- Manual data entry is repetitive and low-value: the ideal candidate to start from.
- Process first, tool second: if documents arrive in chaos, no automation will save you.
- Starting Monday, you can pull the data from a folder of documents and get a ready-to-use Excel.
- When the work becomes critical (integrations, checks, high volumes) it’s no longer a spreadsheet: it’s automation in production.
What manual data entry costs you, in euros
When you look at a stack of invoices and receipts to record, you see an hour of boring work. The real bill is higher. Industry reports on accounts payable estimate that processing a document by hand costs between 15 and 40 dollars, that the average time from receiving it to recording it is about two weeks, and that nearly two in three companies still enter data by hand into their business software.[1]
It’s not just time. About four documents in ten contain an error,[2] and every error drags a correction, a phone call, sometimes a wrong payment behind it. It’s time taken away from what matters: cash, suppliers, decisions. And it isn’t a niche on its way out: the tools that read documents are a fast-growing market.[3]
What it costs you today, in euros
Drag the sliders to your numbers. It’s a starting estimate: the real figure comes from measuring the actual process.
Invoices, receipts, and slips to record
Reading and transcribing by hand
Gross salary plus payroll costs, per hour
Share that needs a correction
The cost of errors here is simplified to 10 minutes of correction per wrong document. It doesn’t count wrong payments or missed deadlines: in reality it weighs more, not less.
Process first, tool second
Before you think about the tool, look at the process. Documents arrive in different formats (PDFs, phone photos, scans, sometimes handwritten); someone opens them, copies the fields into a sheet, sums by category, and files them away. The tool removes just one piece, but the heaviest one: the transcription.
We say it to every client: if disorder reigns upstream, documents that never arrive, categories that change every month, files scattered across five folders, automation won’t put things in order for you. It speeds up the disorder it finds.
Automation amplifies a healthy process. On a messy process, it only speeds up the mess.
If your bookkeeping is still done "by instinct," the first job isn’t the tool: it’s making the process repeatable. How to do it, we explain in how to standardize a process before automating it.
Upload the folder, get the Excel
Let’s say the process is clear. Here’s the version you can try right away, on a folder with this month’s documents. You connect the folder to Claude Cowork, upload the documents, and ask it to extract the data. In one step you get an Excel sheet with one row per document and the totals already calculated.
| Date | Supplier | Total € | Category | To verify |
|---|---|---|---|---|
| 06/03/2026 | Caffè Centrale | 12.40 | Dining | |
| 06/05/2026 | Esso Station | 70.00 | Fuel | |
| 06/06/2026 | photo barely readable | 18.50 | Misc | Yes |
Three things matter more than they seem. It reads crooked photos and handwriting too. It calculates the totals with real formulas, so if you fix a row the sheet updates on its own. And when a value isn’t readable it doesn’t make it up: it leaves it blank and marks it "to verify," so your eye goes straight to the rows to check.
- A "Documents" sheet: one row per document, with date, supplier, amounts, and category.
- A "Totals" sheet: sums by category and by currency, calculated with real formulas.
- The uncertain rows highlighted and marked "to verify," so you only check those.
Generate a new Excel, or update your own
Depending on how you work, you can start from scratch or keep the file you already use alive.
Generate a new Excel
Start from scratch and get the file ready, with a Documents sheet and a Totals sheet.
When it helps: this month’s expense report, or digitizing a backlog of paperwork.
Update your Excel
It works on the register you already have in the folder and adds the new rows at the bottom, keeping your columns.
When it helps: you already have your file and just want to keep it up to date.
Make it run itself, every morning
So far it’s an operation you launch whenever you want. The next step is making it recurring: a routine that checks the folder every morning, works only the new documents, and updates the register. At that point you don’t "upload and ask": you drop the file in the folder and find it already recorded.
Honestly: it needs a minimum of clear rules, where the documents end up, which categories you use, what to do with doubtful cases. It’s not magic, it’s a well-designed process. But it’s the leap that turns "a useful tool" into "work that does itself."
When a spreadsheet is enough, and when you need more
This approach is great for extracting and putting things into a table. It doesn’t replace human review of edge cases, it doesn’t reconcile with your business software or your bank on its own, and it doesn’t handle approval flows.
- Volumes are modest and a single person handles it.
- The data stays in Excel and doesn’t need to talk to other systems.
- There’s no approval flow across multiple people.
- Volumes are high or grow every month.
- Several people work on the same data.
- The register has to integrate with your business software or your bank.
- You need automatic checks, traceability, and compliance.
We always say when AI isn’t needed or isn’t enough too: it’s the fastest way to keep you from spending badly. You’ll find the typical cases in when you don’t need AI. When the work does become critical, it’s a process automation to put into production, integrated and controlled.
Got a transcription job that repeats every month?
On a call we look at it together: we tell you honestly whether it’s worth automating, and how. Fixed price and timeline before we start, a result on your data in 2–4 weeks.
The questions we get asked most
How do I pull data from invoices and receipts into an Excel?
Collect the documents in a folder, connect it to a tool like Claude Cowork, and ask it to extract the fields you need (date, supplier, amounts, category). You get a sheet with one row per document and the totals calculated. It works with photos and scans too.
What if the AI reads a document wrong?
It happens, especially with blurry photos or handwriting. That’s why a well-built extraction doesn’t make things up: it leaves the uncertain value blank and flags it as "to verify." You only check the flagged rows, not all of them.
Can I update my Excel instead of creating a new one?
Yes. You can have a new file generated, or have the new rows added at the bottom of the register you already use, keeping your columns.
When is custom automation worth it instead of doing it by hand every time?
When volumes grow, several people work on the same data, or the register has to talk to your business software and your bank. At that point a spreadsheet isn’t enough anymore: you need integrated, controlled automation, put into production.
I wrote this article myself. The method and the examples come from Yempik’s real work. For the writing I had Claude help me with editing and layout, and the tool in the practical example is Claude Cowork. The substance is mine; the tool is declared.
Sources
- [1]DocuClipper: cost and time of manual document processing and the share of processes still done by hand (2025-2026 statistics). www.docuclipper.com
- [2]Resolve: comparison of manual vs. automated processing, error rate per document. resolvepay.com
- [3]MarketsandMarkets: size and growth of the Document AI market (automated document reading). www.marketsandmarkets.com