AI Build vs Buy: Why SMEs Invest Incorrectly

KI-Strategie · 21. März 2026 · Mohsen Ghulami

AI Build vs Buy in SMEs: Most burn money with wrong strategies. Learn why the hybrid approach is the only solution.

Last week, I sat down with the CEO of a mechanical engineering company from the Teutoburg Forest. A classic hidden champion, 250 employees, world market leader in a niche you've never heard of. He told me, with a mix of pride and despair, about his 'lighthouse project': a self-developed AI for predictive maintenance of his facilities. Budget: 1.5 million Euros. Duration: two years. Result: a half-baked solution that produces more false alarms than a nervous watchdog and is simply ignored by the maintenance technicians on the front line – let's call them Master Kaczmarek. The entire 'AI: Build vs. Buy' debate in German SMEs is a farce. And I'll tell you why: both paths, as they are mostly pursued today, lead directly to a dead end.

Let's be honest: in boardrooms, a dangerous mix of panic and a desire for recognition is rampant. They want to 'do something with AI' because the competitor next door is already boasting about it. So, they put the cart before the horse. Some fall for the 'Build' craze. Driven by the romantic notion of developing the holy grail – a proprietary AI that pulverizes the competition. That's the engineer's soul coming through, I know it. They want to create the thing themselves. The problem: you are not Google or OpenAI. A complete in-house development of an AI solution devours millions (according to W&P analyses, we're talking 5-10 million Euros over three years for an SME project) and carries a gigantic risk. The success rate? A sobering 30-50 percent.

And the other side? They succumb to the siren song of the 'Buy' faction. Let's just buy standard software from one of the big providers, then we'll be on the safe side. Sounds tempting, doesn't it? Fast, supposedly cheap, scalable. The problem: these 'off-the-shelf' solutions are often as flexible as a railway track. Fraunhofer IPA proved this in black and white in a 2026 study (yes, they're looking further ahead): in 70 to 80 percent of cases, standard AI models fail due to the complexity of real production processes, for example, in forecasting energy load peaks. That's like trying to control a highly complex, individualized production process with an app from the App Store. You can do it – but it's just shoddy work.

AI Build vs Buy: The Uncomfortable Truth in Numbers

Let's stop lying to ourselves. The decision for an AI strategy is not a matter of faith; it's hard-nosed business economics. I've looked at the latest analyses from Fraunhofer IPA and the strategy consultancy Wieselhuber & Partner (W&P). And the numbers – they're a slap in the face for every dreamer.

Look at that. Pure in-house development is a financial suicide mission with a ridiculously high error rate. Pure Buy is a lottery where the blank (lack of fit) is more likely than the jackpot. This is not an opinion; these are facts from analyzed projects. And this doesn't even include the follow-up costs for maintenance and the frustrated employees tormented by half-baked tools. We're not talking about peanuts here, but about investments that can shake an SME to its core. There's no getting around it.

AI is a driver and enabler; start with pain points and clean data – then projects experience an AI boost instead of getting bogged down.

— Volker Riedel, Partner at W&P Strategy Consulting

I spoke with Volker Riedel from W&P last week, and he hit the nail on the head. The whole discussion is set up incorrectly. It's not about the technology. It's about the problem. And it's about the raw material – the data.

But... what about the unique competitive advantage?

Now I hear the objection from the tech corner: 'But Mr. Müller, if we only assemble standard components, where does our USP go? Our unique market advantage?' This is the strongest – and at first glance most plausible – argument for the 'Build' approach. The dream of creating an AI so perfectly tailored to one's own secret production recipe that the competition can only stand by in awe.

This is a dangerous fallacy. In my experience, 9 out of 10 companies dramatically overestimate the uniqueness of their processes. The true, insurmountable competitive advantage today rarely lies in the algorithm itself (many of which are open source or easily replicable), but in two things: the unique – and clean! – data you have collected over decades, and the deep integration of AI logic into your core processes. And this is precisely where the hybrid approach comes in. Buy the standard building blocks – data connectivity, visualization, basic algorithms. But build your own highly specific logic on top of that, fed by your expert knowledge and your unique data. This is not capitulation. This is simply clever.

What I see out there: An AI Strategy from the Ivory Tower

When I walk through the factory halls across the country, I see a recurring pattern. Upstairs, in the executive suite, a PowerPoint battle is waged over 'Digital Transformation' and 'AI-First Strategy.' Budgets are approved, consultants are hired, and lighthouse projects are announced. Downstairs, on the factory floor, where value is created, Master Kaczmarek shakes his head. Why? Because the expensively purchased or painstakingly self-developed AI is fed with data that slumbers in twenty-year-old Excel spreadsheets, in proprietary PLC controls from the nineties, or – no joke, I've seen it myself – on the personal laptop of a master who will retire in six months.

The thing is: without a central, clean, and accessible data basis, any AI strategy is like trying to fuel a Porsche with mud-contaminated heating oil. It won't just not drive fast; it will break down. The success of companies like REWE, which were able to reduce their material costs by 10-20% through the integration of standard AI solutions, is not based on a magic algorithm. It is based on the hard work of getting their data under control. Clean data is the true raw material of digitalization. Everything else is expensive fortune-telling.

Do you even know your customer?

And that leads me to an even more fundamental problem. Many companies invest millions in optimizing internal processes with AI, but don't even know exactly who their ideal customer is. Before you put a single euro into an AI project, you need to define crystal clear who you are doing all this for and what problem you are solving for this customer. Otherwise, you are optimizing into a void.

Amplifa ICP Playbook: Find Your Ideal Customer Sharpen your Ideal Customer Profile (ICP). The foundation for any successful sales and product strategy – and the prerequisite for a meaningful AI investment.

Once the ICP is established, it's about understanding the market. How big is the potential really? For which segments is the effort of AI-supported personalization worthwhile?

Amplifa TAM Analysis: Realistically Assess Market Potential Stop guessing. With our Total Addressable Market (TAM) analysis, you quantify your market potential and identify the most lucrative target segments for your AI-driven expansion.

Enough with the War Games: What Needs to Happen Now

Enough analysis. What does this mean specifically for you as a CEO, CTO, or Head of Digital in an SME? It means you have to stop chasing buzzwords and instead roll up your sleeves. Here's your five-point plan – call it the 'Müller Manifesto for Practical AI':

1. Step 1: Radical Pain Point Analysis instead of Technology Infatuation. Lock up your IT people and AI evangelists for a day. Instead, bring the production manager, sales manager, and Master Kaczmarek to the table. Ask: 'Where are we bleeding money? Where are we losing time? What problem keeps us awake at night?' Identify the three biggest pain points. Only for these should AI even be considered.
2. Step 2: Brutal Data Inventory. Immediately. Impose a three-month moratorium on all new AI projects. Use this time to accomplish one single, unshakeable task: getting your data in order. Where is it? How current is it? Who has access? Create a 'Single Source of Truth'. This is the most unattractive but most important task of the entire digitalization.
3. Step 3: Define the Hybrid Model as the Gold Standard. Based on your pain points, evaluate according to the W&P framework: What is our 'secret recipe' (unique process data, expert knowledge), and what is standard (data connectivity, visualization)? Buy the standard components from established providers. But insist on open interfaces (APIs). On this foundation, you then build the crucial 10% of custom logic that makes the difference, with a small, specialized team or an external partner.
4. Step 4: Pilot Project with Built-in Exit Strategy. Choose the smallest of the three pain points and start a clearly defined pilot project. Budget: maximum 100,000 Euros. Timeframe: maximum six months. Define tough KPIs beforehand (e.g., 'Reduce false alarms by 50%'). If the goals are not met after six months, the project is buried. No discussion. The biggest danger is the sunk-cost fallacy – holding on to a failing project because so much has already been invested.
5. Step 5: Consider Compliance from Day 1. The EU AI Act is not coming maybe, it's coming for sure (from 2027 it will get serious for mechanical engineers). Ignoring regulations is not a minor offense, but a potential showstopper. Use the templates and experiences from projects like the KIRR Real real lab in Baden-Württemberg. They have already shown how to implement legally compliant AI solutions by combining validated buy elements with a custom-validated integration. This reduces risks by up to 50% according to researchers there.

And Don't Forget the Human Element

The smartest AI is absolutely useless if your sales team doesn't understand how to convert the added value it creates into euros and cents for the customer. An AI that accurately predicts delivery times is only an advantage if your salesperson can sell it as a guarantee. Technology is just the enabler. Humans still have to sell.

Amplifa Sales Coaching: Turn Your Team into a Revenue Machine Technology doesn't sell, people do. Our coaching transforms your sales team so they can confidently position the value of your innovations – including your AI solutions – with the customer.

Frequent Questions about AI Strategy in SMEs – Plain Talk, Not Consultant Jargon

What are the real costs of an AI project?

Forget the glossy brochures. For in-house development (Build), expect at least 5 million Euros over the first three years, including horrendous maintenance costs. A pure purchase (Buy) might start at 500,000 Euros, but the risk of it not fitting and you writing off the money is enormous. The realistic and most successful path – the hybrid approach – usually lies between 1 and 3 million Euros, but has the decisive advantage of a positive ROI often within 12 to 18 months.

I have old machinery. Can I even retrofit it with AI?

Yes, absolutely. And that's often the smarter way. The keyword is 'Retrofit'. Companies like IBHsoftec have developed frameworks to connect even older machines without modern interfaces via standards like OPC-UA for data. This is often 50% cheaper than a complete new purchase and can, as examples at Siemens in Amberg show, lead to energy savings of up to 70%. Your old machines are not a burden – they are a goldmine of historical data if you tap into them.

Is 'Pure Build' ever a good AI strategy for SMEs?

In very, very rare exceptional cases. If your core business is based on an absolutely unique dataset that no one else in the world has – for example, the analysis of specific material properties in a secret chemical process – then complete in-house development can make sense. For 99% of SME applications (production planning, quality control, predictive maintenance), however, it is financial and technological overkill. Concentrate your developer power instead on perfectly adapting a standard solution to your process.

I bet that in three years we won't be talking about 'Build vs. Buy' anymore. We'll be talking about 'Smart Adapt vs. Dumb Install'. The winners will be those who have understood that AI is not a product you buy, but a capability you have to acquire – built on a solid foundation of clean data and standard technology.

So the question is not 'Build or Buy?'. The real question you need to ask yourself today is: Do you have the courage to do your homework first before throwing your hard-earned money at the next AI charlatan? Discuss with me on LinkedIn or send me an email. I look forward to your urgent letters and success stories.