US-based Arva AI, which builds AI models and agents to automate decisions for banks and fintechs, has raised USD 20 million in a Series A funding round led by Arthur Ventures, with continued participation from Y Combinator and Gradient Ventures. The funding will expand Arva Research and extend Arva's models and agents across the bank back office, from financial crime compliance to lending.
Models for high-accuracy decisions
Arva Research, launched in September 2026, builds proprietary models for the decision steps where accuracy matters most. Its initial model, Arva Intel, researches businesses and individuals online and, according to Arva's own testing, outperformed general-purpose models on accuracy by 13%. Paired with AgentCore, Arva's continual learning infrastructure, the model adapts to each bank's operations and policies.
Arva AI's deployments include a large US bank, an S&P 500 payments company, and Wise. The company says banks make thousands of critical decisions each day, most of which are still reviewed manually, and that general-purpose AI models are not accurate enough to make these decisions without human checks, especially where outcomes must be explained to regulators.
Trust and auditability for banks
Rhim Shah, Co-founder and CEO of Arva AI, said customers started handing real critical decisions to Arva in 2026, and that how far this goes depends on accuracy, which is why Arva Research is the main priority for the funding. He added that banks still run most of their back office manually, and that Arva is building models that let them hand this work to agents and trust the result.
Patrick Meenan, General Partner at Arthur Ventures, said banks need AI they can trust with their most consequential decisions, and that Arva combines domain expertise, technical capability, and a commitment to accuracy and auditability. According to the company, its models are designed to produce outcomes that banks can explain to regulators.