AI for business: where it actually pays off
AI pays off in a Dutch SME where a task is high-volume, rule-based and document-heavy: quotation intake, invoice processing, customer email, order entry and reporting.
Most AI advice for companies is either a tool list or a transformation deck. Neither answers the only question that matters to an owner: which specific process in my business would this earn money on, and how would I know before spending anything. This page answers that by function and by company size.
Last updated: 5 August 2026
AI pays off in a Dutch SME where a task is high-volume, rule-based and document-heavy: quotation intake, invoice processing, customer email, order entry and reporting. It reliably does not pay off on low-frequency judgement work, on processes that are not written down, or on unreliable data. The practical test before spending anything: count how many times a week the task happens and how long it takes. Under roughly two hours a week in total, no automation will return its build cost.
Where it pays
The order of deciding
Before you buy anything: which two numbers move, in which department, and what is realistic at your size.
Two numbers
Which number must move, and who reports it today? Without both there is no business case.
Where it pays
Sales, admin, production and support each leak in a different place. Start where it is measurable.
What is realistic
At ten staff a different project is feasible than at a hundred. The limit is ownership, not budget.
What comes first
The order that works: measure, then one process, then extend. The reverse rarely returns anything.
What schemes cover
WBSO and similar cover development, not purchase. That distinction decides whether you qualify.
The two-number test, before you buy anything
Before any tool, supplier or budget, get two numbers for the process you have in mind: how many times a week does it happen, and how long does one instance take. Multiply.
Under about two hours a week in total, stop: no build recovers its cost, however good the technology. Between two and ten hours a week, an off-the-shelf tool or a small automation is usually right. Above ten hours a week, and especially where several people do the same thing, custom work starts to make sense.
This single calculation prevents most failed AI projects, and it is the first thing our €2,500 audit does. You can do it yourself for free in an afternoon.
Where it pays off, by department
Ordered by how reliably it returns money in Dutch SMEs, not by how impressive it sounds.
| Function | What works | Payback |
|---|---|---|
| Finance & admin | Invoice processing, purchase-order matching, expense handling, dunning | Fastest. High volume, rule-based, tolerant of a review step |
| Sales | Quote intake from email/PDF, quotation drafting, follow-up chasing | Fast, and often the largest revenue effect rather than cost effect |
| Customer service | Drafting replies from your own knowledge base, triage and routing | Fast for drafting; slower if you insist on fully autonomous answers |
| Operations / logistics | Document processing, order entry, exception handling | Fast where document-heavy. See AI for logistics |
| HR & recruitment | Job-ad drafting, structuring applications, onboarding documentation | Moderate, and note AI in hiring decisions carries EU AI Act obligations |
| Management & reporting | Pulling recurring reports together, summarising, first-draft analysis | Moderate; the value is calendar time, not headcount |
| Forecasting | Demand, stock, capacity planning | Slowest. Genuine value but needs clean history, a year-two project |
What is realistic at your headcount
Company size changes what is sensible more than industry does.
| Headcount | Realistic first move | What to avoid |
|---|---|---|
| Under 10 | Off-the-shelf tools plus training. A licence and two hours of coaching beats any build. | Custom development. The volume is not there yet. |
| 10–50 | One bounded automation on the single most repetitive process, €1,500–€15,000. | Company-wide programmes. Do one thing, prove it, then do the next. |
| 50–250 | Custom integration against your ERP where the process is genuinely yours; a proof of concept first. | Buying a platform before knowing which process you are fixing. |
The order that works
Almost every successful Dutch SME AI project we have seen follows the same sequence, and almost every failed one skipped a step.
- 1. Literacy before tools. Staff who understand what these systems can and cannot do choose better problems. This is also an EU AI Act obligation under Article 4. See our training provider comparison.
- 2. One process, measured. Apply the two-number test. Pick the winner, not the most interesting one.
- 3. Buy before you build. Check whether existing software already does it. Frequently it does.
- 4. Prove on your own data. Demos succeed on clean data; your data is not clean. Insist on a proof of concept using real cases.
- 5. Then production, with an owner. Guardrails, monitoring, and a named person who notices when output drifts.
What the Dutch schemes will and will not cover
Custom AI development usually qualifies as R&D under WBSO, but note it reduces your own payroll tax; it does not pay a supplier invoice. Regional innovation vouchers may apply depending on your province. SLIM funds learning and development structures, not standalone commercial courses.
Our free subsidiecheck shows which schemes stack for your province, company size and goal, with source links and check dates. No email required.
Frequently asked questions
Where does AI actually pay off in a business?
Is AI worth it for a small business?
Where should a company start with AI?
What AI solutions exist for SMEs?
What does AI cost for a Dutch company?
Do we need to comply with the EU AI Act to use AI?
Not sure which process to start with?
Twenty minutes, no deck. We will run the volume test with you, and tell you if a tool beats a build.
Book a free consultation →