No template required
Rule-based extraction breaks the moment a supplier changes their layout. We extract by understanding the document, so a new format is handled rather than rejected, and adding a supplier is not a project.
Documents
The documents somebody still types in by hand
Invoices, purchase orders, claims forms, onboarding packs, contracts. Somebody opens each one, reads it, and retypes the numbers into a system that will never know where they came from. This is the most reliably automatable work in most businesses, and the category older automation could never touch.
From$7,500
One document type extracted, validated against your systems and posted, with a human review queue.
Indicative starting price. The fixed fee for your scope is quoted after the two-day diagnosis, before any build begins.
What makes it work
Three things we insist on
Validated before it lands
Extracted values are checked against your own systems - does this PO exist, does the total match the lines, is this supplier on file - before anything is written. Wrong data caught at the door is cheap; wrong data in your ledger is not.
The review queue is a real place
Anything below the confidence threshold goes to a person with the document and the extracted fields side by side. Corrections feed back into evaluation, so the exception rate falls over time instead of quietly plateauing.
Today Person
- 1 Open the document
- 2 Read and interpret it
- 3 Retype into the system
- 4 Check it against the order
- 5 Chase whatever does not match
After Automated Human review
- 1 Read and extract
- 2 Validate against your systems
- 3 Post the clean ones
- 4 Queue the ambiguous ones
Capabilities
What is actually included
- 01
Extraction across document types
Invoices, POs, receipts, contracts, claims, IDs, application forms and correspondence.
- 02
Poor-quality input handling
Scans, photos, faxes and handwriting, with quality flagged rather than guessed at.
- 03
Business-rule validation
Cross-checks against your ERP, finance system or database before posting.
- 04
Straight-through posting
Clean documents written directly into the system of record, with a full audit trail per field.
- 05
Throughput reporting
Volume, touch rate, exception reasons and cycle time, so you can see what automation actually bought.
In detail
What this covers, specifically
A category name is not a scope. These are the individual pieces of work inside this practice - take the two that apply to you and ignore the rest.
- 01
Invoice and AP automation
Extracting supplier, dates, line items, tax and totals from any invoice layout and posting them, without a per-supplier template.
- 02
Purchase order matching
Two-way and three-way matching between order, receipt and invoice, with discrepancies explained rather than just flagged.
- 03
Contract abstraction
Pulling parties, term, renewal dates, notice periods, liability caps and payment terms into a structured register you can query.
- 04
Claims processing
Reading claim forms and supporting evidence, checking completeness and consistency, and routing by complexity.
- 05
Application and form intake
Turning submitted forms, onboarding packs and applications into validated records in the system of record.
- 06
Identity and KYC documents
Reading passports, licences and proof of address, checking expiry and consistency, and flagging anything that needs human eyes.
- 07
Poor-quality input handling
Scans, phone photos, faxes and handwriting, with confidence reported per field rather than a single score for the page.
- 08
Classification and routing
Working out what a document is before anything else happens, and sending it to the right process automatically.
- 09
Validation against systems
Cross-checking extracted values against your ERP or database before posting, because wrong data caught at the door is far cheaper than wrong data in the ledger.
- 10
Human review queue
An interface showing the document and the extracted fields side by side, where corrections feed back into evaluation.
What you receive
Concrete artefacts, not a slide deck
- Document taxonomy with volumes and current handling cost
- Extraction and validation pipeline in production
- Human review queue and interface
- Field-level audit trail and compliance documentation
- Throughput and exception dashboard
01 Diagnose Days 1-2
02 Prove Week 1
03 Integrate Weeks 2-3
04 Operate Ongoing
Around three weeks end to end. That comes from scoping tightly to one workflow - not from skipping a phase. Each still ends in evidence you can check.
Further reading
Written up in more detail
- Context windows and why long documents still fail A large context window is not a document strategy. Why retrieval accuracy drops in the middle of long inputs, and what to do with a 200-page contract instead.
- Document extraction: what to verify before you trust it Field-level accuracy hides which field fails. The validation layer, confidence routing and reconciliation checks that make extracted data safe to post.
Let's talk
Send us two hundred real documents
Messy ones. We will run extraction against them and show you the field-level accuracy, which document types are reliable today, and which need a human in the loop - before you commit to a build.
Related practices