To stop typing delivery notes and invoices by hand, the path is this: decide which fields you really need (date, supplier, number, net amount, VAT, total, line items), get every document to arrive through a single place —an email inbox or a folder—, run OCR that reads those fields, have a person review only what the system read with doubt, and push the result into your spreadsheet with the original archived alongside. You don't need to change software: the automation sits on top of what you already have.
Moving a delivery note into a spreadsheet is one of the most rewarding tasks to automate. It's repetitive, it has clear rules, and it eats a kind of time you don't see but do pay for: two minutes a document that, multiplied by hundreds a month, becomes whole days of someone copying numbers from paper into a cell.
What do you need before you start?
Before building anything, two decisions: what data you want to capture and where the documents will come in.
The data first. Take five real delivery notes and five invoices and underline what you actually use afterwards: almost always date, supplier, document number, net amount, VAT, total and the line items (description, quantity, price). Don't try to capture everything "just in case": every extra field is a column someone will have to check. Define few columns, and make them the ones you really look at.
The intake second. The automation needs a single point where documents arrive: an inbox invoices are forwarded to, a shared folder where PDFs are dropped, or a photo taken with a phone when the delivery note lands in the warehouse. If they currently reach you by five routes —some by email, some on paper, some by WhatsApp— the first job is to bring them all through one door. Without that, there's nothing to automate.
How do you do it, step by step?
With the fields decided and a single door for intake, the process is five steps.
- Decide which fields you really need. Date, supplier, number, net amount, VAT, total and line items. Those are your columns. The tighter the list, the more reliable and cheaper the automation.
- Get everything to arrive through one place. An inbox, a folder or a photo from the phone. A single channel through which all documents arrive, as PDF or image.
- Extract the fields with OCR. An OCR or document-reading service turns the PDF or photo into data: it reads the supplier, the date, the amounts and the line items. This step is the one that eats the typing. The technology exists and is mature; what needs tuning is which fields to read and how.
- Review only the doubtful ones. A serious OCR doesn't just read: it says how confident it is about each thing. Configure the system so that anything read with low confidence —a blurry amount, an odd date— jumps out for a person to validate. The rest passes through on its own. That selective review is what separates a reliable automation from a gadget that quietly feeds errors into your accounts.
- Push to the spreadsheet and archive the original. Each document, one row with its fields in columns. And store the original PDF or photo with a link in that row. That way you have the data to work with and the proof for when someone asks "where does this number come from?".
How much does it cost and how long does it take?
Building it depends on two things: how many different formats you receive and whether you want validation in the loop. A simple case —PDF invoices from a handful of suppliers into a Google sheet— is built in a matter of days. A case with many paper formats, skewed photos and human validation takes longer, because you have to tune the OCR and build the review screen.
The saving is on the other side of the scale. Typing a document by hand is a couple of minutes between opening it, reading it, copying the fields and checking you didn't slip. With hundreds of documents a month, that's hours every day of someone doing a task that adds nothing beyond the data itself. Automated, that person goes from typing everything to reviewing only what the system flags as doubtful: from hours to minutes.
| By hand | Automated | |
|---|---|---|
| Time per document | ~2 min of typing | Seconds; you only review the doubtful |
| Typing errors | The usual from copying by hand | The ones that slip through get flagged |
| Traceability | The paper, in a folder | The original linked in each row |
| When it pays off | Four documents a month | Tens or hundreds a month |
When is it NOT worth it?
Don't automate for the sake of it. There are cases where typing by hand is still the sensible thing:
- Low volume. If you receive four documents a month, building and maintaining the automation costs more than the time to type them. The maths is simple: how much time you save a year versus what it costs to build.
- Many formats, on paper, and few of them. If each supplier sends a different format, handwritten or on crumpled paper, and there are only a few, tuning the OCR for all of them isn't recovered. There, typing is the efficient choice.
- The final data is barely used. If you push delivery notes into a sheet no one then looks at, the problem isn't the typing: it's that the record is redundant. Automating something useless just makes it useless faster.
I say this because automation is sold as always good, and it isn't always. It's excellent when there's volume and repetition; it's an expense when there isn't.
What mistakes do people usually make?
The three stumbles I see again and again.
The first, wanting to capture everything. People try to read every field on the document "while we're at it", and every extra field is a source of errors and reviews. Fewer columns, better system.
The second, trusting the OCR blindly. Pushing whatever the machine reads without anyone reviewing the doubtful cases feeds silent errors into the accounts, which are the worst kind: the ones you find three months later. Good automation doesn't remove the person; it reserves them for the exception.
The third, not keeping the original. If you push the data and throw away the paper or lose the PDF, the day someone disputes an amount you have nothing to defend it with. The original linked in each row is what turns the sheet into something you can trust.
Frequently asked questions
Can I automate this without changing my accounting software?
Yes. The automation sits on top of what you already have: it reads the documents that come in and writes into your spreadsheet or current program. You don't need to migrate systems to stop typing.
Is OCR reliable for reading invoices and delivery notes?
For legible documents, very reliable on date, supplier and amounts. It struggles more with handwriting, skewed photos or odd formats. That's why the system should flag anything it reads with low confidence for a person to check: it's not about blind trust, but about typing only the exception.
How much time does automating delivery notes save?
It depends on volume, but typing a delivery note by hand takes a couple of minutes between opening it, reading it and copying it. With hundreds a month, going from typing everything to reviewing only the doubtful ones usually cuts the work from hours to minutes a day.
When is it NOT worth automating?
When you receive four documents a month: it takes longer to build the automation than to type them. It also doesn't pay off if each supplier sends a different format on crumpled paper and there are only a few: the effort of tuning the OCR isn't recovered.
Automate the rule, type the exception
Moving delivery notes and invoices into a spreadsheet automates well because it's repetitive and has clear rules. The key isn't that the machine reads everything, but that it reads most of it and warns you about the doubtful cases so you only review those. That's how you gain the time without losing reliability.
Start with the highest volume —the supplier who sends the most invoices—, measure the time you save, and grow from there. And always keep the original: data without the proof is worth half.
How many hours a month go into typing documents at your company?
If someone on your team spends their afternoons copying delivery notes or invoices into a spreadsheet, chances are a good chunk of it can be automated. Tell me how your documents come in and we'll look at it.
See the automation service Let's talk about your case