Rhim Shah used to lead the financial crime product team at Revolut Business. He knew how many people a bank needs just to sit and clear alerts all day. In 2024 he started a company to hand that work to software.
That company, Arva AI, has now raised $20 million in a Series A round led by Arthur Ventures. Earlier backers Y Combinator and Gradient, Google’s venture fund for AI startups, both put in more money. The news came out on November 17, 2026.
A Series A is a startup’s first large round of venture money, usually raised once real customers are paying for the product.
Arva has those. A top-10 US bank. An S&P 500 payments company. And Wise, the money transfer fintech.
The alerts that land on someone’s desk
Every bank has to check its customers and payments against sanctions lists and watch for transactions that look like money laundering. The screening systems raise alerts, and each one needs a review. At most banks, a person still does it.
Arva builds AI agents, which are programs that carry out a job with several steps on their own instead of answering a single question. Its agents research the business or individual behind an alert, read the documents and work the case through to a decision.
The company started smaller than that. It came through Y Combinator’s summer 2024 batch with a tool for speeding up the checks a bank runs before it accepts a new business customer. A $3 million seed round led by Gradient followed in January 2025. Agents have spread well beyond compliance since then, as with Desible.ai and its bet that a phone call is only the front door, but few jobs carry the penalty for a wrong answer that this one does.
The year customers let go of the wheel
Plenty of banks will test an AI tool. Far fewer will let it decide anything. According to Shah, that changed for Arva in 2026: customers began handing real decisions to its agents, with their own staff setting the policy and overseeing the results instead of reviewing every case by hand. For a company two years old, selling into one of the most heavily regulated corners of finance, that is the whole story of this round. An investor can count customers, but the harder thing to buy is a bank’s willingness to stop double-checking, and Arva says it now has that at three institutions large enough to name by category, plus a partnership with Fiserv, signed in June, to bring the same automation to banks and credit unions.
“This year our customers started handing real critical decisions to Arva.”
Rhim Shah, co-founder and CEO, Arva AI
Shah sends the money to the lab first
In September, Arva opened Arva Research, a lab that builds the company’s own models for the steps where a mistake costs the most. The company’s press release says the new funding goes there before anywhere else.
It is a choice other young AI companies have made too. Biomatter spent its Series A on a lab of its own for much the same reason.
The lab’s first model is Arva Intel, which digs through online information about people and businesses. Arva says it beat leading general-purpose models on accuracy by 13%. That figure comes from Arva’s own testing.
A second system, AgentCore, learns each bank’s policies and adjusts the agents over time. Oliver Wales, the chief technology officer, came to Arva from Seal, where he was chief product engineer.
Some of the money is headed for people as well. Axios reported that Arva plans to grow its sales team in New York, where it has an office alongside London.
So the software decides, and the humans write the rulebook?
That’s the part I keep coming back to.
Shah spent years inside a compliance operation, so he knows what a missed sanctions hit costs a bank. And his first move with $20 million is
