Invoices and receipts
Supplier, date, lines, VAT and totals, checked against the purchase order.
Document processing that knows when to ask
If someone in your business retypes documents into a system, AI can probably do most of it. We build document processing that reads the page, pulls out the fields, checks them against what you already know, and hands the doubtful ones to a person.
What it reads
Delivery notes and bills of lading
References, consignee, quantities, matched to the right customer and shipment.
Statements, forms and applications
The fields you need, with the rest kept searchable.
Scanned filings and accounts
Figures read from scans and checked against the totals on the page.
Emails with attachments
The attachment is read, the email is filed and the right person is told.
Before and after
Before
- A document arrives by email, post or scan
- Someone opens it, reads it and retypes it
- Mistakes surface weeks later
- The paper is filed and hard to find again
After
- The document arrives by email inbox, copier scan, phone photo or upload
- AI reads it and extracts each field with a confidence score
- Fields are checked against your records: the customer exists, the order matches, the totals add up
- Confident results are filed straight into your system; unsure ones go to a person on one review screen
A person checks - The original is kept untouched, searchable and linked to the record
Where it runs
Tier 1
A hosted model, through a UK or EU endpoint
The best quality for the least effort. Before any of your data goes to a provider, we check that its business terms rule out training on it, and we tell you which region does the processing.
Tier 2
Inside your own cloud account
The model runs in your AWS or Azure account, under your access controls and your logs.
Tier 3
A private model on hardware in the UK
Open-weight models on your hardware or ours. Nothing goes to an outside AI provider. Our own research and document reading run this way.
Plenty of documents don’t need a large language model at all. Proven text recognition plus simple rules is cheaper and more predictable. We use the simplest approach that meets your pass mark.
Storage location and processing location aren’t the same thing. We’ll tell you both, in writing.
How we know it works
Tested on your documents
We build a test set from your real documents and measure accuracy field by field.
Thresholds you set
You choose how sure the system must be before it files something without review.
The sums have to add up
Lines are checked against totals, and references against your records.
Originals stay untouched
We never alter the evidence. What we extract sits alongside it.
Honest numbers
We don’t turn a handful of examples into a percentage. Small tests are reported as counts.
Document capture at the loading dock
Our document-capture product for warehouses reads phone photos and copier scans every minute and matches each bill of lading to the right customer. It splits a multi-document scan only when every page break is clear; otherwise the whole scan goes to a person.
What it costs
-
Pilot on one document type
From £6,000
Up to six weeks on one job, with the test set and pass mark agreed up front
Running cost: estimated per 1,000 documents before you start, and reported monthly after launch.
Prices exclude VAT where VAT applies.
Questions
Can it read handwriting?
Printed and typed documents work well. Handwriting is less reliable, so we test it on your examples before promising anything.
Which systems can it feed?
Anything with an API or an import: accounting, ERP, warehouse and CRM systems, or a spreadsheet if that’s where you are today.
Is this generative AI?
Sometimes. Often the best tool is text recognition plus rules, with a language model only for the messy cases. We pick per document type and tell you which.
What happens to our documents?
They stay where we agree, usually in your own storage. The original is never changed. We don’t use them to train anything.
Get in touch
Send us three sample documents and we’ll tell you what’s possible.
One paragraph is enough. We’ll reply with honest thoughts on whether we can help, and what it would cost.