Skip to content
TrustList
Funding news

Arva AI raises $20 million to automate bank back-office decisions

Editorial

By TrustList Editorial

Arthur Ventures led the Series A, with Y Combinator and Gradient returning; Arva says its agents already decide cases for a top-10 US bank, an S&P 500 payments firm and Wise.

About Arva AI raises $20 million to automate bank back-office decisions

Arva AI raises $20 million to automate bank back-office decisions

17 November 2026: Arva AI says it has raised a US$20 million Series A to build AI agents that take over the case-by-case decisions banks still make by hand, from financial crime checks to onboarding. Arthur Ventures led the round, and Y Combinator and Gradient, both earlier backers, took part again. The money goes into Arva's own model lab and into extending its agents to disputes, payment operations and lending.

Not yet independently verified. The round, its investors and the customer list rest on Arva AI's own announcement of 17 November 2026; we have not seen a post from Arthur Ventures or any customer. The 13% accuracy figure is from Arva's own testing. We will update this when it can be confirmed, and remove this note.

Decisions banks still review one at a time

Arva's argument is about the queue. Banks make decisions every day in financial crime compliance, onboarding, payment operations and lending, and the company says most are still reviewed manually, one case at a time, while the backlog grows. Its agents are meant to handle a case from start to finish, with people setting the policy and overseeing the results rather than working each file.

The company says that over the past year it has deployed at scale with a top-10 US bank, an S&P 500 payments institution and Wise, and that this year customers "began handing real critical decisions to Arva's agents". It does not name the bank or say how many decisions the agents now take. In June it announced a partnership with Fiserv to bring its agents to the banks and credit unions on Fiserv's platforms.

Why it built its own models

Arva's case for raising is accuracy. It says general-purpose models "weren't built for these decisions", are not accurate enough and often cannot show how they reached an outcome, which matters when a bank has to explain a decision to a regulator. In September it opened Arva Research, an in-house lab that builds models for the steps where accuracy matters most. Its first model, Arva Intel, researches people and businesses online, and the company says it beat leading general-purpose models on accuracy by 13% in its own testing. A layer it calls AgentCore adapts the agents to each bank's operations and policies over time.

Where the money goes

Arva lists two uses: more models from Arva Research and further work on AgentCore, and agents for more kinds of back-office decision, naming disputes, payment operations and lending. It says it will publish more of the lab's work, including how it evaluates its models, over the coming months.

For compliance and operations teams weighing tools like this, that promised evaluation method is the part to ask for. The company's own accuracy figure is measured against general-purpose models rather than against a bank's existing analysts or rules engine, and it is that second comparison a model risk review will want to see.

Company profile on TrustList: Arva AI

Sources

Categories & features

TrustList Weekly

The week in software and IT, in one email

The news that matters to buyers, new rankings and our own research. Every Thursday, free, and easy to leave.

We will email you to confirm. Unsubscribe with one click in any issue. Privacy policy

You may be interested in