Wall Street firms know the pains of having to satisfy regulators, and with advancements in artificial intelligence, there's a whole new level of scrutiny and complexity.
Banks have to ensure that the AI and machine-learning models deployed to employees and clients won't have unintended consequences, like an AI model producing racially inaccurate depictions of historical figures. It involves quality-and-assurance teams who test, verify, validate, and challenge the AI models before they go into production and report issues back to the engineering team to fix. It also includes people dedicated to monitoring solutions in real time to make sure they're behaving as expected.
In some cases, banks' technology teams are spending at least half their time documenting models for compliance, Jonas Jacobi, cofounder of fintech startup ValidMind, told Business Insider.
It's not uncommon to find a data scientist, a "Ph.D., fully loaded, costing you $400,000 a year," inside a bank spending 20 hours a week documenting what models do and how they work, Jacobi said, adding that ValidMind's software can cut that down to four hours.
ValidMind is a startup that speeds up time-intensive parts of the risk-management process. The two-year-old fintech automates the collection of information from the quality-and-assurance tests and produces a digital document that plots out charts, diagrams, and tables.
The Palo Alto, Calif.-based ValidMind announced Wednesday it raised $8.1 million in a seed round led by Point72 Ventures, the venture-investing arm of Steve Cohen's hedge fund.
The company was founded in March 2022 by Jonas Jacobi, its CEO; Mehdi Esmail, its chief product officer who previously spent five years at American Express; and Andres Rodriguez, its CTO. The new funding pooled from eight VCs, including Third Prime Ventures and New York Life Ventures, brings ValidMind's total funding to $11.1 million. ValidMind declined to disclose its valuation.
Catching up the back office
As AI continues to gain momentum on Wall Street, the pace of banks' tech development is still being hamstrung by the back office, according to Jacobi.
Technologists on Wall Street have experienced a rise in tools that speed the development of AI solutions and machine-learning operations, but the "AI risk process inside the bank did not mature" enough to "keep up with that pressure from inside the business units to deploy more AI," Jacobi said.
A key piece to proving compliance is documentation. AI models are often compared to a "black box" because it can be difficult to understand their decision-making or how the models arrive at their results. For finance firms, "showing your work" consists of two aspects: the quantitative information being produced by the models (i.e. the output) and qualitative information explaining the reason for the output, Jacobi said.
ValidMind takes the qualitative and quantitative aspects and spins up documentation for the data scientist to review, edit, and sign off on. With the new injection of capital, ValidMind will build more features, like one aimed at automating the risk management of third-party vendor solutions. The startup is working with four banks testing proofs of concept, but Jacobi declined to disclose client names.
