Data Insights · Automation & AI
How to automate invoice processing with AI without losing control
AI can read invoices well enough. The question is not whether it can, but where a person stays in control. How to automate invoice processing — and what the coming e-invoicing rules do and do not change.
Definition
Automating invoice processing means setting up a process in which incoming invoices are read automatically, matched against the purchase order and prepared as a booking — while a person approves the payment. Den Otter Solutions builds that process for small and mid-sized companies, with the reading step handled by AI and the decision to pay left to a person.
Of all the processes in a mid-sized company, invoice processing is the most obvious one to automate. High volume, heavy repetition, a hard check at the end. And at the same time the process people are most wary of — because it touches money leaving the building.
That wariness is justified, and it is solvable. The answer is not in whether you let AI read the invoices, but in where you keep a person in control. This article shows where that line sits.
That is the average cost of processing one invoice manually — opening, retyping, checking, booking, archiving. Automated, it drops to €2–€3 (APQC benchmark, 2024). At five hundred invoices a month, that is the difference between half a working week and half an hour.
Where the AI is — and where it is not
"Invoice processing with AI" sounds like one smart action. It is five, and only one of them is really AI. That distinction determines what is reliable and what is not.
- Intake rule work The invoice arrives via the mailbox, a supplier portal or a scan. The system recognises that an invoice has come in and queues it. Fixed rules, no AI.
- Reading and understanding AI This is where the AI sits. Every supplier has a different layout; a consolidated invoice looks different again. AI reads the supplier, amounts, VAT and line items — even when the layout deviates. This is what replaces the old, maintenance-heavy templates.
- Matching against the purchase order rule work Does the amount match what was ordered and delivered? A fixed comparison. Match: proceed. No match: exception.
- Catching the exception AI Price difference, missing order, duplicate invoice. AI recognises what deviates and summarises it, so a person sees in seconds what it is about instead of investigating it.
- Approving and paying human The booking sits ready next to the underlying order. A person looks, signs, pays. This stays human — not because it cannot be otherwise, but because it should not be.
AI reads the invoice. A person signs the payment. That is the whole line — and it is enough.
The line: where the person signs
The honest line
The temptation is to make the process "touchless": from arrival to payment without anyone looking. Technically possible. Rarely wise.
The workable line: automate everything up to the payment, keep the payment itself signed by a person. The booking is prepared, matched and checked by the system; releasing money stays a human act. For recurring, trusted suppliers below an agreed amount you can lower that threshold — but that is a choice you make and record, not a setting that is on by default.
That way you keep the time saving of automation and the control point that stops fraud and duplicate payments. Not everything has to be AI, and not everything has to be touchless.
Will mandatory e-invoicing change this?
A fair question, because something is coming. For cross-border B2B transactions within the EU, structured e-invoicing becomes mandatory from 1 July 2030 (the ViDA directive, adopted March 2025). For domestic invoices in the Netherlands nothing is decided yet: the Ministry of Finance received advice in March 2026 to make it mandatory domestically too, with a public consultation planned for late 2026 and legislation targeted around 2028. That is not settled policy.
But here is the point that is often missed: e-invoicing changes the front door, not the whole house. An e-invoice arrives as a structured file instead of a PDF. That makes the reading step easier — the first AI step shrinks. Matching against the purchase order, catching exceptions, coding to the right cost centre and approving the payment: all of that stays.
And the transition is years of hybrid. You will receive structured invoices from one supplier and PDFs from another, at the same time. A process that already handles mixed formats well is exactly the process that absorbs that transition effortlessly. Wait to automate "until e-invoicing arrives" and you automate under time pressure instead of at your own pace.
How to start without losing control
Not with the tool. With the process as it runs now. Den Otter Solutions starts an invoice engagement by mapping the actual flow: where invoices come in, who checks what, where they get stuck, which exceptions really occur. That step often reveals that some checks are being done twice.
Then you build the five steps, on your own accounts, with the payment left to a person. You run it alongside the old way for a month and measure: processing time, error rate, lead time to payment. Those numbers show up in your dashboard — not as a feeling, but as evidence for the next step.
Den Otter Solutions delivers this work remotely for clients in the Netherlands and abroad, at a fixed price per process.
The bottom line
Invoice processing is almost always the best process to start with: high volume, a hard check, a line that draws itself. AI reads and prepares, a person signs the payment. The coming e-invoicing rules do not make that redundant — they shrink the reading step and leave the rest of the process standing.
Process first, tool second. And AI only where it adds something: in the reading and the catching, not in the signing.
Frequently asked questions
-
On average €12 to €15 per invoice once you count the whole process: opening, retyping, checking, booking and archiving (APQC benchmark, 2024). Automated, that drops to €2 to €3 per invoice. At five hundred invoices a month the difference runs into several thousand euros a month, quite apart from the time freed up for work that does require human judgement.
-
The process has five steps: intake, reading, matching against the purchase order, catching exceptions and approving. AI handles two of them — reading invoices with varying layouts and recognising deviations. The rest is fixed rule work. Den Otter Solutions sets it up so that AI reads the invoice and prepares the booking, while a person approves the payment.
-
A person. The workable line is: automate everything up to the payment, keep the release of money human. The booking is prepared, matched and checked by the system, but approval stays a human act. For recurring, trusted suppliers below an agreed amount you can lower that threshold — as a deliberate, recorded choice, not a default setting.
-
No. For cross-border EU transactions e-invoicing becomes mandatory from 1 July 2030; for domestic invoices in the Netherlands nothing is decided yet. More importantly, e-invoicing only changes the format in which an invoice arrives, not the matching, coding, exception handling and approval. Those steps remain, and the transition is years of a mix of e-invoices and PDFs.
-
As a rule of thumb, manual processing becomes more expensive than automation somewhere around a few hundred invoices a month. But volume is not the only factor: many different suppliers, varying layouts and a high error rate make the business case positive at lower numbers too. Den Otter Solutions calculates this up front on your own figures, so you do not invest on an assumption.
-
Yes. Den Otter Solutions is based in the Netherlands and delivers invoice automation remotely for clients internationally. The process audit, the build and the handover all run remotely, in English or Dutch, at a fixed price per process rather than per hour.
Read next
AI automation
Fixed price per process, built on your own accounts, handed over documented.
Pillar 4Which processes to automate with AI — and which ones not
The five criteria for deciding whether a process qualifies. Invoice processing is one.
Pillar 1Why your accounting software gives you no steering information
Your package records what happened. Steering runs on something else.
Next step
Automate invoice processing, with the payment left to a person.
Den Otter Solutions starts with a process audit: the invoice flow as it actually runs, the five steps, and a clear line between what AI does and what you sign yourself. Fixed price per process, delivered remotely, no open-ended engagement.