Loading demo…
Document intelligenceComing Soon
Document classification API
Send any document image and get back the document type, confidence score, and suggested next step. Supports 50+ categories — from passports and invoices to tax forms and utility bills.
- 99.4% classification accuracy
- ~100ms median latency
- 50+ document categories
- Chains with Document AI for end-to-end parsing
Availability
Document classification is in private beta
The category set usually needs mapping onto your own routing logic before it is useful, which is a conversation rather than an upload button. Early access includes API keys and help doing that mapping.
No public demo yet
We run early access as a design-partner programme — you get API keys and we get feedback from a real integration before general release.
How it works
How document classification works
Classification is the cheapest question you can ask about a document, and asking it first makes every later step better informed.
- 1
The document arrives
A scan, a PDF page, or a phone photo — whatever your intake channel produces, with no guarantee it is the document the user was asked for.
- 2
Layout is assessed
Visual structure carries most of the signal. A passport data page, an invoice, and a bank statement look distinct long before you read a word of any of them.
- 3
A category and confidence come back
One label from 50+ categories with a confidence score, in around 100ms — a fraction of what full extraction costs.
- 4
Your pipeline routes on it
High confidence flows into the right extraction path automatically; low confidence goes to review before anything expensive or irreversible happens.
Coverage
What it recognises
Over 50 categories across four broad families. Each result carries a confidence score, and genuinely unfamiliar documents come back uncertain rather than forced into the nearest label.
| Family | Examples |
|---|---|
| Identity | Passport, national ID card, driving licence, residence permit |
| Financial | Bank statement, payslip, tax form, invoice, receipt |
| Proof of address | Utility bill, council tax notice, tenancy agreement |
| Business | Contract, purchase order, delivery note, certificate of incorporation |
Pipeline
Where it sits in a document pipeline
Classification is a gate, not a destination. Its value is what it lets you avoid doing.
1. Classify
Identify the document type in ~100ms and reject anything that clearly should not be here.
2. Extract
Parse fields appropriate to the now-known type, instead of guessing what the document might contain.
See the product3. Verify
Apply the checks that type demands — for identity documents, comparing the portrait against a live selfie.
See the productUse cases
What teams use classification for
KYC intake triage
Confirm the user uploaded a passport and not a selfie or a utility bill, before it reaches the extraction and verification stages.
Mailroom automation
Sort inbound scanned post into the right department queue without a person opening each item to see what it is.
Pipeline cost control
Filter out irrelevant uploads with a cheap call so full extraction only runs on documents that warrant it.
Archive organisation
Label an unstructured document store retroactively so it can be searched, retained, and disposed of by type.
FAQ
Document classification FAQ
- What is document classification?
- Classification identifies what kind of document you have received — passport, invoice, bank statement, utility bill — before anything tries to read it in detail. It is the triage step that decides where a document goes next.
- Why classify before extracting?
- Because extraction, validation, and routing all depend on the document type. Knowing a file is a passport rather than a driving licence tells you which fields to expect, which checks apply, and which queue it belongs in. Classifying first also catches the wrong-document-uploaded case early, which is one of the most common failure modes in any intake flow.
- How many categories are supported?
- Over 50, spanning identity documents, financial documents, and general business paperwork. The output is a category label with a confidence score rather than a bare guess.
- What happens with an unrecognised document?
- It comes back with low confidence across all categories rather than being forced into the nearest label. That is the signal to route it to manual review — a confident wrong classification is far more expensive than an honest uncertain one.
- How does it chain with Document AI?
- Classification runs first and decides the type, then Document AI extracts fields appropriate to that type. Running them in that order means the extraction step knows what it is looking at, and you avoid paying for full parsing on documents that turn out to be irrelevant.
- Does it read the whole document?
- It does not need to. Classification works from layout and visual structure alongside a light read of the text, which is why it returns in around 100ms — roughly a third of the cost of full extraction. That gap is the whole reason to run it as a separate gate.
- Can it handle photos rather than scans?
- Yes. Phone photos of documents classify reliably as long as the whole document is in frame. Since classification depends on overall layout, a partial crop that cuts off the document structure is harder than a slightly blurry full-page shot.
- When will the API be available?
- Document classification is in private beta. Early access includes API keys and support for mapping the category set onto your own routing logic. Contact us to be considered.
Related: Document AI, OCR, and everything that is live today.
Ready to ship identity verification?
Talk to our engineering team about your use case. We'll get you up and running in under a day.
SOC 2 Type II · GDPR Ready · ISO 27001