Five steps from idea to scale

Why most SMEs use AI but have no AI strategy

Only 21% of German companies have an AI strategy, even though 40% are already using AI in some form.¹ Using AI is not the same as having a strategy.

The rest buy licences because a competitor is supposedly doing something with AI too. Individual departments experiment because management likes the idea. A few ChatGPT accounts here, a Copilot licence there. Six months later, whoever controls the budget asks what it actually delivered.

Nothing measurable. Because tools without a plan do not produce results.

A good strategy defines a specific problem, sets out a path, and names measurable actions. Here is what that looks like in practice, in five steps.

Step 1: AI readiness check

Before you spend a single euro on AI, you need to know where your company stands. Not technologically. Organisationally.

Map your shadow AI. 78% of your staff are already using AI, whether you allow it or not.² During an AI readiness check in one of my workshops, we found 104 different AI tools in use at a mechanical engineering company, everything from ChatGPT to Suno.AI to Manus. Marketing writes social media posts with AI, sales generates email drafts, HR screens applications. All of it uncontrolled, without governance, without a compliance check.

GDPR violations cost up to 4% of annual turnover. EU AI Act violations go up to 7% of global turnover.³ Banning it will not solve the problem. Structuring it will.

Assess your data quality. Yes, your data is probably not perfect. That is normal. But "we need to clean up our data first" is an excuse to avoid starting. A first proof of concept (PoC) often works fine with small, idealised datasets. Most inconsistencies only surface once the project is underway, and that is the right moment to deal with them.

Check team readiness. AI only works if people can get involved, contribute ideas, and understand where it is going. If half the team is worried about their job and the other half says "we have always done it this way", you do not have a technology problem. You have a change problem.

The readiness check is not a 200-page report. It is an honest stocktake that takes one to two weeks. What tools are already in use? What is the state of the data? What is the mood in the team?

Step 2: Identifying the right use cases

Most AI projects do not fail during implementation. They fail because the wrong problem was chosen.

Typical first ideas: "AI should optimise our entire production process." "Automatically analyse all customer data." "Fully automated quality control for everything." Then people are surprised when the project takes 18 months, costs 500,000 euros, and does not work in the end. Almost every company I work with starts out thinking too big.

The right questions look different:

1. What cost your team 90 minutes of unnecessary work in the last two weeks?
2. Which task is repetitive or takes a disproportionate amount of manual effort?
3. Is there a concrete way to use AI to handle exactly this task? Not "AI will just do it."
4. What will you measure to know the use case is working?

The AI Design Sprint solves exactly this problem. In Opportunity Mapping, the team identifies real pain points, not "where should we use AI?" but "what annoys us the most?". Twelve problems get narrowed down to three priorities. IT gets a clear list. The teams see things moving forward.

Start with 10,000 euros, not 10 million. Successful companies start small, learn fast, then scale.

Step 3: AI pilot project with a clear scope

This is where most people fail. Not for lack of will, but for lack of scope.

An AI pilot project needs to be three things: small enough to deliver a result in weeks, not months. Specific enough to define measurable goals. Relevant enough to scale if it works.

The framing session takes the most important pain point and analyses it at department level: one process, measurable problems, measurable goals. Not "we want to use AI in manufacturing" but "AI should detect defects at goods receipt and flag the delivery on the dashboard." Or "AI should categorise incoming customer enquiries and route them to the right team." A specific scope beats a vague wish list.

In AI Concept Development, the team then builds out the AI solution itself. What does the AI get as input, and from which data source? What format should the output be in? Who does the result go to, a person or another system? This is also where I kill ideas that are too expensive, too slow, or technically not feasible. Sounds harsh. Saves money and awkward conversations with the board.

The PoC has to be self-contained and time-boxed. No production-ready architecture, no year-long projects. CSV files, idealised data, or a quick database are enough. What comes out at the end is a working prototype that shows whether the idea holds up. Not a strategy paper. A result.

I walk through exactly how this works in practice, step by step, in my article on the AI Design Sprint.

Step 4: From prototype to scalable system

You have a working PoC. Now comes the point where a lot of people get stuck, because the road from PoC to a production-ready MVP is long.

ROI can be measured on several levels. Pilot ROI answers: did the PoC prove feasibility? This is usually measured as "knowledge gained." Scale-up ROI asks: how much do I need to invest, and what does the application contribute to my business goals? Is there a non-AI alternative that performs better? Post-implementation ROI shows whether the model holds up long term and at what point it turns a positive ROI.

In my workshops, SMEs often expect a fourfold return on their investment within a year. That is ambitious. AI is an organisational change, not a clean-cut business case like a new machine on the shop floor. The value often shows up indirectly: employees quit less often because their work gets less frustrating. Throughput goes up because bottlenecks get caught earlier. Customers stay because their requests get handled faster.

Scaling depends on cross-functional collaboration: manufacturing, IT, sales, and leadership at the same table. AI is not an IT project. Treat it like one, and you will find out that IT alone neither understands nor can solve the problems that live in the business units.

AI strategy for SMEs: don't forget governance and training

AI without governance is like driving without a licence. It works for a while, until it doesn't.

Get compliance right. The EU AI Act is here. The AI competence requirement under Article 4 has applied since February 2025. From August 2026, the full requirements for high-risk AI kick in.³ Your company needs to know which AI applications fall into which risk category. GDPR compliance has to be built in from the start, not bolted on afterwards.

Get training right. AI training is not a ChatGPT tutorial. Employees learn prompting, maybe advanced prompting if they're ambitious. That is not enough for real enablement. Real enablement starts where teams understand which processes actually suit AI, where it makes no sense, and where the limits and risks are. Employees don't need to become developers. But they do need to be able to judge what makes sense.

Keep the knowledge in-house. Traditional consulting creates dependency. Consultants show up, do the work, take the knowledge with them, and leave. Or they stick around on retainer with endless PowerPoint blueprints. I train your people to become the AI experts in their own departments. Then I leave. No retainer, no embedding. After the AI Design Sprint, teams can identify, evaluate, and plan use cases as a PoC on their own.

Document and communicate your policies. Which AI tools are allowed, and in what contexts? Which data can go where? Who is responsible? An AI policy nobody knows about is not a policy.

What sets a strategy apart from buying a licence

Stop treating AI like a shopping list. Stop waiting for the perfect model. The next big release will not solve your problems if there is no clear plan behind it.

What you actually need fits on a napkin: a problem, a concrete idea, a prototype. Everything else comes after.

An effective AI strategy for SMEs does not start with technology. It starts with a decision: for a concrete approach, with a clear use case, with the people who do the actual work rather than against them.

Companies that use AI operationally today do not have better technology than you. They stopped planning sooner and started taking a concrete first step. That is the difference.

What is your first concrete step?

Your next step

Want to know where your company stands on AI readiness and what to do next? Get in touch, or take a look at how the AI Design Sprint takes you from problem to prototype in 3 to 4 days.

¹ HKA/KARL Study 2025: AI in German SMEs, Karlsruhe University of Applied Sciences, Prof. Dr. Steffen Kinkel (517 companies surveyed), h-ka.de
² Microsoft Work Trend Index 2024: AI at Work, microsoft.com
³ Regulation (EU) 2024/1689 (EU AI Act), Article 4 (AI competence) and Article 99 (penalties of up to 7% of global annual turnover); Regulation (EU) 2016/679 (GDPR), Article 83 (penalties of up to 4% of global annual turnover), eur-lex.europa.eu

Back to overview