Many AI experiments, few results

Most SMEs have by now started with AI somewhere. Someone uses ChatGPT for emails, the marketing team tests an image generator, and the bookkeeper tries a smart tool. According to Statistics Netherlands (CBS), in 2024 already 22.7 percent of businesses with ten or more employees used at least one AI technology, almost 9 percentage points more than a year earlier. Scattered experiments, in other words. The problem is that these experiments rarely come together into something that structurally improves the organisation. A lot is tried, but little is measured.

An AI strategy does not have to be complicated or expensive for an SME. The difference between tinkering and results lies not in the number of tools, but in the choice: where does AI solve a real problem, and how do you measure whether it works? This article shows how to move from scattered experiments to a strategy that pays off.

Start with the problem, not the tool

The most common mistake is starting from the technology. Someone reads about a new tool and goes looking for a use for it. That is backwards. A working AI strategy starts with the question of where in your business time leaks away or opportunities are left on the table.

Think of recurring, predictable work: customer questions that come back every week, quotes drawn up by hand, leads not followed up quickly enough. Those are the places where AI delivers value immediately, because it is work that lends itself to automation without quality suffering. A good first step is mapping those repetitive processes, not trying out as many tools as possible.

Three layers where AI already helps SMEs

A usable AI strategy divides applications into layers, from low-threshold to structural.

  1. Productivity. Helping individual employees work faster with AI assistants for text, summaries and research. Quickly introduced, but the gain stays personal and hard to measure.
  2. Processes. Automating whole workflows, such as customer service, lead follow-up or planning. Here the gain becomes visible in the numbers: shorter lead times, faster responses, fewer mistakes. Read how customer service shifts from reactive to proactive.
  3. Visibility. Customers increasingly do not search through Google alone, but ask their question to an AI assistant. Anyone who wants their business to appear in those answers has to make their own website readable for AI. This is called AI visibility and is still uncharted territory for many SMEs.

The mistake is usually that businesses stay stuck at layer one. There it feels modern, but it does not change the numbers. The real gain is in layers two and three.

Measuring is half the strategy

An experiment without a metric is not a strategy but a hobby. Before you deploy an AI application, you define what you want to improve and how you will see it. Response time to customer questions? Number of leads followed up? Hours spent on quotes?

This sounds obvious, but it is rarely done. Precisely by taking a baseline measurement up front, you can later prove whether the investment pays off. It also makes it easier to stop something that does not work, instead of continuing to pay for a tool nobody uses anymore.

Transparency and compliance are part of it

Deploying AI brings obligations. The European AI Act sets requirements for transparency: in certain cases you have to make clear that customers are dealing with AI, for example with a chatbot. For SMEs that is no reason to back off, but it is a reason to set it up properly from the start.

The same goes for security. Anyone who has customer data processed by AI systems must know where that data goes and how it is protected. This relates directly to the requirements of the Cybersecurity Act, the Dutch implementation of NIS2. See what the NIS2 Cybersecurity Act means for SMEs. An AI strategy that includes compliance from the start avoids expensive corrections later.

A workable approach in five steps

  1. Map repetitive work. Where does most time go to predictable tasks?
  2. Choose a process, not a tool. Pick a process with measurable pain, such as lead follow-up.
  3. Take a baseline measurement. Record how it performs now, so you can demonstrate improvement.
  4. Implement with human control. AI does the work, people guard quality and exceptions.
  5. Measure, adjust and scale. Does it work? Expand to the next process. Does it not? Stop and choose something else.

Conclusion

An AI strategy for SMEs is not about as many tools as possible, but about targeted choices: start with a real problem, pick a process with measurable pain, and guard quality and compliance. That is how AI turns from a collection of scattered experiments into something that improves the numbers. The businesses that tackle this structurally now are building a lead that experimenting competitors will find hard to close.

Sources

  • Statistics Netherlands (CBS), AI Monitor 2024 (use of AI technology by Dutch businesses, published 27 February 2025), cbs.nl
  • European Commission, AI Act (Regulation 2024/1689), digital-strategy.ec.europa.eu
  • National Cyber Security Centre, Cybersecurity Act and NIS2, ncsc.nl