30 to 40 billion dollars. That's how much companies worldwide spent on generative AI in 2025. The result, according to the MIT study "The GenAI Divide: State of AI in Business 2025": 95% of these pilot projects deliver no measurable business value. No revenue growth. No demonstrable ROI. Nothing that shows up on the balance sheet.
The obvious conclusion: AI is overrated. A hype that will fade soon enough.
The honest answer: no. The problem lies somewhere else entirely. And once you look closely, it actually makes a lot of sense.
The cause has nothing to do with the technology
When AI projects fail, most people suspect one of three culprits: not enough budget, not enough technical know-how, or the technology itself simply not being mature enough.
The MIT researchers checked all three. They reached a different conclusion.
Over 80% of the companies surveyed have tested ChatGPT or Microsoft Copilot. Only 40% actually use these tools in production. And even within that 40%, the effect was modest: individual employees became more productive, but no measurable impact on company profit.
The reason is pretty simple. Generic AI tools are flexible. They can do a lot. But they don't know your processes. They don't learn from your workflows. They remember nothing beyond the current conversation. They don't know that your invoices follow a specific format, that your customers fall into three categories, or that your warehouse has a special case only two employees really understand.
Generic tools solve generic problems. Your business doesn't have generic problems.
What the 5% do differently
The companies where AI actually moves the needle take a different approach. They don't buy the tool everyone is talking about. They first figure out where in their business time, money, or nerves are being lost, then build a solution that targets exactly that spot.
Suppose you run a mid-sized company. Sales receives daily inquiries by email, some as PDFs, some as plain text. One employee spends ninety minutes every morning transferring these inquiries into the CRM and routing them to the right department.
A generic AI can help a little here. It can summarize text or suggest replies. But it doesn't know your CRM. It doesn't know which fields need to be filled. It doesn't understand that inquiries from the southern region go to colleague Maier and anything over 50,000 euros goes straight to management.
A custom-built solution does. It reads the email, extracts the relevant data, checks it against existing customer records, creates the entry in your CRM, and routes it to the right person. Ninety minutes a day become ten minutes of review. That's over 300 working hours a year, free to create value elsewhere.
It's not magic. It's just not an off-the-shelf product.
The four conditions where most projects fail
The study points to four things missing in the 95% of failed projects:
- A clear business case. Not "we're doing something with AI now," but: where exactly do we save time or money, by how much, and how do we measure it?
- The right infrastructure. Where does the data live, who has access, how is the whole thing maintained?
- The AI adapts to your processes, not the other way around. Your business has worked a certain way for years. The AI should fit that, not force your team to adapt.
- Realistic expectations. No AI replaces an entire department overnight. But it can take out specific bottlenecks reliably.
Companies that take these four points seriously end up almost automatically in the 5% group where the investment pays off.
Why this is different from two years ago
Back in 2023, custom software with AI integration was expensive. Really expensive. Today the picture has changed. Modern development tools and stronger models have lowered development costs significantly. What used to be a six-figure project can often be done for a fraction of that.
At the same time, ongoing SaaS costs have gone up rather than down. If you pay monthly for half a dozen licenses, three of which only partly do what your business actually needs, the annual total often makes a custom solution economically attractive.
The math has gotten simpler. And the risk of falling into the 95% trap can be avoided if you ask the right questions from day one.
The next step
If you want to find out whether your business has a concrete use case that would genuinely pay off with a custom AI solution: book a free initial consultation at BleyIT.com/de/contact. 30 minutes where we go through your processes and give you an honest assessment of where the effort is worth it and where it isn't.