Process example, without demonstration
AI reads fixed fields and flags deviations. An employee checks and decides before anything is booked.
Start with one invoice and its job sheet. Have differences flagged and check them against the source. See the administration workflow.
Administration: invoice and purchase order
Invoice: 12 filters. Purchase order: 10 filters.
Invoice number, supplier and quantities next to the original order. Difference: 2 filters.
The workflow reads fixed fields from a document, compares them against the source and shows the differences. Booking, approving and rejecting stay with an employee. This works for document types with a consistent structure; on one-off exceptions the workflow stops and the case goes to the process owner.
In many SME offices the same document type keeps coming back: an invoice, work order, quote, registration form or attachment to an email. Someone opens the file, looks for the fixed fields, retypes them into a system and checks whether anything is missing. That retyping is slow, error-prone and hard to hand over when the regular employee is away.
You want less retyping between document and system, without AI booking or paying on its own.
You want the fields, the gaps and the differences with your own records already lined up, so that you only have to approve.
The workflow takes over that preparatory work. AI-first here means: first look at what a model can prepare, only then who does what by hand. The employee keeps the decision: is this correct, may it proceed, or should someone look at it.
| Now | After delivery |
|---|---|
| Open the file, find the fields, retype into the software | The agreed fields are ready, with the source next to them |
| A missing attachment or deviating amount is noticed later | Gaps and differences are visible before approval |
| Booking happens in the same motion as checking | Nothing is booked until an employee approves |
By email, upload, shared folder or an existing software package. One document type per implementation, not all incoming post at once.
The model extracts the agreed fields: date, amount, reference, party, missing attachment. What is not in scope stays outside the workflow.
Code compares the fields with your own records: customer status, rate, outstanding balance, mandatory documents. That check is not a question to the model, so the same rule applies to every case.
The employee sees the proposal, the source used and what deviates or is missing, before anything is booked or sent.
Uncertain or deviating cases go to a human via an agreed escalation path. Every step is logged.
Digital documents plus Exact, AFAS, Twinfield or a folder. If the software already covers the whole route, you use the software. An integration with a second system counts as scope, at most two.
Only the fields of this document type go in. Comparison with your own records happens in code. Booking, paying or sending stays with the employee. The design choices are on security.
From the agreed start and complete information: baseline and design, followed by build, test on real documents of this type, final measurement and handover within ten working days. The full route is on the how it works page.
€1,500 excluding VAT, ten working days, one document type. Check the numbers in the calculator.
No workflow for a document that rarely occurs, for free text without fixed fields, or for a route in which AI books, pays or commits to a customer on its own. And no second document type in the same implementation: that is a new process.
The employee remains owner of the outcome and handles the cases the workflow does not understand. The goal is less retyping and less searching for missing documents, not automatically fewer people. Whether that goal is reached on your process is shown by the baseline and the final measurement. The way of working is on the how it works page.
No. The workflow produces a verifiable proposal: fields, missing documents and differences compared with your own records. An employee approves before anything is booked, sent or committed to.
Use that feature. I never build what the software already does well. Custom work is only worth investigating when documents, exceptions and a second system together form a process the software does not cover.
Above, you can see how AI compares an invoice with a purchase order and flags a difference. I build this workflow for your document types, formats and systems.
First step
Send a short message: what you do, what you are struggling with and what you want to improve. I personally tell you where best to start.
Message JoostOr send an email