We tried OCR years ago and it was useless. What is different?

The old tools matched templates, so a new supplier or a changed layout broke them. What we build reads the document the way a person does, so a layout it has never seen is the normal case, and phone photos of paper and neat handwriting read too. We test it on your real intake in the first week, not on a sample deck.

Accurate enough that the job becomes checking exceptions rather than typing. Every extracted field carries a confidence score, anything under the threshold you set is routed to a person with that field highlighted, and on anything financial you can set the threshold as strict as you like. It is a setting, so it moves as your confidence does.

Where does our data go?

Where you decide, and it is written down before a single document is sent anywhere. For most business paperwork a mainstream model behind a data processing agreement is the right call. Where documents carry special category data or a client contract forbids it, the build changes to keep them inside your own environment, and that is agreed before anything moves.

A fixed price, quoted in writing before anything starts, whatever the job turns out to involve. It runs in your own accounts and under your own name, so the systems and the data stay yours.

What do you need from us?

The brief, the brand and the access: what arrives, where each field should land, and logins to the inbox, folders and accounting system it feeds. Everything after that is us.

Document processing automation for invoices, forms and PDFs

Every invoice, purchase order, form and PDF read the moment it arrives, checked against your own records and filed where it belongs. Nobody retypes anything. You bring the brief and the access. Everything after that is ours.

What happens to every document

Six stages, all of them ours. You approve the field map once, then it runs.

It arrivesEmail attachment, shared folder, scanner or supplier portal. Every source feeds the same line.
It is readAny layout, any supplier, including phone photos of paper taken at an angle. Nothing is templated.
The fields come outSupplier, dates, line items, totals, references, terms. Whatever your process needs, mapped once.
It is checkedAgainst your own records: supplier known, PO matched, totals adding up, invoice number never seen before.
It is filedInto Xero, QuickBooks, Sage, your CRM or the job record, with the original attached to the entry it created.
The doubtful ones are flaggedAnything under your confidence threshold lands with one person, the exact field named. Checking, not typing.

The checks that pay for the build

Duplicate invoices caught before they are paid.

Duplicate invoicesThe same invoice number, or the same amount from the same supplier, stopped before it is paid twice.
Totals that add upLine items, VAT and the header total reconciled on every document.
Matched to the PO and the supplierKnown supplier, open purchase order, quantities and prices as agreed.
Changed bank details flaggedNew account details on a familiar supplier go to a person before anything is updated.

Every check runs against your own ledger, purchase orders and supplier list, in your own accounts. A fixed price, quoted in writing before anything starts.

Book twenty minutes →Send us the brief

Where it reads from, and where it files to

Claude does the reading. Everything else is your own inbox, folders and ledger, connected in your accounts and your name, so nothing of ours needs switching off if we ever part ways.

Claude
Gmail
Microsoft 365
Google Drive
Dropbox
Xero
QuickBooks
Sage
HubSpot
Sheets
Airtable
n8n

The original document stays attached to the record it produced, so a year from now the number and its source are still together.

What happens once the invoice has been read →

The day-to-day, all of it automated

Fourteen jobs a specialist does by hand. The run does every one, on a schedule, inside the tools you already use. You see the results, not the work.

Dext, illustrated with example figures: Dext: Costs inbox showing processed supplier invoices with extracted amounts and Ready to publish status. — Dext: Costs inbox showing processed supplier invoices with extracted amounts and Ready to publish status.

Incoming supplier invoices extracted and checked automatically
Purchase orders matched against line items on bills
Paper documents scanned and filed into correct folders
Delivery notes checked against stock and invoice quantities
Customer application forms read and checked for completeness
Bank statements imported and converted into clean data
Key details validated against existing client database records
Missing fields and duplicate submissions flagged for review
Tax rates and company details verified on receipts
Email attachments processed and sent to relevant teams
Historical paper files digitised for instant search
Approval workflows triggered automatically based on invoice amounts
Expense receipts checked against company expense policy
Final documents archived securely with full audit histories

Xero, illustrated with example figures: Xero: Bills to pay list with extracted supplier details, due dates and Awaiting approval status. — Xero: Bills to pay list with extracted supplier details, due dates and Awaiting approval status.

Google Drive, illustrated with example figures: Google Drive: organized folder tree showing supplier invoices named automatically by date and vendor. — Google Drive: organized folder tree showing supplier invoices named automatically by date and vendor.

Illustrations of the tools in use, drawn for this page for an example business, Marlow & Finch. Every figure in them is an example, not a client's data, and any person in them is generated. Tap a picture to open it full size.

DextInvoices and receipts read
XeroBills created and matched
QuickBooksSame job, QuickBooks shops
Google DriveFiled and named, every time
DocuSignSigned documents in the flow

The night shift

Nobody in the building. This is the job doing itself at eleven at night, and what you see in the morning.

Every sheet read on the glass, filed in the tray, posted to the ledger. The pile never grows.

Questions we get asked about this

We tried OCR years ago and it was useless. What is different?: The old tools matched templates, so a new supplier or a changed layout broke them. What we build reads the document the way a person does, so a layout it has never seen is the normal case, and phone photos of paper and neat handwriting read too. We test it on your real intake in the first week, not on a sample deck.
How accurate is it?: Accurate enough that the job becomes checking exceptions rather than typing. Every extracted field carries a confidence score, anything under the threshold you set is routed to a person with that field highlighted, and on anything financial you can set the threshold as strict as you like. It is a setting, so it moves as your confidence does.
Where does our data go?: Where you decide, and it is written down before a single document is sent anywhere. For most business paperwork a mainstream model behind a data processing agreement is the right call. Where documents carry special category data or a client contract forbids it, the build changes to keep them inside your own environment, and that is agreed before anything moves.
What does it cost?: A fixed price, quoted in writing before anything starts, whatever the job turns out to involve. It runs in your own accounts and under your own name, so the systems and the data stay yours.
What do you need from us?: The brief, the brand and the access: what arrives, where each field should land, and logins to the inbox, folders and accounting system it feeds. Everything after that is us.

Got a question the list did not answer? Ask it on a call, or send us the brief and we will come back with a fixed price.

Book twenty minutes →Send us the brief