LONDON, September 16 — Arva AI, the fintech startup founded in 2024, has launched a research division dedicated to building the models and infrastructure that let banks automate high-risk decisions. The company announced the Research Lab on its LinkedIn page, saying it was built to close the gap between general-purpose AI models and the accuracy and control banks require for high-stakes automation. The shift here is not another chatbot.
It is the machinery underneath the decision. After more than 5,000 hours of research, training and evaluation, the Lab has produced proprietary models for data enrichment, transaction analysis and evidence-based reasoning.
It has also built infrastructure called AgentCore, which turns analyst corrections, insights and case outcomes into system improvements that are backtested, evaluated and versioned before they reach a live decision.
Arva Intel and the precision gap
Arva’s first model, Arva Intel, was developed to research suspicious individuals and businesses online. Evaluated independently against frontier models, it outperformed general-purpose models by 13% on precision. The benchmarking did not rest on the final case outcome alone; it measured the accuracy of each component inside the decision.
Founder and CEO Rhim Shah framed the problem plainly. “Banks keep humans in the loop because no AI has been accurate enough to remove them safely—that’s the problem the Lab solves,” he said.
“Our models and AgentCore let us automate these decisions with the accuracy and control banks require, and this is just the start.” The launch lands as financial institutions explore ways to use AI to manage and review financial crime and fraud cases, along with other back-office operations where cumbersome manual review by large teams can expose firms to financial losses, customer harm or regulatory breach. Keeping a human in the loop has become a requirement in many AI deployments, but general-purpose models remain too inconsistent and inaccurate for high-stakes automation.
Already live in production
“We built the Lab to close that gap,” the company noted on LinkedIn. “Proprietary decisioning models purpose-built for the highest-risk parts of a decision, and AgentCore, infrastructure that turns analyst corrections and case outcomes into tested, versioned system improvements. These are already live in production with leading financial institutions and we’re excited to finally be talking about it publicly.” Financial crime and fraud are the initial focus for the Lab.
Arva anticipates that the Lab will expand over time into payment exceptions, disputes and other customer-related investigations, and the company said it plans to publish its benchmark methodology and research at academic venues soon. Arva AI made its Finovate debut at FinovateEurope 2026 in London.
The company is backed by Google’s Gradient Ventures, Y Combinator and other leading funds. Watch this space. The next signal to track is that published benchmark methodology — because a precision claim only becomes an industry standard once rivals can test it.
