The AI that reads the paperwork nobody wants to - bank statements, tax forms, insurance claims - and hands finance back its afternoons.
Exhibit A: Heron's calling card. Somewhere behind this image, a transformer model is quietly parsing a PDF that a human swore they'd get to "first thing Monday."
It is Monday morning at a small-business lender, and the inbox is already losing. Bank statements as photographed PDFs. Tax forms scanned upside down. Applications half-filled by people who hate forms as much as the underwriter who has to read them. Somewhere in that pile is a good loan and a bad one, and a human has to tell them apart by Friday.
Heron Data built the thing that opens the pile. Its software ingests the documents, reads them, pulls out the numbers that matter, checks them, and drops clean, structured data straight into the lender's CRM - before the coffee is cold. The underwriter still makes the call. They just no longer spend three hours retyping a checking-account balance to get there.
“We want to free people from the repetitive, mindless work that makes you dread getting out of bed on a Monday.”
Lending, equipment finance and insurance share an unglamorous secret: the work is mostly paperwork. Every deal, every claim, every application arrives as a document, and not a tidy one. The data exists - it is just trapped inside formats designed for human eyes, not machines.
For years the answer was people. Armies of analysts keying figures from one screen into another, the digital equivalent of moving water with a teaspoon. The teaspoon, it turns out, scales poorly. Heron's bet was that the teaspoon was about to be replaced - not by more people, but by models that could actually read.
“Many of the repetitive parts of white-collar work will disappear in the next decade.”
Johannes Jaeckle, Dominic Kwok and Jamie Parker started Heron in 2020 - early enough that betting a company on large language models still counted as a slightly eccentric thing to do. The conviction arrived quietly. One night, co-founder Dom retrained an open-source transformer model on transaction data. By morning it was reading financial language with a fluency that looked, unsettlingly, human.
That was the pivot. Heron - then operating under the deeply literal legal name Open Credit Technologies - went LLM-first well before the rest of the world discovered chatbots. The discipline that followed was almost more impressive than the timing: the team reached $1M in annual recurring revenue with three founders and two employees. Five people. Capital efficiency as a personality trait.
“$1M in recurring revenue. A team of five. The kind of math that makes investors lean forward.”
Jaeckle, Kwok and Parker launch the company (as Open Credit Technologies) to help fintechs make sense of bank-transaction data.
Backed by Y Combinator, BoxGroup, Flex Capital, Musha Ventures and a cluster of fintech-founder angels.
An open-source transformer retrained on transaction data reads like a human. Heron commits to an LLM-first approach.
The team proves the model and the margins before adding headcount - lean to the point of stubborn.
Heron expands into insurance and specialty finance, scaling document automation across new workflows.
Insight Partners leads, with Y Combinator, BoxGroup and Flex Capital following on. Fuel for the next phase.
Heron describes what it builds as AI colleagues - software that takes a specific, tedious task and just does it. The platform is a pipeline: intake, parse, extract, enrich, check, and relay. Documents go in messy; structured, verified data comes out and lands in the customer's system of record.
Pulls documents from email and portals and reads 50+ types - statements, tax forms, applications and more.
LLMs turn unstructured pages into structured fields, enriched with context for underwriting.
Runs compliance, fraud and consistency checks, flagging issues before a human ever looks.
Syncs clean data directly into the customer's CRM and workflows. No retyping required.
Heron's case isn't a demo - it's throughput. The platform processes more than 350,000 documents a week for over 150 customers, among them FDIC-backed lenders and insurers. The work it removes is measured not in features but in hours that humans get back.
Bars scaled for comparison, not to a single axis. The point of the last one is how short it is.
The capital reflects the traction. A $1.2M pre-seed in 2020 grew into a $16M Series A in July 2025, led by Insight Partners with Y Combinator, BoxGroup and Flex Capital following their earlier checks. Revenue had tripled the year before. Customers like FundThrough run on the pipeline.
“350,000 documents a week is a lot of Mondays nobody has to dread.”
Heron's stated ambition is bigger than parsing PDFs. The founders believe a large share of repetitive white-collar work will simply vanish over the next decade, and they want to be the ones who retire it - in financial services first. The framing is deliberate: not "replace people," but free them from the part of the job no one chose.
There's an equity argument folded in, too. Faster, more consistent document handling means decisions - who gets a loan, how a claim is settled - run on the same data, checked the same way, every time. More efficient, more accurate, and arguably more fair.
The frontier of AI tends to land first where it's flashiest. Heron is aiming it somewhere less photogenic and far more useful: the back offices of the lenders, financiers and insurers that keep small businesses running. That's the expansion the Series A is meant to fund - more segments, more document types, more hours returned.
Back to that Monday morning. The inbox is still full - documents don't stop arriving. But the pile no longer sits and waits for a human to dread it. It gets opened, read and routed before anyone reaches for the coffee. The underwriter still decides. They just start the week ahead of it, for once.