Invoice automation: from the inbox to the ledger without manual retyping
An incoming invoice passes through several pairs of hands in a company and is retyped at least once. Most of those steps can be removed — and with mandatory e-invoicing in Slovakia from 2027, it is worth doing sooner rather than later.
Invoice automation means invoice data is no longer retyped anywhere. The invoice arrives by email, the system extracts the supplier, the amount, the due date and the payment reference, writes it into your system and flags only what does not add up.
The same applies in the other direction: issuing an invoice after delivery or after payment without anyone clicking it out by hand.
An important caveat: invoice automation is not the same as being ready for mandatory Slovak e-invoicing. That is a legislative requirement with its own rules — I write about it separately in eFaktúra 2027 na Slovensku (in Slovak). Automation will, however, make preparing for it considerably easier, because it forces you to put your documents in order now.
Where the time goes on invoices
The invoice itself takes little time. The problem is that there are more steps around it than it seems:
- The invoice arrives in an inbox — somebody has to spot it and download it.
- It is saved somewhere accounting will find it. Usually a folder whose name you have to remember.
- The data is retyped into the system — supplier, amount, date, payment reference.
- It is checked against the order or the delivery.
- It goes into the payment queue and somebody watches the due date.
- Once paid, it is matched against the payment on the bank statement.
- At month end, somebody checks that everything arrived and adds up.
At ten invoices a month that is bearable. At a hundred it is a job of its own — and most of those steps are pure retyping, which is where the typos come from too.
What can and cannot be automated
| Step | Can it be automated? | Note |
|---|---|---|
| Catching the invoice from email | yes, reliably | by sender, subject or attachment |
| Saving into a folder structure | yes, reliably | named by supplier and date |
| Extracting data from a PDF | yes, with checks | electronic PDFs reliably, scans less so |
| Writing into the system | yes, if the system has an API or an import | this tends to be the main constraint |
| Matching against an order | partly | with clear identifiers yes, otherwise a suggestion to approve |
| Suggested posting | partly | a suggestion from history, approved by the accountant |
| Due-date checks and reminders | yes, reliably | simple and very useful |
| The decision to pay | no | stays with a person and should stay there |
| Judging tax correctness | no | that is the accountant's job |
The key distinction: automating the transfer and preparation of data is safe and pays back quickly. Automating the decision is a risk. That is why a well-designed system prepares everything and leaves a person a single click — approve.
Data extraction: what works and what does not
This is where expectations are least realistic, so plainly:
- Electronic PDFs (generated by an invoicing system). The text can be read directly and extraction is very reliable. This is most incoming invoices.
- A scan or a photo. This needs OCR and accuracy depends on quality. A clean scan is usable; a photo of crumpled paper at an angle is not.
- Unusual layouts. Foreign invoices and uncommon templates need more tuning. A language model copes with them better than fixed rules, but not for free.
- Checksums are mandatory. I always set up validation: does the line-item total match the invoice total? Is the date in a sensible range? Does that supplier exist? If not, the invoice goes for manual review.
Never let extracted data reach accounting unchecked. Both the model and OCR can be wrong — most often on the amount or the date. So the system proposes and a person confirms. The time saving stays practically the same and the risk drops to zero.
The connection to mandatory Slovak e-invoicing from 2027
From 1 January 2027, Slovakia introduces mandatory electronic invoicing in B2B. That is not the same as emailing a PDF — it means a structured format that systems can read without OCR.
For you that has two practical consequences:
- Extraction stops being necessary for invoices that arrive in a structured format. The data will be usable directly, which simplifies the whole process.
- The rest of the automation stays the same. Routing, matching, due-date checks and approvals will still be needed — and that is most of the benefit.
So invoice automation built today is not a wasted investment. Quite the opposite: companies whose documents and process are already in order will have a markedly easier transition. I go through the legislative detail in eFaktúra 2027 na Slovensku (in Slovak).
Nexflow does not provide tax or legal advice and does not replace a certified e-invoicing service provider. Check your specific obligations with your accountant and in the official sources for your country.
How to build it
- Map how invoices reach you. One inbox? Several mailboxes? Supplier portals? This drives the complexity more than anything else.
- Check what your accounting system can do. Does it have an API? An import from a format you can generate? This is the most common obstacle and the first thing to establish.
- Start with one supplier. Ideally the one sending the most invoices in a consistent format. Extraction and validation get tuned on them.
- Set the validation rules and decide what happens on uncertainty — where the invoice goes and who handles it.
- Extend to the others only once the first one has run unattended for at least a month.
What it costs and when it pays back
- from €390Catching invoices from email, saving them into a structure, and due-date reminders.
- €700 – 1,200Data extraction with validation rules and writing into a system that has an API or an import.
- from €1,200Matching against orders, suggested postings, an approval process, several invoice sources.
A model calculation (illustrative, not a client's result): 80 incoming invoices a month × 4 minutes to download, save and retype = roughly 5.3 hours a month. At a cost rate of €15/h that is about €80 a month. A €700 solution pays back at that volume in roughly nine months — and that is without counting time saved hunting for documents and fixing errors.
Run it on your own numbers. Below roughly 30 invoices a month, full extraction usually does not pay back — but catching, saving and due-date reminders on their own pay off almost always, because they are cheap.
When it is not worth automating invoices
- You have few invoices. Below roughly 30 invoices a month, full data extraction usually does not pay back. At most, catching, saving and due-date reminders make sense.
- Your accounting system has no API or import. You would still be retyping the data into it by hand. First find out what the system can accept.
- Invoices arrive as photos or poor scans. Extraction will be unreliable and manual checking will eat most of the saving.
- Most of the time goes on judgement, not retyping. The decision to pay and tax correctness stay with a person. If that is the main work, automation will take little of it away.
Frequently asked questions
Can an invoice be read automatically from a PDF?
Yes. With electronic PDFs generated by an invoicing system, extraction is very reliable because the text can be read directly. With scans and photos, OCR is used and accuracy depends on the quality of the original — which is why validation rules are always added and an uncertain case goes for manual review.
Can the system get the extraction wrong?
Yes, most often on the amount or the date. That is why extracted data is never sent to accounting unchecked — the system proposes, a person confirms. Checksums and supplier verification catch most errors before that.
Will it work with our accounting system?
It depends on whether it has an API or an import from a format that can be generated. That is the most common obstacle in these projects and the first thing I establish — before quoting.
How does this relate to mandatory Slovak e-invoicing from 2027?
They are two different things. E-invoicing is a legislative requirement for a structured document format. Automation is a way of processing documents without manual work. They overlap in that structured documents remove the need for extraction — the rest of the automation stays useful.
From how many invoices is it worth it?
Full data extraction usually pays back from around 30 incoming invoices a month. The simpler parts — catching from email, saving into a structure and due-date reminders — pay off at almost any volume, because they are cheap to deploy.
Does it replace an accountant?
No. It removes retyping and hunting for documents, but judging tax correctness, deciding to pay and carrying responsibility stay with a person. In practice it frees the accountant's time for work that cannot be automated.
Where do our documents go in the process?
It depends on the design. If an external language model is used for extraction, the invoice content is sent to the provider — I tell you that up front, and it can be limited or built so the documents never leave your infrastructure. Nexflow does not provide legal advice; with sensitive data I recommend an assessment with your DPO.
Sources
- Financial Administration of Slovakia — official information on electronic invoicing
- n8n — official documentation — nodes for working with email, files and APIs
Related services
Related articles
How many invoices a month do you retype by hand today?
Tell me how invoices reach you and which accounting system you use. Free of charge I will tell you what can be connected and whether it pays back at your volume.
Get a free consultation →