It has been a good summer to work at ThetaRay. In July, CNBC named ThetaRay to its World’s Top Fintech Companies 2026 list, in the Regtech category. This week, Datos Insights named Ray, ThetaRay’s agentic AI investigator, Silver Medalist for Best Financial Crime Investigation and Reporting Innovation in its 2026 Fraud & AML Impact Awards.

The second one is personal, because I lead product for Ray. The Datos awards only consider innovations that are actually in production, and they judge them on innovation, fit with market needs, risk mitigation, business impact and the strength of the roadmap behind them. 

“This year’s Datos Insights’ Impact Awards in Fraud & AML celebrate innovations that deliver on all three—solutions powered by AI and real-time decisioning that lower false positives while also reducing risk, integrate disparate data sources, and deliver compliance teams with the intelligence they need to keep pace with threat actors,” said Chuck Subrt, Practice Director, Fraud & AML at Datos Insights.

This category in particular rewards work that modernizes alert investigation, case documentation and regulatory reporting so that  analysts can redirect their capacity toward complex, high-risk cases. That is a precise description of what we set out to do, and having independent analysts say we are doing it means a lot to a team that spent the last year inside a bank’s security perimeter making it real.

Recognition is a moment, though, not the point. What I want to use it for is to describe what “agentic investigation” looks like once it is running inside a real compliance operation, in our case first at a tier-1 global bank, because that is where the phrase either means something or it doesn’t.

What an investigation actually is

A transaction monitoring alert fires. Before any judgment happens, the analyst assembles: who this customer is, what the KYC file says, what the activity looked like over the period, what previous alerts on them concluded, whether any screening hits matter, what the attached documents say. Then they write all of it down in a form that will survive an audit two years later.

Most of the day goes to assembling and documenting, not to deciding. Datos Insights put it precisely: institutions are under pressure to detect suspicious activity and to document and defend every investigative decision, at a volume manual workflows struggle to sustain. Two analysts working the same alert can reach different conclusions. Evidence capture is incomplete. Quality assurance finds the gaps after the fact, when they are most expensive to fix.

What “agentic” means here, and what it does not

An agentic investigation flips the order. When the alert fires, Ray runs the investigation before anyone opens the case. It gathers the profile, the activity, the history, the screening context and the documents, correlates them against a domain model of more than fifty risk indicators mapped to red flags and typologies, and writes a structured report in which every finding cites the evidence it rests on. The analyst opens a prepared case, reviews it, challenges it, and decides.

Datos described the shift as moving “from analyst-driven execution to AI-led investigation with analyst validation.”

That is exactly right, and the second half matters as much as the first. Ray recommends. It does not close cases, it does not escalate them, it does not change a state. Nothing leaves the analyst’s hands. We built it that way on purpose, and in a regulated operation that is not a limitation. It is the feature that makes adoption possible.

What changes for the analyst

In production, the job changes shape. Less assembling, more judging. A new analyst reviews a structured, evidenced case instead of building one from a blank screen, and documentation becomes a by-product of the investigation rather than an after-the-fact chore. Because Ray runs inside the bank’s own environment and every finding cites its source, the first line, the second line and the auditors can all see exactly how a conclusion was reached.

That is the point of the award category, and it is the point of the product: capacity moves from gathering to the complex, high-risk analysis that genuinely needs a human.

The medal recognizes a team, a customer willing to go first, and an approach: AI that does the investigative legwork in the open, cites its evidence, and leaves the decision where it belongs, with the people accountable for stopping the crime behind the alert. None of that is specific to the largest banks. Any institution that has to defend its decisions at volume, from a global bank to a fintech, has the same problem. Agentic investigations are not coming. They are already running in production. The interesting work now is making them ordinary.