Marketing automation for SMEs: what to automate and what to keep yourself

The twelve top-ranking Dutch-language articles about marketing automation have one thing in common. Six come from software vendors or comparison sites, four from agencies that sell the service. Every one of those twelve has an interest in you automating, and not a single piece says when you are better off not doing it. That is exactly the question an SME asks first. I answer it here, with the thresholds attached.

What is marketing automation exactly?

Software that runs marketing and sales processes along a fixed path without anyone having to start them again each time. An event is the trigger, a series of steps follows, and the system records what happened. It is not artificial intelligence: a fixed if-this-then-that path infers nothing and adapts to nothing. That distinction matters later on.

The confusion usually sits in the layer underneath. A CRM is the data layer: the Belgian statistics office describes it in its own survey form as software “for managing information about customers” that “helps track the customer’s interests and buying behaviour”. Marketing automation is what you run on top of that layer. Without the first, the second does not work, and that is not a detail but the most common reason projects stall.

The term is in our glossary too, alongside the vocabulary you will meet in quotations.

How many Belgian companies already do this?

Fewer than the stories suggest, and the real figure does not exist. No official statistic measures marketing automation separately. What we do know from the Statbel figures for 2025: two in five Belgian enterprises use CRM software, and that is the precondition, not the automation itself. Among companies with more than 250 employees it is 77.1%.

One figure from that same publication is sharper than all the adoption percentages combined. 27.7% of Belgian enterprises run data analysis on customer data such as purchase information, place of residence and preferences. Fewer than three in ten, in other words. That is precisely the raw material marketing automation runs on, and at seven in ten that raw material is not there.

Two things to read those figures correctly. They cover enterprises with ten or more employees, because that is the lower bound of the survey, and most of the Belgian SME population sits below it. And they measure CRM and data analysis, not automation.

The gap with large companies is real. According to the Digital Decade figures for 2024, which the sector federation Agoria cites in its analysis, 66.27% of large Belgian companies used AI against 23.09% of SMEs. Agoria is the federation of the technology industry and its members sell digitalisation, so read it with that interest in mind. The CEO of that federation put it this way when the figures were published:

“These good results are partly thanks to the efforts of the three regions. But getting 75% of companies to use AI by 2030 remains a challenge. We have to maintain the momentum, especially among SMEs.”

— Bart Steukers, CEO of Agoria (translated from Dutch)

Which task do you automate first, and at what volume does it pay off?

The task you do often, always in the same way, and where a mistake costs little. All three conditions together, not separately. Something you do twice a year is not worth the setup, however annoying it is. Something that runs differently every day will not fit a fixed path. And something where a mistake costs you a customer belongs with a human in the loop.

Those three together give you a usable test. Multiply how often a task recurs by how long it takes, and subtract how bad it is when it goes wrong. Whatever ends up on top is what you automate first.

On volume I would rather give no number than a number that is right for nobody, because the threshold differs per process. The calculation that does hold for your situation: count how often you did that task last quarter, multiply it by the time per instance, and set it against the time that setup and maintenance cost. If the first side comes out lower, it is too early. Less satisfying than a threshold, but it is your threshold.

If you are still short of that point and want more enquiries first, lead generation for SMEs is the logical first step. Automating what is barely there changes nothing.

When are you better off not doing it?

When your data is wrong, when the process is still changing, or when you do not actually know the problem. In those three cases automation only speeds up the existing problem, and that costs more than doing nothing. It sounds obvious, but the figures show this is the rule rather than the exception.

In a survey of American marketing leaders from January 2026, only 30.3% say their company has consolidated customer intelligence in a way that integrates customer data across all touchpoints. That is American data and therefore indicative, not Belgian. But the two thirds without a coherent view of the customer are exactly the group every vendor tells to “start today”.

The brake is rarely on the technology, by the way. From the Eurostat figures for 2025, among EU enterprises with ten or more employees that considered AI but do not use it: 70.89% cite a lack of relevant expertise, 52.52% a lack of clarity about the legal consequences and 48.83% concerns about data protection. Only 20.68% simply found the technology not useful.

There is one more reason to wait that nobody likes to write down: the process you want to automate may be unnecessary. A weekly report nobody reads becomes, through automation, a weekly report nobody reads and that you also have to maintain.

What breaks when a flow has been running for months?

Usually nothing visible, and that is precisely the problem. A field gets renamed, a connection expires, a colleague adjusts a form, and the automation quietly stops working. No error appears, because technically nothing is broken. You notice only when someone asks why those emails are not going out any more.

That is not an edge case. In the European Central Bank figures for the fourth quarter of 2025, 26% of euro area firms cite incompatibility with existing systems as a reason not to work with AI more intensively, after a shortage of skills (40%) and limited usefulness (28%). Automation runs aground on connections, not on ambition.

There is also a technical layer that breaks many setups and rarely appears in a quotation. Anyone sending more than five thousand messages a day to Gmail addresses must, according to Google, keep spam complaints below 0.30%, offer a one-click unsubscribe link and have SPF, DKIM and DMARC correctly in place. Miss that and a flow that is perfect on paper simply does not arrive. It also explains why the quality of your list weighs more than its size.

Three things you build in from day one, and that none of the twelve top-ranking articles mentions: a weekly check that the flow actually ran, a counter that tells you how many messages really went out, and one person who owns it. Without an owner an automation usually fades unnoticed.

If you would rather not work out which connections you need yourself, that falls under marketing automation as a service, which ClickForest delivers for SMEs in Flanders.

What may you send automatically in Belgium?

Less than most tools suggest. Advertising by email is prohibited in Belgium without prior consent. That is stated literally in article XII.13, §1 of the Code of Economic Law, as the Belgian Data Protection Authority itself quotes it: “The use of electronic mail for advertising is prohibited without the prior, free, specific and informed consent of the addressee of the messages.”

There is one exception, the soft opt-in, and it has three conditions that must all apply together. You obtained the email details directly in the course of selling a product or service, you advertise only your own similar products or services, and at the moment you collect those details you give the customer a free and simple way to refuse. That second condition is the trap: it must be what you supply yourself. Promoting a partner service falls outside it.

Two practical points many setups get wrong. The right to object to direct marketing is unconditional, according to the European privacy supervisors in their guidelines of October 2024, which still carry the status of a consultation version. The individual does not have to give a reason. And according to the Belgian supervisor’s checklist, mentioning that right in your privacy policy is explicitly not enough.

These are facts from the sources themselves, not legal advice. If you have a concrete case, read the recommendation of the Belgian Data Protection Authority itself; the Dutch-language version still carries the note that it is a consultation version and that the final text is for now available only in French.

What changes now that a model can decide for itself?

The branching moves from the flowchart into the model. Where you once had to draw every exception in advance, a language model can now classify, summarise and phrase within a step. The path stays yours, the filling in does not. That is the real shift since the classic flow builders, and none of the twelve top-ranking Dutch-language articles covers it.

Which tools are usable for that today is covered in the overview of AI tools for marketers. The difference with an AI agent is that an agent picks its own path while an automation follows yours. How that works exactly and when you need one or the other is fully worked out in what is an AI agent. What the broader AI layer can mean for marketing is covered in the AI marketing guide.

The soberest advice comes from a party that would stand to gain from telling you the opposite. Anthropic, which supplies language models itself, writes in its own engineering guide:

“we recommend finding the simplest solution possible, and only increasing complexity when needed”

— Erik S. and Barry Zhang, Anthropic

Add the judgement of an analyst house to that. Gartner surveyed four hundred and thirteen martech leaders between June and August 2025. Among those with AI agents in pilots or production, 45% say the agents offered by their vendor do not meet the promised business performance. Half report that their organisation lacks the technical and data readiness. Two caveats: the geography of those respondents is not stated, and these are organisations with a martech leader on staff, so larger than the average Flemish SME.

“AI agents are rapidly reshaping the marketing landscape, with leaders eager to harness their potential for content production and campaign optimization. However, business value is the yardstick CMOs should use to evaluate these investments, rather than vendor hype.”

— Benjamin Bloom, VP Analyst at Gartner

The link between a model and your own systems these days often runs through an open protocol; what MCP actually is explains that layer without requiring you to be technical.

Is an SME really behind large companies here?

On adoption yes, on deep use no. That is the most surprising finding of this research, and it appears in none of the twelve top-ranking articles. In the fourth quarter of 2025 more than 70% of euro area firms reported using AI, but only 7% describe that use as significant. That comes from the SAFE survey among more than five thousand firms, described by economists at the European Central Bank.

They put the lesson this way:

“The advent of AI has been widely hailed as a driver of productivity growth. Yet simply adopting AI does not guarantee measurable improvements in firms’ efficiency. What does matter is what they use the new technology for.”

— David Chaloupka, Tibor Lalinský and Paloma Lopez-Garcia, European Central Bank

And then the figure I found most surprising myself. In an ECB working paper from 2026, based on surveys of around six thousand firms across twelve euro area countries, the researchers write that this 7% share is similar across size classes, pointing to “a common ceiling on intensive adoption”. The paper does not represent the views of the ECB itself, as it states explicitly.

That refutes the assumption many SME stories run on. You are not behind large companies when it comes to deep use; they hit the same wall on this point. My conclusion from that: what an SME lacks is scale, not insight.

A Flemish practitioner describes the same pattern up close, in an interview at the launch of a paid course his organisation is involved in:

“In companies, AI applications today often get stuck in a pilot phase because there is not enough knowledge to scale them into solid business cases.”

— Grisja Lobbestael, director of Flanders Make (translated from Dutch)

What does such a project look like in practice?

I never start with the tool. The first conversation is about which process you repeat today and who does it, because that determines whether there is anything to automate. Only then comes the question of which systems it has to talk to.

At ClickForest I build those systems with Claude Code in VSCode, topped up with Make.com where that makes sense, and always connected to the tools that are already there. An example you can read up on: the order reporting for Bastiano ran on a tool whose maintenance depended entirely on one external maker. We took it over, rebuilt it self-hosted and extended it into a daily working engine with route planning per driver. One clearly defined task, and maintenance back in our own hands.

If you would rather not build it yourself, the overview of Belgian agencies that build AI agents shows which parties demonstrably offer that service, and which two questions to ask in a first conversation.

Two things I consistently advise against. Starting with a large platform before you know which process you want to change, because then you buy features you do not use. And tackling everything at once, because then nobody maintains it.

If you are looking for someone to set your AI course rather than automate one process, AI consultancy for SMEs is the better entry point. If you mainly want your email flows for a webshop sorted, email marketing automation for e-commerce covers that fully. If your automation revolves around customer questions, AI chatbots for customer service is the starting point, and if it is about what you show on your site, AI personalisation and CRO covers that angle.

Conclusion: which question do you start with?

This one: which process do I repeat in exactly the same way this month, and how often? Answer that first, and only then which tool belongs with it. In that order most SMEs end up with one task worth doing and three that are not, and that is a perfectly good outcome.

What struck me most while writing this is not that SMEs are behind. It is that the ceiling on genuinely deep use is equally low for everyone, including companies with a department for it. Anyone who automates one process well today and puts an owner on it is therefore doing more than most companies that have been paying for a platform for years. ClickForest sets up those projects for SMEs in Flanders, so read this with that in mind.