Paul Stokes spent nine years turning down investors. On Wednesday, the co-founder and chief executive of Prevalent AI explained what had changed: not his company, but the market's understanding of what it was selling.
The London firm said it has raised $22 million from Integrity Growth Partners, the first primary capital it has taken since it was founded in 2017. Prevalent says it has been profitable since landing its first customer, has never taken growth equity, and saw annual recurring revenue more than double over the past twelve months. It declined to disclose the revenue figure, the valuation, or IGP's stake. The only prior change to its cap table came in July 2021, when Istari, the Temasek-backed cybersecurity platform, took a minority position through a secondary that put no new money in.
"We believed it was possible that, if we raised money too early, we might have been raising it at the wrong time, since the market had not yet recognised what was required," Stokes told Tech Funding News.
What Prevalent actually sells
Not a model. A data fabric that reaches into hundreds of separate enterprise systems — security platforms, cloud logs, identity providers, asset inventories, internal databases — cleans and deduplicates what comes back, and reassembles it as a continuously updated knowledge graph of what exists inside an organisation and how it connects. The company calls the result sovereign: the graph sits inside the customer's own infrastructure rather than a shared third-party cloud, and the customer decides where the underlying data physically lives.
The construction is deliberately deterministic. Prevalent does not use large language models to build the graph. "You can't use AI for that because it would lead to assumptions and deviations in the graph that aren't accurate," Stokes said. "We construct the graph in the same way each time." Human analysts and AI agents query the finished artefact.
Roughly 200 people work on it. Customers include global banks, telecoms carriers, insurers and critical national infrastructure operators, typically organisations of more than 5,000 employees and sometimes north of 100,000. The company cites an international banking group that improved incident detection by more than 80% and a global insurer that cut the time needed to produce executive security reports by 95%. Neither customer is named and neither figure has been independently verified. The founding group carries British intelligence heritage: Sir Iain Lobban, director of GCHQ from 2008 to 2014, has sat on the board, and co-founder Andrew France, a former deputy director for cyber defence operations there, went on to run Darktrace.
Why It Matters
The pitch is that enterprise AI is failing at the plumbing layer, not the model layer, and the evidence has accumulated fast enough to make data reconciliation look like a growth story.
Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027, blaming escalating costs, unclear business value and inadequate risk controls. MIT Media Lab's NANDA initiative reached a blunter conclusion in its 2025 report "The GenAI Divide: State of AI in Business," finding that roughly 95% of enterprise generative AI pilots produced no measurable impact on profit and loss — and attributing the failures not to model quality but to the integration gap between the tools and the organisations deploying them.
Stokes' version of the diagnosis predates the round. "Large enterprises do not have a shortage of tools or data. They have a shortage of context," he said. "Security teams are being asked to make decisions across thousands of systems, controls, identities, and data sources that were never designed to work together."
The distinction matters commercially because it reverses the direction of the fix. If agents fail because models are not clever enough, the remedy is to wait for a better model, which costs the customer nothing. If they fail because the enterprise cannot tell the agent what it owns, who has access to it and which controls are live, the remedy is a multi-year integration project somebody has to be paid for. Frontier models, in Stokes' framing, are already plenty smart enough. An agent handed a wrong picture of a company simply reaches the wrong conclusion faster than a human would.
Ryan Anderson, managing partner and co-founder of Integrity Growth Partners, priced that thesis at roughly a tenth of his fund. "Paul, Arun, and the team have built something rare: genuinely differentiated, AI-native technology that the most sophisticated enterprises in the world rely on, all while maintaining remarkable capital discipline," he said. The Santa Monica firm, which backs bootstrapped software and tech-enabled services companies, closed an oversubscribed $220 million fund in December.
What To Watch
The money buys the thing Prevalent has never had: a sales organisation. Stuart Barnard joins as chief financial officer and Mike East as senior vice president of global sales, with the capital funding a formal go-to-market structure spanning sales, marketing, customer success and partnerships, plus a deeper push into the United States — expansion that cuts against recent traffic, with US data-security firms including Rubrik making London their European base.
The second test is scope. Prevalent intends to extend the graph past security into financial crime analysis, compliance and wider operational risk — selling to buyers who have never heard of it, against incumbents who own those budgets. Revenue has roughly doubled each year; Stokes concedes that gets harder at scale.
Then there is the moat. Rivals in AI-native data tooling insist deterministic reconciliation is easily replicated; Prevalent's counterargument is nine years of doing it the boring way, which is not a claim a competitor can refute quickly, or a customer verify quickly either. The deeper question is whether the discipline that made this round unusual survives a quota-carrying sales team.
“Large enterprises do not have a shortage of tools or data. They have a shortage of context.”— Paul Stokes, Co-Founder and CEO, Prevalent AI