Hey everybody! 👋

We’re Alberto, Carmelo and Adrian, the founders of Invofox.

When we first launched on YC four years ago, Invofox extracted data from invoices.

Over these past few years we expanded to all types of documents, grew to serve +150 software teams across Europa and the US, and processed tens of millions of documents for them.

Today we're excited to launch something new: a fully self-serve API to turn virtually any business document into structured, validated data.

https://www.youtube.com/watch?v=cuvpexh-WTw

TL;DR

Send Invofox a PDF or image. We handle the document processing pipeline and return clean, structured JSON.

Mortgage documents, Invoices, bank statements, payslips, receipts, bills of lading, insurance documents, healthcare documents, or your own custom document type.

And starting today, anyone can sign up and try the entire product without talking to us first.

Upload your own documents, see the results, define what you want to extract, and integrate the API.

For the launch, new users get 1,500 pages free.

The problem

LLMs have made extracting information from a document look deceptively easy.

Upload a PDF. Write a prompt. Get some JSON back.

The hard part starts when you need to do this reliably in production across thousands or millions of real-world documents.

Documents arrive rotated, scanned badly, bundled together, in different languages and layouts. You need to split them, classify them, extract nested fields and line items, validate the output, deal with edge cases, measure accuracy, handle failures and continuously improve the system.

And one wrong field can be enough to break the workflow downstream.

We kept seeing software teams build increasingly complicated document-processing infrastructure around models that were supposed to eliminate that infrastructure in the first place.

We are extraction-first

There’s an important distinction in how we’ve built Invofox.

Many document-processing products started with parsing: OCR, converting PDFs into text, understanding layout, or making documents searchable. Data extraction then became a natural extension of that product.

We came at the problem from the opposite direction.

Our core problem has always been: how do we get the exact data a customer needs out of a document, correctly, every time?

Parsing is one component of that problem, but it isn’t the end product.

That difference has shaped almost everything about Invofox.

Our technology is optimized around extraction accuracy, not simply producing a good representation of the document.

Our validation layer is designed to catch extraction errors before they reach downstream systems.

Our pricing is built around successfully processed documents rather than tokens, model calls, or pages of parsed text — in fact, if you use Invofox, documents were we make any mistake are completely free.

And our operations are built around continuously finding and fixing the difficult edge cases that appear in production.

The goal is not to give you a cleaner PDF.

The goal is to give your software the right data.

Especially on the documents where extraction gets difficult.

What we built

Invofox gives developers one API for the entire document processing pipeline.

Send us a document and we handle:

• Parsing and OCR
• Splitting files containing multiple documents
• Document classification
• Structured data extraction (with support for custom schemas)
• Tables and line items
• Validation and confidence checks
• Production monitoring and continuous improvement
• Delivery through our API and webhooks

You can use one of our pre-built document types or define exactly what you want returned.

The output is structured JSON that can go directly into the rest of your application.

Send your documents to our API and don’t worry about the details: which models to use, how to prompt them, what OCR or parsing techniques will improve your results. You don’t really care about any of that: you just want to get the right data, reliably, at the right price.

Today, 150+ companies use Invofox across accounting, financial services, mortgage, insurance, healthcare, logistics and other document-heavy industries.

Why we obsess over accuracy

After processing tens of millions of real-world documents, we've learned that field-level accuracy alone is a misleading metric.

If you extract 50 fields from a document and get one important field wrong, that document may still require a human to review it.

So we measure whether the entire document is correct.

Our production accuracy is above 99%, we guarantee this level of quality as part of our SLAs, and we back that up commercially: if you report an incorrectly processed document, you don't pay for it.

We want the incentive to be simple: we make money when the automation actually works.

Why self-serve now?

Until today, trying Invofox generally meant talking to our team first.

That increasingly felt wrong.

We're infrastructure for software teams. The best way to evaluate infrastructure isn't to watch our sales deck or take our word for how accurate it is.

It's to throw your own data at it.

So now you can.

Create an account, upload the ugliest documents you have, inspect the output, and decide for yourself whether it's good enough to put into production.

No sales call required.

A little backstory

The three of us met studying mathematics and computer science in Spain.

We originally came across this problem while building software that needed to extract data from business documents. Like a lot of engineers today, our first reaction was essentially: surely this has already been solved.

It hadn't.

We joined YC in S22 with a product focused on invoice extraction.

Since then, customers kept bringing us adjacent problems: bank statements, mortgage documents, payslips, insurance documents, logistics paperwork, healthcare documents and many others.

The document types changed, but the problem was always the same: reliably turning messy, unstructured documents into data that software could actually use.

Over time, what started as an invoice extraction product became infrastructure for processing unstructured business documents in general.

This launch is the first time we're making that entire product available for anyone to try on their own.

Our ask 🙏

We'd love two things from the YC community:

1. Try Invofox on your hardest documents. We're giving new users 1,500 free pages for the launch. If you find something we can't handle well, I genuinely want to see it.
2. If your company is building a workflow that involves humans reading PDFs or images and typing information into software, we'd love to talk. There's a good chance that workflow can now be automated.

You can get started at invofox.com.

And if you try it, please send us feedback, especially the bad stuff!