment layer, in his framing, rather than the orchestration layer.
When we talk about the “compliance brain” this is where Dabbiru becomes most animated. Anyone who has worked in financial services recognises the figure he is reaching for: the person who has been in the building for thirty years, has seen every problem that can exist, and has developed a sixth sense for where to start pulling the thread. Dabbiru wants that instinct merged into the machine, so an analyst cannot merely query a case but seek active guidance on it, in seconds rather than hours. He talks about lifting the speed and quality of decisions a hundredfold.
His second goal is a bit different but tells you as much about him. He wants the analyst's daily experience to stop feeling like software built in 2006, and describes himself, without much exaggeration, as maniacally obsessed with product experience, the kind of founder who lectures his team about it before breakfast. His conviction is that compliance has been written off as a pure cost of doing business, when a better compliance experience actually brings in more business. Onboarding is only the start. A held payment released quickly is what keeps a customer, and persuades them to send more payments, which is exactly what happened to the firm that tripled its volumes. The financial crime analyst was never standing in the way of growth, contrary to what the rest of the business was suggesting.
The real reason was ten open tabs and a lack of context.
