Invofox · YC S22
ONE API to extract data from any document 17M+ invoices processed a year 100+ software companies as customers 99.2% extraction accuracy Motto: Process. Deliver. Erase. Founded by 3 Madrid math grads HQ moved Madrid → San Francisco Y Combinator Summer 2022 ONE API to extract data from any document 17M+ invoices processed a year 100+ software companies as customers 99.2% extraction accuracy Motto: Process. Deliver. Erase. Founded by 3 Madrid math grads HQ moved Madrid → San Francisco Y Combinator Summer 2022
Company Profile · Fintech Infrastructure

Invofox teaches machines to read the paperwork nobody wants to

The YC S22 startup does one unglamorous thing: turn any invoice, receipt or shipping document into clean, structured data through a single API. Three friends who studied mathematics together built it - and 100+ software companies now quietly run on it.

Somewhere in the world right now, a person is squinting at a supplier invoice, typing a number into a box, and then typing the next number into the next box. It is one of the largest jobs on earth and almost nobody lists it on a resume. Invofox, a company that came out of Y Combinator's Summer 2022 batch, decided that job should not exist - and built an API to make it disappear.

The pitch fits on a business card: one API to extract data from any document. You send Invofox a PDF or a photo - an invoice, a receipt, a payslip, a bill of lading - and you get back clean, structured JSON that software can act on. No re-typing. No copy-paste. Often, no human at all.

What is interesting about Invofox is not that it reads documents. Plenty of tools claim that. It is who Invofox sells to, and what it refuses to do.

01 / WHAT IT ACTUALLY DOESThe plumbing, not the faucet

Invofox does not sell to accountants. It sells to the software companies that accountants use. That distinction is the whole strategy. When a fintech app pays a bill automatically, when an accounting platform files a receipt, when a lender reads a mortgage statement - Invofox can be the layer underneath, reading the document so the product on top never has to think about it.

The founders describe the company as document-processing infrastructure. In plain terms: it is the plumbing. You do not admire plumbing. You just notice, sharply, when it fails. Invofox's entire product is the promise that it will not.

STEP 01
Upload
A PDF or image arrives - invoice, receipt, shipping doc - one file or a batch.
STEP 02
Classify & Split
The engine detects document type and splits multi-doc uploads apart.
STEP 03
Extract
Dual-pass OCR + ML rebuild tables, normalize entities, score confidence.
STEP 04
Verify
Rules and human reviewers catch and fix the shaky fields.
STEP 05
Deliver & Erase
Clean JSON goes out by webhook - then the file is deleted.
The assembly line for chaos. A messy document goes in one end; structured data leaves the other. The last step is the unusual one - the file gets thrown away on purpose.

That last step is the part worth pausing on. Invofox's motto is three words: Process. Deliver. Erase. After it hands back your data, it deletes the document. In an industry that mostly hoards data, choosing to forget became a feature - one that regulated customers in finance and healthcare will pay for.

One API to extract data from any document. — Invofox, company tagline

02 / THE FORTY-YEAR PROBLEMWhy OCR never finished the job

Optical character recognition has been around since roughly the 1980s. The promise was always the same: point a scanner at a page and let the paperwork vanish. It never quite happened. OCR could tell you the shapes on a page were the characters "1,240.00." It could not tell you that number was the total, that the currency was euros, or that the line item above it was mislabeled.

Reading characters and understanding a document turn out to be very different problems. The gap between them is where every accounts-payable clerk has lived for four decades. Invofox's founders - Alberto Gimeno, Carmelo Juanes and Adrian Ramirez - bet that modern machine learning could finally close it: not just recognizing text, but classifying the document, validating fields, filling in what is missing, and flagging what looks wrong.

17M+
Invoices processed / year
99.2%
Stated accuracy
100+
Software-company customers
7+
Languages supported

There is a tidy incentive built into how Invofox charges. Pricing is per page, and errors cost the customer nothing. When your revenue depends on being right, "mostly accurate" is not a marketing line - it is a bill you eat. Few document companies publish an error rate at all. Invofox prices around it.

03 / THE HUMANS IN THE MACHINEWhere the robots hand off

The unfashionable truth about document AI is that pure automation gets you to "mostly right," and mostly right is useless when the number is going to move money. Invofox wires people into the pipeline exactly where the models get shaky - a design its founders call human-in-the-loop. Rules check the fields, humans review the doubtful ones, and only then does the data flow downstream where automation can run unattended.

It is a quiet answer to the loudest debate in the field. Invofox's own blog has asked whether the future of accounting is human insight or automation. Their product is the boring, correct answer: both, arranged so each does the part it is good at.

Thinking bigger to achieve more ambitious goals, maintaining focus by going deeper rather than broader, and making complex concepts simple. — Alberto Gimeno, Co-Founder & CEO

04 / WHO PAYS FOR ITThe customers you never see

Because Invofox lives underneath other products, its customers are companies, not people. Fintechs and accounting platforms use it to automate bookkeeping and payments. Lenders and mortgage firms use it to read financial statements. Insurers, marketplaces and logistics operators feed it the paperwork of their trades. A shipping company's bill of lading is, from a machine's point of view, a uniquely messy document - and messy documents are exactly the ones Invofox wants.

Fintech & Accounting

Automated books

Invoices and receipts become ledger entries without a human keying them in.

Lending & Insurance

Faster decisions

Financial statements and forms parsed into data underwriters can act on.

Logistics

Paper that ships

Bills of lading and shipping docs turned into structured records across borders.

Platforms & ERPs

Embedded & white-label

Software products drop Invofox in under their own brand and never build a parser.

05 / THE MARKET AROUND ITNarrow, on purpose

The category has a clunky name - intelligent document processing, or IDP - and no shortage of players. Legacy OCR vendors sit at one end. Cloud giants like AWS Textract and Google Document AI sit at the other. In between are focused startups such as Rossum, Mindee, Nanonets and Docsumo. It is a crowded room.

Invofox's answer to the crowd is to go narrow. Not an all-in-one platform with a hundred features; one thing - document to clean JSON - done deep, sold as infrastructure to builders rather than as an app to end users. In a market that keeps adding surface area, choosing to add depth is the contrarian move.

Where Invofox sits in the document stack
Raw OCR
reads text
Cloud APIs
text + fields
Invofox
understand + verify + erase
Your app
acts on clean data
Depth beats breadth. Most tools stop at reading. Invofox pushes further down the stack - understanding, human verification, then deletion - so the product on top just receives trustworthy data. Illustrative positioning, not a benchmark.

06 / THE PEOPLE AND THE MOVEFrom a Madrid classroom to San Francisco

The three founders met studying mathematics and computer science in Madrid. Gimeno, the CEO, carries dual degrees in mathematics and computer science and had already tried his hand at a book-rental service and a hardware-security company before Invofox. Juanes runs technology as CTO; Ramirez leads the machine-learning work. The team is small and heavily technical, the kind of shop where most of the roster could argue about model architecture.

In 2024 Gimeno moved the company to San Francisco, keeping the Madrid operation running. The relocation reportedly cost more than $30,000 in immigration legal fees - a price paid not for the weather but for proximity. His reasoning is refreshingly unromantic about it.

You meet your heroes. You learn how the best founders think. And you build a startup in the most intense, inspiring environment. — Alberto Gimeno, on moving Invofox to San Francisco

By public accounts the bet has been paying off: press interviews describe Invofox serving more than 100 software companies, processing tens of millions of documents a year, and posting several million dollars in annual recurring revenue with strong month-over-month growth. Those figures come from the company and its coverage, so treat them as the founder's scoreboard rather than an audited one - but the direction is clear.

The larger idea underneath Invofox is almost philosophical. Every business runs on documents, and documents are where information goes to get stuck. A shipping manifest, a utility bill, a payroll slip - each is a small dam holding back data that some system downstream wants. Invofox is in the business of opening those dams, quietly, one page at a time, and then walking away without keeping the water.

#DocumentAI#IDP#OCR#Fintech #YCS22#APIfirst#InvoiceProcessing #StructuredData#HumanInTheLoop#Invofox