01
Look behind the scenes
Free
Half an hour, by invitation
See everything we build running in our sandbox, the Proeftuin. No sales pitch, unless you ask for one.
Request an invitationBlog
An AI process has fixed and variable costs. Work out the cost per transaction, document, model call and exception, and set limits that apply while the work is running.
A software budget often starts with a familiar number: so many staff, so many licences per month. That works for office software, CRM packages and many other subscriptions. In many AI processes, usage costs come on top of fixed costs for capacity, software and support.
That does not have to make costs unpredictable. It does mean you need to know the cost per process as well as per user. Two teams with the same number of users can end up with very different usage.
With a language model, the cost is usually made up of several parts: the text you send in, the text that comes back, documents that first have to be made searchable, calls to other systems, storage of data you choose to keep, and logging and monitoring. Sometimes there is fixed capacity or a managed environment as well.
Not every process uses all of these. That is exactly why a price per user is often too crude a measure. Base the calculation on a specific task. For an order email you can count how many emails come in, how many pages the attachments have on average, how many cases can go straight through and how many go to an employee. For an agent that answers customer questions, you count questions, sources consulted, average answer length and cases that need handling a second time.
A monthly report shows quiet periods, normal load and peaks. The peak matters more than an average figure from a spreadsheet. That is where you see whether a process needs a cost ceiling, a capacity limit or a different way of working.
A budget alert that only goes off at the end of the month does little good. A limit that applies during processing works better. For example, a task must not cost more than an agreed amount. An unusually long document goes to a queue first. An agent may only consult the sources that belong to its task. An answer is capped in length when a short one is enough for the next step.
These rules do more than protect the budget. They also make a solution faster and easier to check. There is no need to attach a complete archive to an order when the item code, customer number and agreed terms are enough to create a draft order. Sending less data cuts usage, as long as the task still gets the data it needs.
The cheapest model call is not always the best choice. A model that costs little but often leads to rework shifts costs to the order desk or the service team. The reverse is also true: more expensive processing need not be a problem when it replaces a lengthy manual step and the outcome is reliable enough.
So compare the cost per correctly handled task with the effort the same task takes today:
cost per correctly handled task = total process costs / tasks that reach the right next step without rework
Include waiting time too, if it leads to a missed delivery date, a customer who rings to chase or an invoice that is booked late. The process owner usually knows which of these matters. Finance can help make that assumption explicit.
The organisation using the solution must be able to see what makes up the bill. Ask for costs per process or task type, not just one total for the cloud environment. Also ask who may change a usage limit and what happens when that limit is reached. Does the process pause, does the task go to an employee, or does it quietly keep spending?
That conversation matters most when a supplier offers a fixed monthly price. A fixed price can work well, but the assumptions behind it should stay visible: expected volume, number of sources, response time, support and exceptions. Otherwise the risk simply moves into a change request that turns up on a later invoice.
So an AI budget is less a sum set aside for a new subscription than an agreement about which work may be processed, within what limit, and how you can tell that the costs are still in proportion to the outcome.
Three ways to start
Start small, with one process that comes up every week. Choose how: from half an hour behind the scenes to a session at your offices.
01
Free
Half an hour, by invitation
See everything we build running in our sandbox, the Proeftuin. No sales pitch, unless you ask for one.
Request an invitation02Recommended
Free
Half an hour on Teams
You tell us what you already do with AI. We tell you what fits, or whether it is still too early.
Call with Marcel Logtenberg
03
From €495
At your offices, excluding VAT
An AI quick scan, an AI assessment or an AI assessment plus inspiration session, at a fixed price.
Get started