date

category

Announcement

read time

4 mins

Today we're announcing our $20M Series A, led by Arthur Ventures with continued participation from Y Combinator and Gradient.

Banks still make most of their back-office decisions by hand. That is starting to change. This year our customers began handing real critical decisions to Arva’s agents, while people set the policy and oversee the results. Arva Research is how we make that accurate enough for banks to trust.

Over the past year we’ve deployed at scale with banks and fintechs including a top-10 US bank, an S&P 500 payments institution and global fintech Wise. Thank you to our customers, our investors and our team for getting us here.

Why accuracy comes first

Banks make decisions every day in financial crime compliance, onboarding, payment operations, lending and many other places. Most are still reviewed by hand, one case at a time, and the queues keep growing.

AI can take these decisions on, but only if the bank can rely on every outcome and explain it to a regulator. General-purpose models weren’t built for these decisions. They aren’t accurate enough= and they often can’t show how they reached an outcome. The more accurate the system, the more decisions a bank can hand over.

How Arva gets to accuracy

Arva deploys AI agents that handle critical banking decisions end-to-end, across financial crime compliance, onboarding, payment operations and lending. Accuracy at this level doesn’t come from a single model. It comes from the models, the agents and the way they learn working together.

In September we launched Arva Research, a dedicated lab building our own models for the steps in a decision where accuracy matters most. Our agents use those models to handle each case from start to finish, and AgentCore, our continual learning infrastructure, adapts them to each bank’s own operations and policies over time.

Arva Intel, our first model, does deep research online on individuals and businesses. In our testing it outperformed leading general-purpose models on accuracy by 13%.

What the funding is for

The funding will go toward:

  • Arva Research: building more models for the steps in a decision where accuracy matters most, and extending AgentCore, our continual learning infrastructure.

  • More of the bank back office: extending our agents into more types of decisions, including disputes, payment operations and lending.

We'll share more of Arva Research's work over the coming months, including how we evaluate our models.

If you run financial crime, fraud, operations or lending at a bank or fintech and want to see what Arva could take off your team's queue, get in touch. If you want to work on these problems with us, we're hiring.

The latest

News & Research

Research notes, product updates, and company developments — straight from the team.