Financial infrastructure company Plaid unveiled Thursday a new transactional foundational model, Chief Technology Officer Will Robinson shared with This Week in Fintech exclusively.

The San Francisco-based startup, which is best known for connecting bank accounts to financial applications, says it is entering a new phase of development centered on artificial intelligence. Robinson said Plaid is positioning the technology as the basis for what it is describing as “intelligent finance.” And it’s using de-identified data across its network to train the model.

In an interview discussing the initiative, Robinson described the shift as a natural evolution from the fintech’s past decade of open finance. The next stage, he said, will center on making those connections smarter, more personalized, and more responsive to user behavior.

Interestingly, Plaid chose to build its new foundational model internally rather than rely on third-party providers. Robinson cited three main reasons behind that decision: access to large-scale financial data not publicly available, the need for specialized training tailored to financial use cases, and Plaid’s long-standing experience navigating the regulatory and security complexities of the financial ecosystem.

“We think that having AI make use of financial data properly requires a bunch of idiosyncratic-like specific things,” Robinson told This Week in Fintech. “Models are only as good as the input data that they have and we’ve got a unique data set, which we can provide custom training on.”

General-purpose AI systems are better designed broad reasoning and conversation tasks, he said, whereas financial interpretation requires domain expertise – such as distinguishing income from transfers, assessing payment risk, or detecting fraud patterns. 

“We have more than a decade of experience dealing with all the stakeholders in this space – consumers, financial institutions, fintech applications, regulators of all different stripes, said Robinson, who worked at Coinbase as its VP of engineering for nearly 4 years prior to joining Plaid. ”You can't just move fast and break things and ignore the complexities of the finance space – just like the health space, or other spaces where there's a lot of sensitivity, and a lot of regulation – you need that industry understanding and the patience and the ability to really engage with that productively.”

Plaid acknowledges that AI companies are building “powerful” general-purpose AI models that are useful for understanding concepts, generating insights, and improving financial literacy. However, the company believes that it “operates at a different layer of the stack.”

“We build the financial data infrastructure and specialized models that understand how money moves across accounts, transactions, income patterns, fraud signals, and payment behavior,” Robinson said.

Despite the fact that Plaid is building the model’s core capabilities in-house, it confirmed interoperability with major AI ecosystems, including systems from OpenAI, Anthropic, and Mistral AI.

Early results

Founded in 2013, Plaid got its start as a company that connects consumer bank accounts to financial applications but has since been gradually expanding its offerings to also include lending, identity verification, credit reporting, anti-fraud, and payments. The company has raised about $1.3 billion in funding over its lifetime and at one point was going to be acquired by Visa before that deal fell apart due to regulatory concerns. 

Early results from the company’s AI-driven tools have already shown measurable improvements, Robinson claims, including increased fraud detection performance and stronger payment risk assessment accuracy.

“We’re still protecting you from fraud,” he said. “We're just doing an even better job of it.”

The model, the company says, can also more reliably distinguish between payroll deposits and peer-to-peer transfers, resolve fragmented merchant names across institutions, and identify subscription services even when naming conventions vary.

For consumers, the changes will largely appear as gradual improvements rather than dramatic new features. Financial apps that rely on Plaid’s infrastructure are expected to become more reliable and secure, while new AI-driven financial services – ranging from personal finance assistance to investment tools – may emerge as developers build on the enhanced capabilities.

The company said adoption of the new model is already underway in production environments, with broader integration planned through a staged rollout. Executives emphasized that changes will be implemented cautiously to maintain system stability and data protection.

Security and privacy remain central to the strategy, Robinson emphasized, pointing to layered protections, strict permission controls, and expanded investment in infrastructure designed to sensitive financial data.