New York-based Arva AI has raised a $20M (about £15.1M) Series A led by Arthur Ventures. Existing backers Y Combinator and Gradient Ventures, Google's early-stage AI fund, also took part.

Banks still make most back-office decisions by hand. Analysts screen customers against sanctions lists, research businesses at onboarding and review flagged transactions, repeating the same checks thousands of times. The work is slow and expensive, and mistakes carry regulatory risk. That has made banks wary of handing it to general-purpose AI.

Arva deploys AI agents that take on those decisions, while people set the policy and oversee the results. In September it launched Arva Research, an in-house lab building proprietary models for the steps where accuracy matters most. Its first model, Arva Intel, researches businesses and individuals online. Arva says its own testing shows the model beats leading general-purpose models on accuracy by 13%. A continual learning layer, AgentCore, adapts the models to each bank's operations and policies.

Arva was founded in 2024 by CEO Rhim Shah, who previously led financial crime product at Revolut Business, and CTO Oliver Wales, formerly chief product engineer at Seal. The company went through Y Combinator's Summer 2024 batch and raised a $3M seed led by Gradient in January 2025. Its agents are now running at a top-10 US bank, an S&P 500 payments company and Wise.

The new capital goes first into Arva Research. The aim is to extend Arva's models and agents across bank back-office work, from financial crime compliance to lending.