Johannes Jaeckle

The man teaching machines to read the boring stuff

Somewhere right now, a loan officer is squinting at a PDF bank statement that arrived as an email attachment, retyping numbers into a system that will spit out a decision. Multiply that by a few hundred thousand a week. That blizzard of paperwork is Johannes Jaeckle's raw material - and his opportunity.

Jaeckle is the co-founder and CEO of Heron Data, a New York company that uses large language models to do the part of financial services nobody brags about: opening emails, reading the attachments, classifying what they are, pulling out the numbers, and pushing them into a CRM or a loan platform. The software does the typing, the sorting, the chasing-for-missing-info. Humans get to keep the judgment.

It is not a flashy pitch. It is a $300 billion one. "Many industries still rely on human teams to process unstructured documents," Jaeckle has said, "creating a $300B+ global spend on business process outsourcing." Heron is going after that spend with a stubbornly practical promise to its customers: don't change anything. "We don't ask customers to rip out their infrastructure or adopt a new system." The AI slides in behind the inbox they already have.

The numbers do the convincing. Heron processes more than 350,000 documents a week for over 150 customers, a roster that runs from small specialty lenders to FDIC-insured banks and commercial insurance carriers. One customer cut its submission-to-decision time by 60%. Another, Jaeckle likes to point out, "scaled revenue 4x without hiring." In insurance, the platform now triages more than 80% of incoming submissions automatically.

Three friends, an LLM, and a two-year head start

Here is the detail that separates Heron from the 2023 gold rush: Jaeckle and his co-founders started building with large language models in 2020. ChatGPT did not arrive until November 2022. By then, this team had already spent roughly two years shipping LLM-based products for financial services - learning the hard way where the technology breaks and where it holds.

The motivation was almost embarrassingly simple. "We started the company because we were friends fascinated with technology and startups," Jaeckle has said. The three friends were Jaeckle, Dominic Kwok, and Jamie Parker. Kwok had been Head of Data Science at Revolut, with earlier stops at Facebook and Spotify. Parker had run a multimillion-dollar P&L at AlphAsights. They went through Y Combinator's Summer 2020 batch and named the company after a bird.

Jaeckle did not arrive as a first-time operator. He had been an early team member and UK Managing Director at Segovia Technology, then went on to run the remittance fintech Taptap Send - the unglamorous, high-stakes business of moving money across borders for people who cannot afford for it to go wrong. He studied PPE at Oxford, then stayed for a Master's in Development Economics. The throughline is unmistakable: systems that have to work, for people who are counting on them.

Why he won't promise 100% accuracy

In a market crowded with founders claiming their AI never makes a mistake, Jaeckle made candor a feature. "Anyone who tells you they use AI to automate work with 100% accuracy is probably lying to you," he says. It is a striking thing for a CEO to put in print - and exactly the kind of thing a bank's risk team wants to hear before it lets software near its decisions.

That realism shapes the product. Heron is not, in Jaeckle's framing, a fancy text scanner: "We're not just doing OCR." The system understands documents well enough to act on them, but it is built to flag what it is unsure about rather than guess. The bet is that reliable-with-honest-edges beats perfect-on-paper, especially in regulated corners of finance where a confident wrong answer is worse than a flagged one.

A Series A that closed in roughly a month

In July 2025, Heron announced a $16.6 million Series A led by Insight Partners, with Y Combinator, BoxGroup, and Flex Capital joining. The round brought total funding to $23.3 million. The pace told its own story - by Jaeckle's account, about a month of raising plus a month of due diligence. When the product is quietly chewing through 350,000 documents a week, the deck mostly writes itself.

The plan from here is less about a single moonshot and more about patient expansion: more verticals, more workflows, the same wager that armies of people doing manual data entry would rather be doing something else. SMB lending and insurance came first. The map keeps unfolding.

The CEO and the Rockaways

Ask Jaeckle about a New York summer and the answer is not a productivity hack. It is swimming at the Rockaways and eating at Tacoway Beach - the far-flung surf-and-tacos corner of Queens that most Manhattan founders never make it to. There is a tidy symmetry to a man who spends his workdays deleting other people's drudgery and his weekends in the ocean. The mission, after all, is to give people back a lifetime of boring work. He seems intent on spending some of his own.