AI AUTOMATION
AI in the systems a business already uses
2026
AT A GLANCE
- Project: anonymous
- Sector: Financial administration
- Location: the Netherlands
- Year: 2026
- Built: AI inside an existing bookkeeping setup: reading invoices, coding them to ledger account, cost centre, VAT and period, and getting them ready to post. A person signs off.
- Services: AI & automation
- Live: no public link
- Built by: the AMASO team
Most companies don’t need a new system. They need the one they already pay for to stop eating hours. The software is usually fine. The logins are fine. What costs the month is the part where someone reads something off one screen and types it into another.
Nobody wants to hear that the fix is a migration. A week of retraining, a year climbing back to where you already were, and a bill at the end of it. So most of the time, that is not what we do.
Two ways in
Build on what is there
Your systems stay. We find the part that is still done by hand and put the model behind that part. The tools stay, the logins stay, the reports people already trust stay. What changes is that the typing stops. Nobody has to learn anything new, and that is usually the difference between a project that lands and one that quietly gets abandoned.
Rebuild it properly
Sometimes the old system really is the problem. The process lives in a spreadsheet, in one person’s head, or in software that was never going to bend. Then we build it from scratch, as a real application, with the AI designed in rather than bolted on later. That is the more expensive door, and we’ll tell you when you are standing in front of it.
What this looks like in practice
Less abstract. A document arrives and the system reads it, instead of someone retyping it. A form fills itself with data the business already holds somewhere. A report someone put together by hand every Monday puts itself together on Sunday night. A queue that was worked through item by item arrives already sorted, coded and checked, with the few genuinely unclear cases set aside for a person. That person stays. They move up a level, from typing to deciding, which is what they were hired for.
Code from the document, not the supplier
The clearest example is invoice processing, inside bookkeeping that was already in place rather than beside it. A supplier document comes in, gets read, gets coded to the right ledger account, cost centre, VAT treatment and period, and is ready to post. The rule that makes it reliable is the one a careful bookkeeper already works to: code from the document, never from the supplier’s name. A description on a statement will easily send you to the wrong account, and the period a service was delivered in is often not the period named in the text.
A person still signs off
This is the part we don’t bend on. Automation that quietly books the wrong thing is worse than no automation, because you lose the mistake and the trail to it in one go. So the flow keeps its human step. Anything the system isn’t sure about is held rather than guessed, with the document and the reasoning attached, so whoever picks it up can see why it stopped. Approval chains stay exactly where they are. We don’t remove the person who signs. We give that person a queue of decisions instead of a queue of typing.
The bar, whichever door
Before we call it done, it has to be faster than what it replaces, measured against that old process and not against a slide deck. It has to leave a trail someone can audit, so there is an answer when a number is questioned six months later. And someone who doesn’t write code has to be able to run it, because a system only its maker can operate has a single weak point, and that is a person.