Automatic document entry. They go in on their own, in the format you choose.
Orders, delivery notes, invoices from foreign suppliers, receipts. They arrive through different channels and in different formats, and someone retypes them by hand into the business software. This piece reads them, puts them into the format you need and enters them one by one, after a check.
They file themselves. The document arrives, is recognised, and the data you need is extracted and written where it belongs. Nobody has to reopen a PDF to copy a quantity.
You choose the format. The fields to extract, what they're called and the format they come out in are agreed once, and they're yours, not the incoming document's.
There's a check before the data goes in. Every field comes with a measure of how confident the system is that it read it correctly. Above the threshold it goes in; below it, it stops and shows you. Nothing goes in on a guess.
This page covers a single piece of the system. The other pieces, and how we choose which one to start with, are on the services page.
What changes in practice
Retyping disappears from the working day. An order that arrives at 9 is in the system at 9, with the lines, quantities and codes already in place, and the person who used to key it in now has a document already entered, to check in ten seconds.
The second change is less visible and matters more: all the data goes in the same way. Same naming, same codes, same units of measure, whichever supplier sent the document and however they wrote it.
The time you save isn't in reading the document. It's in never having to open it a second time.
You choose the format
During the analysis we decide what to extract, field by field, and what to call it. If your business software calls the field quantità ordinata, that's what the system writes, even when the incoming document says ordered qty.
The same goes for the output format. Where your business software can be connected to from outside, the data is written straight in; where it can't, the system prepares a file in exactly the layout that software imports, and the final click stays with you.
We extract a short list, agreed together, not everything on the document. Extracting everything produces fields that need cleaning up, and cleaning them up is the same manual work you wanted to get rid of.
What the system does, step by step
Every document comes in through the same point, whatever channel it arrived on, and every step leaves a log entry recording what happened and when.
| Step | What happens | What you get |
|---|---|---|
| Recognitionwhat kind of document it is | The document is classified by type before it's read, and documents that already arrive structured are taken as data, not as images. |
Orders, delivery notes and receipts each follow their own route, without anyone having to sort them. |
| Extractionfield by field | The agreed fields are extracted, with the names they have in your business software and in the format you need. |
The data arrives already in your own internal terms, ready to use without renaming anything. |
| Checkhow confident it is | Every field comes with a measure of how confident the system is that it read it correctly, and the fields that involve money have the highest threshold. |
You know which data is solid and which isn't, instead of finding out from a customer. |
| Stopbelow the threshold | An uncertain document doesn't go in: it's shown with the doubtful field highlighted next to the original, ready to correct. |
Ten seconds of checking on a handful of documents, instead of a week chasing a wrong order. |
| Entryinto the business software | Documents above the threshold are entered. Where the business software can't be connected to from outside, the system prepares the file to import. |
The business software stays yours, and nobody has to change it to make the rest work. |
The check, which is what makes the rest usable
A model that reads documents always gives an answer, even when it hasn't understood. Faced with an illegible quantity, it writes a plausible number, and that number looks exactly like one that was read correctly. That's why the check isn't an optional extra: without it, the system moves errors around instead of removing them.
So every field comes out with a confidence score, and the threshold is agreed together, field by field. An uncertain delivery address can be fixed later. An uncertain quantity, net amount or order number stops the whole document and sends it to a person.
Stopping is normal, and it's worth knowing that before you buy anything: in departments that already process documents automatically, the average exception rate measured by Ardent Partners in January 2026 is 18.4%, at an average cost of $9.84 per document and 8.2 days of processing time. Anyone promising zero exceptions is describing a system that guesses.
The useful comparison is with what happens today without a system. Manual transcription gets about 1% of values wrong, and rereading the document by eye doesn't reduce that: a study by Barchard and Pace in Computers in Human Behavior, comparing data entry methods across 195 people, measured exactly this.
The point is what those errors look like, and the authors' lab puts it in so many words: “most of those errors are within the allowable range for the variables”. They're plausible numbers instead of the right ones, so they get past any check done by eye. The automatic check tells you which values are uncertain, and that's the one thing the eye can't do.
The share of documents stopped is tracked, so if it rises you'll know. It's the sign that documents are arriving that differ from the ones the system was calibrated on, and you see it before it becomes a problem.
In Italy, invoices already arrive structured, and that's worth knowing
There's nothing to read on Italian invoices, because they already arrive as data. The Agenzia delle Entrate, the Italian Revenue Agency, states that since 1 January 2019, all invoices issued for supplies of goods and services between parties resident or established in Italy can only be electronic invoices. In 2025, 2.4 billion of them went through the exchange system, issued by 5.5 million businesses and individuals.
That means Italian invoice data already arrives structured, and reading it with a vision model would introduce errors where there are none. The system takes these invoices for what they are and does the real work elsewhere.
The real work is in orders, which arrive as PDFs or typed into an email, delivery notes, which may still be on paper, the receipts gathered at the end of the month, and everything that comes from foreign suppliers and customers, which doesn't go through the exchange system. That's where the retyping happens today.
Same mechanism, different name in every trade
The mechanism is identical; the document being retyped isn't. It's worth looking at your own case, because that's where you see what it's costing you today.
| Sector | The document being retyped | Where you see it |
|---|---|---|
| Food and agricultureand export | An order from a foreign distributor, typed into an email in English, with product codes that aren't yours and outside the Italian e-invoicing system. |
|
| Restaurantsand bars | Suppliers' delivery notes, which arrive on paper every morning and have to be checked against what actually came into the warehouse. |
|
| Hospitalityaccommodation and events | A whole season's expense claims and receipts, gathered in an envelope at the end of the month and retyped one by one before they go to the person who keeps the books. |
What this piece doesn't do
It doesn't replace your business software or impose a new one. If yours can't be connected to from outside, the system stops one step earlier and hands over the file to import.
It doesn't do the bookkeeping. It prepares the data and hands it over in order, which shortens the monthly gathering, and the bookkeeping stays with the person who signs it off. Monthly reporting works along the same lines, turning that data into something readable once a month. The reasoning behind keeping a person in the loop is on our page about the principles we build by.
It doesn't decide what to do with the document. Enquiries that come in alongside documents and need a reply are a different problem, handled by Inbox AI, the first reply to every enquiry.
Questions and answers
What documents can it read?
Orders as PDFs or typed into the body of an email, delivery notes (even photographed ones), receipts and expense claims, invoices from foreign suppliers, price lists and order confirmations. In general, any recurring document for which there's a reasonable sample to start from.
Italian e-invoices are taken as data rather than images, because since 1 January 2019 they have gone through the national exchange system and arrive already structured.
How can I trust what it writes into our business software?
Because every extracted field comes with a measure of how confident the system is that it read it correctly, and you set the threshold field by field. Above the threshold the data goes in; below it, the document stops and is shown to you with the doubtful field highlighted next to the original.
The fields that involve money, such as quantity, net amount and order number, have the highest threshold of all. A wrong figure in your business software is worse than a missing one, because nobody ever reads it again.
Does the data come out in our business software's format?
Yes, and that's decided during the analysis. The fields to extract, what they're called and the output layout are agreed once and stay yours, even when the incoming document calls them something else or uses another language.
Where your business software can be connected to from outside, the data is written straight in. Where it can't, the system prepares a file in the layout that software imports.
Is it for bookkeeping? Does it replace the accountant?
No. The system prepares the data in an orderly form and hands it over ready to use, which cuts down the gathering work that's done by hand every month.
Bookkeeping entries, the tax treatment of a document and any decision that affects the tax return stay with your accountant, and we give no advice on that side.
How do you measure whether it's working?
By the share of documents that go all the way through without anyone touching them, agreed as the metric before we start. We measure the time spent today on a batch of real documents, set the threshold the work has to beat not to count as a failure, and compare afterwards.
The second number is the share of documents stopped by the check, which should stay stable. If it rises, documents are arriving that differ from the ones the system was calibrated on.
Notes on sources
- The requirement for electronic invoicing between parties resident or established in Italy from 1 January 2019 is set out by the Agenzia delle Entrate. The 2.4 billion invoices processed and the 5.5 million issuers come from the Agency's own report I risultati 2025, pages 3 and 25.
- The 18.4% exception rate comes from Ardent Partners, January 2026, published by the firm that carries out the research. The survey doesn't state its sample size or geography, and it measures accounts payable departments, mostly large and US-based: we point this out because the readers of this page aren't that sample.
- The error rate of around 1% in manual transcription, and the finding that rereading by eye doesn't reduce it, come from Barchard and Pace, Preventing human error: The impact of data entry methods on data accuracy and statistical results, Computers in Human Behavior 27(5), 2011, pages 1834-1839, summarised by the authors' lab at the University of Nevada. It's a controlled experiment with 195 people transcribing research data, not office staff entering business documents: we cite it for the shape of the error, which stays within a plausible range and so goes unnoticed.
- We don't publish a percentage of time saved. The most widely quoted figure on this topic comes from a 2016 survey of data professionals, and it measures a different job and a different task from retyping business documents. The time is measured on your own batch of real documents, before any quote.
- This page doesn't report results achieved for a client, because this piece hasn't yet been delivered to a client. The tests mentioned are functional checks run in a test environment.
The other pieces in this group
Hours spent doing repetitive work by handFifteen minutes, with your case in front of us.
How many documents come into the business each month and get retyped by hand somewhere? If the answer is that nobody knows, that's already the most useful thing to find out, and it can be counted within thirty days. In fifteen minutes on the phone we'll look at it together and tell you where it makes sense to start, even if we never end up working together.
You'll speak to Mattia Esposito, who then builds the system: there's no salesperson in between. If you'd rather measure things yourself before talking, the Diagnostico (in Italian) is twenty questions and five minutes.