We build your document processing and connect it to what you already use.

Send us a few sample invoices, contracts or IDs.

We build the extraction, usually within one business day. It costs you nothing until it runs on your real documents.

Catalin Crisan, founderTwelve years building data pipelines. He runs every build himself.

supplier "Meridian Supply SRL"number "MS-2026-0416"vat_id "RO99014288"date "2026-07-27"due "2026-08-26"currency "EUR"total 4182.60pages 2

Running in production at

Blitz
USMLEX
Cicada Technologies
Tap Point
MakeIT Future
Solo
Premium Advisors

Around 10,000 documents a month, across all of them.

FIG. 2Before you pay anything

The build is on us.

All of it happens before any money changes hands, and almost none of it needs your time.

Invoicesupplier, number, VAT, dates, total
Contractparties, term, value, effective date
IDname, document number, date of birth

01
We talk for 30 minutes
What documents you handle, what you need out of them, and where that data has to end up. If it is not a fit, we will say so on the call.
30 minutes
02
You send a few representative documents
Whatever actually comes in. We need to see the scans that arrive crooked and the supplier who formats their invoices differently to everyone else.
We sign an NDA if you want one, and redacted samples are fine. Your documents never train anything, and we do not keep them.
03
We build it
The extraction, the output structure, the validation, and the connection into Make, your ERP or your own API. That part is ours to do.
Usually one business day
04
You look at the output
Your documents, your fields, your team checking whether the values are right. That is the only part that asks anything of you, and it is reading results.
Zero hours of setup on your side
05
It goes live, and only then does it cost anything
You pay per document processed. Nothing before that point, and nothing else after it.

Why we do not charge for the build.

We get paid per document. If what we build never ends up processing anything, we have spent a day for nothing, and that is our problem.

It is also why we turn work down. If your volume is small, the metered revenue will not cover the build, and we would rather tell you that on the call than take the project and hope.

Most tools in this space charge you a subscription first and let you discover afterwards whether they cope with your documents. We would rather find out first, at our expense.

Before Fabrx, twelve years of data engineering.

Pipelines: ETL and ELT frameworks at ING, Rabobank, APG and Paddy Power Betfair, moving data from source systems into warehouses people report off.
Scale: Redshift warehouses past 10TB, and ingestion running at thousands of records a second.
Regulated: IFRS9, AnaCredit and COREP reporting, the kind banks get audited on, plus EU AI Act Annex IV documentation.

Connecting an extraction to your ERP, your warehouse or your database is the routine half of this work.

FIG. 3Four years in production

Blitz turned a two-minute contract into a ten-second one.

Over 300 field agents scan an ID and a completed contract comes back, inside Blitz's own application.

Blitz, field sales

We built ID extraction into their existing contract generation app. It has been running there since 2022.

Time to complete one contract

By handabout 2 minutes

With Fabrx10 seconds

300+: Field agents using it every working day
4,000+: Documents processed every month
4 yrs: In production, on the same workflow

We have been using Fabrx.ai for over 4 years to automatically extract data from identity documents within an internal contract generation application used by more than 300 users. The implementation was straightforward, and the extracted data has been consistently accurate.
Rebeca Dogaru, Business Analyst, Blitz

It reduces the time spent filling out contracts from approximately 2 minutes to 10 seconds. It's a secure way of entering data because it eliminates human error associated with manual input.
Bogdan Lucaciu, Regional BD Manager, Blitz

Read the full story

FIG. 4What customers say

Most of it is about how we work.

Responsive

What really sets them apart is their communication. It's rare to find a partner that is both technically capable and so personally invested in their clients' success. If you need efficient document processing and a team that actually picks up the phone, I highly recommend Fabrx.ai.

Craig Sullivan
Founder, USMLEX

90% time saved

The system has proven to be reliable and efficient, which has helped us automate processes that previously consumed a lot of time. Cătălin is flexible, responds quickly, and finds practical solutions for any situation encountered along the way. For us, this type of partnership matters just as much as the technology itself.

Bogdan Rus
CEO, Cicada Technologies

It lands where your data already lives.

You do not adopt a new tool or move your team into somebody else's dashboard. The extraction runs behind whatever you already use.

The same path, whatever the page looks like

Pick a document type. The request in the middle is the same one every time. What changes is the page it reads and the record it writes.

A supplier invoice, the Fabrx extraction request and its response, and the vendor bill the values land in inside Oracle NetSuite. — Supplier invoice/v1/extractNetSuite vendor bill

A supplier packing slip, the Fabrx extraction request and its response, and the goods receipt the values land in inside a manufacturing ERP. — Packing slip/v1/extractERP goods receipt

A bill of lading, the Fabrx extraction request and its response, and the shipment record the values land in inside a freight management system. — Bill of lading/v1/extractShipment record

A motor insurance claim form, the Fabrx extraction request and its response, and the claim record the values land in inside a claims management system. — Auto claim form/v1/extractClaim record

A master supply agreement, the Fabrx extraction request and its response, and the matter record the parties, dates and terms land in inside a legal workspace. — Supply agreement/v1/extractMatter record

A commercial lease agreement, the Fabrx extraction request and its response, and the lease record the tenant, dates and rent land in inside a property management system. — Commercial lease/v1/extractLease record

A patient referral form, the Fabrx extraction request and its response, and the referral record the patient, provider and diagnosis code land in inside a care platform. — Patient referral/v1/extractReferral record

The documents and the destination screens are samples. Each one shows the kind of system the fields land in, not a named customer.

Documents arrive from
- An email inbox, including Gmail
- A shared drive
- An upload in your own interface
- A scanner or a phone camera
Fabrx does
- Classification
- Field extraction
- Validation
Data lands in
- Make
- Your own app, over our API
- Your ERP

We build the API. It lands structured fields in any system that accepts an API call, so a new destination is one more endpoint.

What comes back

Structured fields, same shape every time
A webhook on every result
An exception when a value cannot be read

POSThttps://your-app.example/hooks/fabrx

{  "event": "document.processed",  "type": "invoice",  "fields": {    "supplier": "Meridian Supply SRL",    "number": "MS-2026-0416",    "vat_id": "RO99014288",    "date": "2026-07-27",    "due": "2026-08-26",    "currency": "EUR",    "total": 4182.60,    "pages": 2,  },  "exceptions": []}

Your documents stay yours.

We do not store them: Documents pass through and are gone. Fragments are kept only if you switch on data lineage, and that is your choice to make.
Never used for training: Your documents and the data we pull out of them never train a model, ours or anybody else's.
GDPR: PII detection and audit trails are included in the service.
Google and Gmail access: If you ask us to collect documents from your Gmail inbox, Fabrx requests read-only access (the gmail.readonly scope) and uses it only to find and import the messages and attachments you select for processing. Gmail data is never used for advertising and never trains a model, and you can revoke access at any time from your Google Account. The privacy policy sets out the full terms.

You buy documents, not seats.

A prepaid package of documents, valid for a calendar year, drawn down as you process. We send your rate after the call, once we have seen what your documents take, because before that we would only be guessing.

What you pay for

Documents processed, counted one at a time
A prepaid package, valid twelve months
A better rate as your volume grows

What you never pay for

The build, the integration or the tuning
A monthly subscription or platform fee
Per-seat licences, however many people use it
Compliance, audit trails or observability

Get a rate by email

Three answers are enough to price your documents. Tell us what you process and we will send back a rate, or the one question we need answered before we can give you one.

The small print, in plain words

How it ends.

What we own: The extraction logic and the schema we build stay ours. You have access to them for as long as you are a customer.
What it costs later: A package holds its price for the calendar year. We rarely change rates, and when we have to you get at least a month's notice first.
If you leave: One month's notice. There is nothing of yours for us to hand back, because your documents were never ours to keep.

The ones people ask.

What's the catch?

There is a condition. We only take on work where the volume makes the per-document revenue worth a day of building. If your volume is low we will tell you on the call, instead of building something that does neither of us any good.

What does it cost?

A prepaid package priced per document, valid for a calendar year. The rate depends on the document type and your monthly volume, so we send it after the call. There is no subscription, no per-seat charge and no setup fee at any volume.

How long until it is running?

Usually one business day from the moment we have your sample documents. Setting up the API takes none of your team's time, because we do all of it. The only thing we need from you afterwards is somebody to look at the output and tell us whether it is right.

What if the accuracy is not good enough?

Then you do not pay, because nothing went into production. We either keep working on it or we tell you plainly that your documents are a poor fit for this. Either way it has cost you nothing but the call.

Do we have to move our team into your platform?

No. The extraction runs behind the systems you already use. At Blitz, the 300 people using it every day work inside their own contract application and never see anything of ours.

What happens if Catalin is unavailable?

Once it is built, it runs on its own. The extraction at Blitz has been processing over 4,000 documents a month since 2022 without anyone touching it, through four years of holidays and busy weeks. Catalin is personally involved in the build and in anything that changes afterwards.

Is it safe to send you our documents?

We do not store them and they never train anything. Fragments are kept only if you turn on data lineage for auditability, and those are encrypted. We will sign an NDA if you want one, and redacted samples work fine for the build.

Tell us which documents are eating your week. We will tell you on the call whether we can fix it.

You will be talking to CatalinEvery call goes straight to the founder.

Twelve years of data engineering