It is Tuesday morning at a mid-sized credit union in the Midwest. A wire transfer for $48,200 pings an analyst's queue. Six months ago she would have spent forty minutes piecing together device logs, watchlist hits, and a behavioral pattern that may or may not matter. Today an AI agent has already drafted the narrative, pulled the device fingerprint, charted the counterparty graph, and queued the SAR for her signature. She approves, types two sentences, moves on. The agent was built by a company most consumers have never heard of: Unit21.
That is the trick of being plumbing. Unit21 does not sell to you. It sells to the people who decide whether the money you just moved looks like crime.
The compliance industry was built for a slower decade.
In 2018, Trisha Kothari was an engineer at Affirm watching the same scene play out across the fintech she knew well. New products shipped weekly. The fraud and AML stack underneath them was assembled from vendors that traced their lineage to the 1990s - rules written in proprietary syntax by analysts who had to file tickets to update them, then wait for a quarterly release. The bad guys, meanwhile, were iterating in real time.
Clarence Chio - co-founder, security researcher, and author of an O'Reilly book on adversarial machine learning - saw the same gap from the other side. Detection models were getting smarter every year. The investigation tools wrapped around them were not. An analyst's day was 80% pattern-recognition busywork and 20% actual judgment. The ratio was, charitably, upside down.
Build the infrastructure. Let the analysts write the rules.
Kothari and Chio's bet was a vocabulary shift. Don't sell a fraud product. Sell a risk operating system - one where the people closest to the problem could change the logic themselves. A no-code rules engine. A graph view of every entity and counterparty. A unified case manager. APIs to ingest anything: transactions, logins, device signals, KYC pings.
It is the kind of pitch that sounds boring in a deck and revolutionary in a procurement meeting. Investors who could read a balance sheet caught on. ICONIQ led the Series A in 2020. Tiger Global and Google's Gradient Ventures piled into a Series B in 2021. By 2023 a $45M Series C - again led by ICONIQ - had taken total funding to roughly $137M.
The Unit21 Timeline
A toolkit, then a workspace, then an agent.
The Unit21 platform has three layers a customer ever has to think about. The first is monitoring: transactions, devices, logins, behavioral signals stream in through APIs, get scored by rules and models, and surface as alerts. The second is investigation: each alert lands in a case management workspace that draws the graph of related entities, surfaces evidence, and lets the analyst annotate. The third - the one that defines 2026 - is a fleet of AI agents that do the first two layers' grunt work on their own.
The agents are not chatbots bolted onto an existing dashboard. Unit21 actually rebuilt the platform around them. The agents triage alerts, pull supporting evidence, run graph queries, draft regulatory narratives, and queue everything for a human to sign. The humans keep judgment; the software absorbs the keystrokes.
Customers who do not get to be wrong.
The Unit21 customer list is unusual because every name on it operates under a regulator. Chime, the largest neobank in the US. Brex, the corporate card the venture industry uses to expense its lunches. Crypto.com and Nexo, two of the largest crypto exchanges still standing. Intuit, the company behind TurboTax. Green Dot, FIS, Jack Henry, Sallie Mae, Patriot Bank, Kinecta Federal Credit Union, American First Credit Union. A spread that runs from Silicon Valley fintech to community banks in towns with one stoplight.
The numbers behind the logos do the convincing. Roughly five percent of every Suspicious Activity Report filed in the United States now flows through Unit21's pipes. 4.5 billion events are monitored each month. 48.3 million analyst hours have been returned to compliance teams that used to spend them stitching screenshots together. The 2025 AI rebuild has, by the company's count, reduced false positives by up to 93%.
Where Unit21's Numbers Land
Compliance is the rare industry where shrinking a number on a chart is the entire point.
Who's running on Unit21
A short tour of the customer list - banks, neobanks, crypto exchanges, payment platforms, and the credit unions you've never heard of that nevertheless write the same SARs as the giants.
Software that handles the paperwork. Humans that handle the calls.
If you read Unit21's own writing on the topic, the mission is straightforward enough to fit on a coffee cup: automate the detection, investigation, and reporting of financial crime so compliance teams can focus on real threats. The unstated part is more interesting. Compliance officers are the people who, when something goes wrong, get called into rooms with lawyers. Unit21's pitch is that the right software does not replace them. It moves them from the spreadsheet to the conversation.
There is also a quieter argument about scale. The cross-institution fraud consortium that Unit21 runs lets member banks share fraud signals without sharing customer data - a hard technical problem and a harder political one. The premise is that no single institution is going to win the fraud arms race alone, and that the only counterweight to coordinated criminals is, well, coordinated banks.
Crime adapts. Infrastructure has to adapt faster.
There is a version of this story where Unit21's bet on agentic AI looks obvious in five years and quaint in ten. There is another where it turns out to be the moment the regulated economy stopped trying to fight modern fraud with a modem-era stack. Trisha Kothari has been publicly blunt about the stakes - "we're going to lose if we don't use AI," she told FinTech Global earlier this year. It is not a marketing line. It is a directive to her peers.
The point of building infrastructure is that you do not have to win the news cycle to win the decade. You just have to be the thing other people build on. By that standard, Unit21 - quietly, with a logo most people couldn't pick out of a lineup - is already winning.
Back to the credit union analyst on Tuesday morning. She is done with her queue by lunch. Six months ago that sentence would have ended differently - it would have ended with "she is done with her queue by Thursday." The wire is filed. The narrative is on a regulator's desk. The fraud ring on the other end is going to need a new bank. Unit21 has, again, been the thing nobody noticed and the thing that made the difference.
It is, on its face, a very boring business. It is also, on its face, the kind of business that the next decade of fintech quietly runs on.