How to start with AI: a practical roadmap for SMEs

Most AI initiatives in SMEs stall at the same moment: right after the strategy session. A consultant delivers an 80-page report, you read through it, and then everyone is back to the daily grind. Three months later someone at drinks asks whether anything is still happening with "that AI thing". The answer is usually no.

This piece is about how it does work. No consultancy circus, no strategy without execution, no six months comparing platforms. Just a concrete pilot within two weeks, measured in time and money, on one clearly scoped process. Only then do you scale.

This approach isn't for everyone. If you're building a data warehouse for 200 employees, you genuinely need a strategy phase. But for 90% of Dutch SMEs, and according to CBS figures that's the vast majority, this is the approach that works.

Step 1: first understand what AI can and can't do

Before you pick a tool or call a vendor: make sure you roughly understand what the current generation of AI does. Not to become an expert, but to avoid building a chatbot when you needed a spreadsheet, or deploying an agent when an email template would have done the job.

At minimum you should know the difference between an LLM and an AI agent. A language model like ChatGPT or Claude answers a question. An agent carries out tasks with multiple steps. Anyone who isn't clear on that difference often buys the wrong thing.

What AI is currently good at: summarising text, classifying, rewriting, pulling structured data out of unstructured input, generating code, taking simple rule-based decisions. What AI isn't good at: doing arithmetic without tools, remembering exactly without a memory layer, giving guarantees about factual accuracy, and replacing existing software where deterministic logic is required. Keep that list in mind whenever someone floats "AI" as the solution.

Read the explainer on what AI is for SMEs if you don't know the basics yet. Forty-five minutes of reading saves you three months of work in the wrong direction.

Step 2: find your three time sinks

No abstract exercise, no workshop with post-its. Take a week, set a timer on your phone and note down three times a day what you or your team are doing. Or even simpler: ask the three most productive people on your team where they lose the most time.

You're looking for tasks with digital input and digital output. A quote based on a customer PDF is a good candidate. Counting physical stock isn't. The task also needs to be repetitive, with variation. Not identical, because then a macro is enough, but the same pattern. Think answering emails, rewriting content for different channels, entering invoice data, categorising support tickets. And the third requirement: you have to be able to express it in time. "This costs me 20 minutes on average, I do it 30 times a week." Without that number, you can't measure later whether it works.

According to the Stanford AI Index Report 2025 the productivity gain of LLM tools on repetitive knowledge work sits between 25% and 40%, with peaks up to 70% on writing tasks. So you need a time sink of at least 4 to 5 hours a week to make it worthwhile.

Step 3: pick one pilot

One. Not three at once, not "let's start broad". From your three candidates, pick the one with the highest time investment and the lowest complexity. That's often not the sexiest option, and that's exactly right.

Examples of pilots that worked well in SMEs over the past two years: categorising incoming emails and prepping draft replies in Outlook or Gmail, rewriting product descriptions for webshops from a supplier feed, automatically summarising meetings with action items per person, drafting quotes based on an intake call and a price list, or handling the first line of customer service for the standard questions that in practice make up 80% of the total.

Projects like an all-encompassing chatbot, a forecasting model without data or an agent that takes over the entire customer journey are not good first pilots. Those are phase-3 projects, and that's how you kill your belief in AI in one go.

If you're torn between candidates: do the AI scan, which gives you a ranking in 5 minutes of where the highest return per hour sits for your business.

Step 4: block two weeks in the diary

This is where most initiatives still stall. Not because it's technically hard, but because it has to feel urgent. So block two weeks in the diary, not a day longer.

Week 1 is building. On day one you capture the happy path with ten examples of input and desired output. Days two and three you build the first version. For 80% of pilots in this category you don't need to program anything. ChatGPT with custom instructions, Claude Projects, a Zapier flow or a Make scenario gets you very far. Only once the pilot works and scaling is on the table do you look at custom integrations. On days four and five you test with the real user, you or a colleague, not with an idealised demo. Note down where it goes wrong.

Week 2 is polishing. Days six to eight you fix the most common errors. Not all edge cases, only the errors causing 80% of the problems. On day nine you put it live in the real workflow for one person. On day ten you measure, and that's step 5.

Skip platform comparisons, RFPs and vendor pitch rounds. If the pilot works, in phase 2 you can look at whether it needs to move to a more serious platform. The OpenAI documentation and Anthropic's prompt engineering guide are free and good enough for the first version.

Step 5: measure what it delivers in time and money

Without measurement you don't have a pilot, you have a hobby. The measurement doesn't have to be complicated, but it does have to be factual.

Capture the following numbers before and after: time per task, error rate and user satisfaction. Time per task you measure with a stopwatch, not with a guess. Take five representative cases and time the old way, then clock the same five steps with the new workflow. For error rate you look at how many of those five outputs you have to correct. That matters more than time, because a process that's 90% faster but wrong half the time costs you net time. For user satisfaction skip the formal NPS, just ask: would you want to keep working with this tomorrow, yes or no. If the answer is no, the time saved doesn't matter.

Do the maths in money. A process going from 30 minutes to 8 minutes at 30 times a week is 11 hours saved per week. At an internal hourly rate of 60 euros that's 660 euros per week, or 34,000 euros per year for one pilot. Against 15 to 30 euros a month in tooling, the payback period is usually less than a week.

According to a McKinsey study from 2024, 23% of companies that start with AI achieve measurable revenue or cost impact within the first year. The difference between that 23% and the rest almost always comes down to this point: do they measure concretely, or not.

Step 6: document and train your team

Pilot works, numbers are in. Now comes the part 80% of companies skip: making sure it still works six months from now when you're on holiday.

Write down what the pilot does and doesn't do. Not as an ISO document, just as a single A4 page in Notion or Confluence. What's the input, what's the output, at which point does a human check or step in, and when do you switch back to manual handling. Also train at least two people. One bottleneck isn't success, that's a weak link. The AI training sessions from DataDream are set up specifically for this, but you can also do it yourself with two hours of explanation and a week of shadowing.

Finally, check your obligations under Article 4 of the AI Act. Since February 2025 this has been in force: anyone working with AI systems in your organisation has to be sufficiently AI-literate. That sounds heavy but isn't, you need to be able to demonstrate that the people using it understand what it does. Read the explainer on the AI Act or take the AI Act check to see what applies to your situation.

Documentation and training are often the difference between "nice pilot from last year" and "standard part of how we work".

Step 7: scale or stop

At the end of the pilot you decide whether to continue, stop or replace the approach. None of those options is wrong. What is wrong: ending up in a situation where you're no longer enthusiastic, but haven't officially stopped either.

Decide on numbers, not on gut feel. Time saved of 25% or more with high user satisfaction: scale. Add the second task from your time sink list and start again at step 3. Time saved between 10 and 25% with mixed satisfaction: another two weeks of polishing, no longer. If it's still not right after that, you stop. Time saved under 10% or low satisfaction: stop, learn, pick a different pilot. It's not a failure, it's a learning moment.

One scaling rule: only add a second pilot once the first has been running for at least 4 weeks without your attention. Otherwise you build a pile of half-working systems that all demand your time.

Most AI initiatives stall after the strategy session, not before it. Execution is the whole point.

What this is not

Just to be clear, these are the patterns this roadmap explicitly pushes back on.

No "AI strategy first". An AI strategy without pilots is a PowerPoint without a customer. Strategy follows the patterns you see in pilots, not the other way around. If a consultant tells you that you first need a two-year vision: thank them and start with step 2 above.

No "platform first". Microsoft Copilot, Google Gemini, your own LLM on Azure: those are all fine choices, but only in phase 2. In phase 1 it makes no difference. A pilot that works on ChatGPT also works on Copilot, and vice versa.

No "big bang implementation". Moving the entire company onto AI workflows in one go is how you lose everyone. One pilot, one team, one month. Then the next.

Common mistakes

In the pilots I've guided the same mistakes keep coming back. The scope is too big. "We want to automate our entire customer service" isn't a pilot, that's a programme. Start with the three questions that cause 60% of the volume.

No baseline measurement is done. Without a baseline you don't know whether it delivered anything, and "it feels faster" isn't a measurement. People commit to a platform too early. An annual contract with an AI vendor without a working pilot is almost always more expensive than two months of experimenting yourself. The user gets skipped. The person doing the work every day knows better where the friction is than management does. If they don't help build it, the pilot fails. And the exit criterion is missing. Agreeing upfront when you stop is just as important as agreeing when you scale.

When to bring in outside help

Most of the above you can do yourself. External help makes the difference at a few specific moments.

If you don't have time to focus for two weeks, an external partner is sometimes the only way to attach a hard deadline to it. If the pilot touches sensitive data (customer, financial or personnel data), you don't need to pay top dollar to a GDPR consultancy, but you do want someone looking over your shoulder to check that the setup is legally sound. If you want to scale to more than 5 processes, thinking about architecture becomes smart, and a strategy phase is a good investment. Only then, not earlier. And if your team isn't yet AI-literate under Article 4 of the AI Act, a one-day training helps you meet that requirement and saves you fines and hassle later.

Even with help: stay owner of the problem. The external party does it faster, you know the business better. The combination works, the handover rarely does.

Conclusion

Starting with AI isn't a question of strategy, it's a question of one two-week pilot on one concrete time sink, with hard numbers before and after. Do that once properly, and your second pilot costs half the energy. Do it ten times and AI is no longer a project, it's just how you work.

The biggest mistake I see in SMEs isn't that people pick the wrong tool, it's that they spend months choosing. Two weeks of action delivers more than six months of comparing.

Want to know which pilot delivers the highest return per hour for your business? Take the AI scan, 5 minutes, short and concrete. Or get in touch if you want to spar directly about your three time sinks.