1. Home
  2. Insights
  3. Where to start
Guide · Tim Beckers · 2 min

Where do you start with AI in an SME?

With one task that comes back every week and costs hours. Measure what it costs now, build one working first version on your data, and then decide if you go on.

A path with three stations: a magnifying glass on a process map, a robot arm that builds blocks, and a team at a working dashboard with a flag.
01 / Why it goes wrong

The problem is the organisation, not the model.

What is missing: a result agreed upfront, a link to the real work and someone who decides.

95%

of the generative AI trial projects at companies show no measurable result in the profit and loss account.

77%

of the failures are organisational, not technical.

02 / Four steps

From one task to a measured result.

Four steps up. Pick one task: done when the task fits in one sentence. Measure the baseline: done when today’s number is on paper. Choose the route: done when you have decided, go or no-go. Build the first version: done when the new number is measured. At the top, the new number sits next to the baseline. New number next to the baseline 4Build the first versionDone: new number measured 3Choose the routeDone: go or no-go decided 2Measure the baselineDone: today’s number on paper 1Pick one taskDone: the task in one sentence

  1. Pick one task

    It comes back every week, costs hours and lives in systems you already have. Pick one, not three.

  2. Measure the baseline

    What does the task cost now? Hours per week, errors per month or lead time. Without a baseline, every later result is an opinion.

  3. Choose the route

    After a Discovery call of 45 minutes you decide: go or not. If you go on, you choose a Startsprint, a Project or the ongoing partnership.

  4. Build the first version

    On your data, for one task, with a human who checks the outcome. Then you measure again.

03 / When not

Sometimes AI is not the answer.

Then we say no in the Discovery call, honestly, with advice on what you can do instead.

We say no when the task rarely comes back, when nobody wants to digitise the data, when nobody decides, or when the process itself is the problem. WE SAY NO WHEN The task rarely comes back Nobody wants to digitise the data Nobody decides The process itself is the problem
04 / Further reading

Next: brain or app, and no islands.

Article

A brain first, or an app first?

You do not have to start with the Brain. Here is when you should.

2 min readRead the article →
Article

From vibe-coded tests to scalable systems

How you spot an island, and how a system grows with your documents, agents and people.

2 min readRead the article →
Discovery

What does a solid AI foundation bring your business?

We look at your processes together and identify the first place where AI pays for itself.

Get in touch