Michael Polyani, the British-Hungarian philosopher, economist, and scientist, is perhaps best known today for coining the term “tacit knowledge.” His great observation was that a large part of what constitutes expertise in any given field is never written down. In some cases, it exists only as a kind of professional intuition that even the expert can’t fully articulate. “We know more than we can tell,” was Polyani’s famous catch phrase.
Today, tacit knowledge presents a challenge to companies that want to automate workflows with AI agents. Much—perhaps even most—of the knowledge these agents need is not written down.
Interloom, a Munich-based startup that is aiming to transform traditional business process automation for the AI age, thinks it can crack the problem of tacit knowledge. And it has just raised a new $16.5 million venture capital round to help it achieve that mission.
The funding is being led by DN Capital, with participation from Bek Ventures and existing investor Air Street Capital. The company previously announced a $3 million seed round in March 2024.
Interloom did not disclose its valuation after the new funding.
Fabian Jakobi, Interloom’s founder and CEO, argues that the current wave of excitement about AI agents overlooks the tacit knowledge bottleneck. About 70% of operational decisions have never been formally documented, he said. When a complex support ticket lands on a veteran staffer’s desk, they know the workaround, the right internal team to escalate to, and the resolution—not because it’s in a manual, but because they’ve seen it before.
“The most important person at the bank is the person who knows whether the documentation is right or not,” Jakobi told Fortune. “They’re often the lowest paid. But they determine quality.”
Interloom’s approach is to ingest millions of operational records—support emails, service tickets, call transcripts, work orders—and use them to build what it calls a “context graph,” a continuously updated map of how problems actually get resolved within a given organization. Jakobi likens the concept to Google Maps: just as Google learns optimal routes from real-time traffic data, Interloom maps the paths that operational experts take to solve problems, and uses those maps to guide AI agents and new employees alike.
Jakobi is a serial entrepreneur. He previously founded Boxplot, which in 2021 he sold to Hyperscience, a New York-based AI software company that specializes in extracting data from unstructured documents.
Interloom’s software is already live with several large European enterprises. At Commerzbank, Interloom analyzed millions of customer support emails and checked them against existing internal documentation—finding that much of it was either conflicting or incomplete. The company says it reduced the gap between documented and actual operational knowledge from roughly 50% to 5%. At Volkswagen, it is processing customer support tickets. And at Zurich Insurance, Interloom won a company-wide AI competition—beating out what Jakobi says were 2,000 other AI-native startups—for an underwriting use case.
An underwriting decision at an insurance firm, Jakobi said, reflects that company’s particular risk appetite, its accumulated experience with certain brokers and products, and institutional knowledge that no general-purpose model possesses.
“The Zurich underwriter knows how their broker chat underwriting works much better than Accenture does,” Jakobi said, taking aim at the large consulting firms that have traditionally dominated enterprise process work.
The broader argument is that AI agents, no matter how capable, are useless in large enterprises without organization-specific context. Jakobi frames this as the “corporate memory” problem.
“In software, the compiler tells you if the code works,” Jakobi said. “We don’t have that luxury [in other domains.] The evaluation has to come from a human expert.”
Interloom’s new backers agree with that thesis. Guy Ward Thomas, a partner at DN Capital, said that “an agent is only as good as the expert decisions it can rely on.” And Thomas said that DN Capital has seen with other AI agent startups that when these agents don’t have the right context about the enterprise in which they are being deployed, they rarely work well. “Our experience with vertical AI agents and voice platforms showed us how important context is,” he said.
Mehmet Atici of Bek Ventures previously backed UiPath, which had been the leader in the previous wave of RPA, or robotic process automation. But RPA relied on agents that were, for the most part, hard-coded to follow the same exact workflow in the same exact way every time. “We’ve seen automation’s transformative potential firsthand and we believe AI is now unlocking a new wave of rapid adoption in the enterprise,” Atici said.
Interloom’s timing may be propitious. The so-called “Great Retirement” is seeing roughly 10,000 Baby Boomers retiring daily in the U.S. Walking out the door with them is decades of institutional knowledge—just as companies are trying to deploy AI at scale.
Jakobi sees the competitive landscape in characteristically blunt terms. His biggest rival, he says, is inertia—the assumption within large enterprises that operations will continue to function as they have for the past decade.
Interloom’s next product push is what it is calling internally a “Chief of Staff”—a layer designed to give managers real-time visibility into how their AI agents are performing, complete with version control for agent-driven processes.
But Interloom is hardly the only company trying to create an AI agent management and orchestration layer. Almost every company marketing AI agents, from OpenAI to ServiceNow to Microsoft, has been working on similar kinds of products.
Jakobi, however, said that he thinks Interlooms “context graph” gives it a distinct advantage over these larger players, which he says rarely have insight across an entire complex process.
