AI automation for SMBs: what already works?

AI automation already works well in SMBs on one type of task: turning unstructured input — emails, documents, meeting notes — into structured steps within an ordinary workflow. Not as a replacement for your processes, but as a new process step within them. The pressure to get started is real: for 29.7 percent of Dutch companies, automation is the most important measure against the labour shortage (CBS, 2026). This article separates what already works from what doesn't yet — with the GDPR ground rules and a level-headed order in which to start.

What is AI automation — and what is it not?

AI automation is combining classic workflow automation (rules, integrations, fixed steps) with AI models that can handle unstructured input: text, emails, documents. The rules still drive the process; the AI does the step that used to require a person — reading, summarising, categorising, pulling data out of a document. Not a replacement for your processes, but a new kind of process step within them.

What it is not: sticking a chatbot on your website or 'doing something with AI' as a goal in itself. The level-headed order stays the same as with all automation — first the process, then the tool. A company that still retypes its order flow gets more out of an ordinary integration than out of any AI experiment; see business process automation.

Why now? The figures

The pressure comes from the labour market: nearly two thirds of Dutch companies face a staff shortage, and for 29.7 percent of all companies, doing more with automation — robotisation or AI support — is the most important measure against it (CBS, 2026). The ICT sector leads the way: 44.1 percent invest in automation, against 28.5 percent a year earlier (CBS, 2026).

At the same time, the gap between large and small is considerable: 40.4 percent of large companies invest, against 20.1 percent of small ones (CBS, April 2026). For SMBs that is an opportunity rather than a threat: the technology itself is affordable per user and available through existing platforms — the difference is in the approach. More figures with sources are in Automation in SMBs: the figures.

What already works in SMBs?

The applications that prove themselves in practice share one feature: bounded input, checkable result. Concretely: sorting and summarising incoming email (triage for support or sales), pulling data out of documents (purchase invoices, CVs — see CV parsing, packing slips) and feeding it into your system as structured fields, preparing draft texts (quotation texts, answers to frequently asked questions) and summarising minutes or meeting notes into action points.

In all these cases the AI is one step in an ordinary workflow: the email comes in (trigger), the AI categorises it (step), the rules decide what happens (routing, task, draft reply). The platforms that SMBs already use — Make, Zapier, Power Automate — have by now built in these AI steps; what such an AI agent exactly is, we explain in the glossary.

What (not yet) — and the ground rules

Don't let AI make decisions with financial or customer consequences without human oversight: final quotations, payments, rejections of candidates or claims. AI output is linguistically convincing, even when it's wrong — so every AI step needs a check: a person who confirms, or a rule that validates the result before the system builds on it.

The GDPR ground rules apply in full: having personal data processed by an AI service requires a data processing agreement with that service, and sensitive data has no place in tools where you don't know where the data goes. In practice: choose business versions of AI services with clear data-processing arrangements, and set out which data may and may not go through AI steps.

Where do you start?

Start with a process that is already partly automated and gets stuck on one unstructured step — there the AI step is a jigsaw piece in an existing chain, not a project in itself. The order: pick one task with clear input and output, run it for four weeks with a check afterwards, measure the error rate and the time saved, and only then decide about scaling up.

And hold on to the level-headed test: if an ordinary rule or integration can do the same, the ordinary rule wins — it is cheaper, faster and more predictable. Where the manual work sits in your case, the free Operations Scan maps out; the tool choice is covered in Zapier vs Make vs Power Automate.

In short

AI automation = classic workflows plus AI steps for unstructured input; the rules still drive the process.
Proven applications: email triage, document processing (invoices, CVs, packing slips) and summaries into action points.
For 29.7% of companies, automation is the most important measure against the labour shortage; the ICT sector leads the way at 44.1% (CBS, 2026).
Let AI prepare and people decide — no final quotations, payments or rejections without a check.
GDPR: a data processing agreement with every AI service, and setting out in advance which data may and may not go through it.

Frequently asked questions

What is the difference between AI automation and ordinary automation?

Ordinary automation works with rules and structured data: if this, then that. AI automation adds steps that can handle unstructured input — reading emails, extracting data from documents, summarising text. The rules still drive the process; the AI replaces the reading step that used to require a person.

Which AI applications already work in SMBs today?

Email triage (sorting, summarising, draft replies), document processing (pulling data from invoices, CVs and packing slips into system fields) and summarising conversations into action points. The common feature: bounded input and a result that a person or a rule checks.

Am I allowed to have personal data processed by AI (GDPR)?

Only with the right arrangements: a data processing agreement with the AI service, clarity about where the data is processed and a judgement about which data may go through it at all. Choose business versions of AI services with explicit data-processing arrangements and set out internally what is and isn't allowed.

What does AI automation cost?

The AI steps themselves are usually a modest item: they come as a feature in platforms such as Make, Zapier and Power Automate, or are billed per use. The real cost sits in setting up the process around them — the same order of magnitude as ordinary automation: platform flows for a monthly fee, custom work indicatively €1,500–€15,000 (see What does an API integration cost?).

Does AI automation replace employees?

In SMBs the reality is the other way around: automation — with or without AI — is deployed precisely because there aren't enough people. For 29.7 percent of companies it is the most important measure against the labour shortage (CBS, 2026). The work that disappears is reading, retyping and sorting; the decisions remain human work.

Written by Hugo Eleveld · Updated June 2026. This article is informational; for tailored advice book an intro call.

Prefer this kind of insight in your inbox?

Leave your email address and we'll put you on the list and email you as soon as the next edition on IT, automation and dashboards comes out. You can unsubscribe at any time with a single email.