Why do most AI projects in mid-sized companies fail?
Short answer: it is not the AI. It is where you put it. Most companies stick a chatbot or a copilot on top of the organisation they already have, with the same messy data and the same manual steps underneath. A copilot like that can certainly help, a handy assistant, a few quick wins. But it does not make your organisation ready. The AI works exactly as well as the chaos it runs on.
MIT NANDA, The GenAI Divide, 2025
of enterprise AI pilots deliver no measurable result on the profit and loss account.
Only 5% turn it into real revenue. The rest stay stuck in a pilot that goes nowhere.
And it is not that nobody is trying. McKinsey measures that 88% of companies now use AI, while only 6% get a serious profit impact out of it. Everyone has the tools. Almost nobody has the result.
The Struon thesis in one sentence: real AI transformation rebuilds the organisation, it does not build on top of it. Bolt-on AI on an old organisation gives you an expensive pilot. AI built into a rebuilt organisation gives you a streamlined company. That difference is the whole story, and the rest of this piece shows where it sits.
The real problem: bolt-on AI on an old organisation
Take a transport company. Planning sits in one system, the bookkeeping in another, orders arrive by email and get retyped by hand. Now you stick an AI copilot on top. What do you get? A smart assistant that politely joins the same chaos. The data is half correct, the processes stay manual, and at best the AI hands something over a little faster that still has to travel through three systems afterwards.
That is bolt-on AI. There is nothing wrong with it for a few quick wins, and as a handy assistant it does its job. But you are putting intelligence on top of an organisation that was never built for it, which is why it stalls the moment you want a real result out of it.
S&P Global Market Intelligence saw this clearly in 2025: 42% of companies abandoned most of their AI initiatives that year, up from 17% a year earlier. On average an organisation failed to move 46% of its proofs of concept into production. Not because the AI did not work, but because it had nothing to land on.
Bolt-on AI can work perfectly well as a smart assistant. It only falls short as a lever for your whole organisation, because you are putting intelligence on top of chaos and expecting it to solve the chaos.
The biggest misconception in mid-sized companies right now: treating AI as a tool you switch on, rather than a transformation you carry out. A tool changes nothing about your organisation. A transformation changes the organisation itself.
The real lever is data and process, not the model
This is where it gets concrete. Most people assume a better AI model makes the difference. That is the part that matters least. The model has become a commodity, everyone can reach the same models.
What does make the difference sits underneath: whether your data is in order, and whether your processes have been redesigned around what AI is actually good at. McKinsey is sharp about this in 2025: the strongest predictor of profit impact is fundamentally redesigning workflows. Not more models, not more tools. Rethinking how the work flows. Companies that transform are 3.6x more likely to be doing that deeper work than companies running pilots alone.
Translated to your company that means the following. A quote that currently passes through four pairs of hands and three systems is not something you "speed up with AI" while those four pairs of hands remain. You rebuild the process so the quote emerges in one flow, with AI in the places where it genuinely takes work over, on data that is correct. Only then does the organisation move faster and cost less.
That is also why Dutch mid-sized companies should pay attention to timing. Statistics Netherlands measures that in 2025 29.8% of small and mid-sized businesses used AI, against 66.2% of large companies. The mid-market is behind, and the companies that get the transformation right now build a lead while the neighbour is still trying a copilot that will stall within six months.
What real AI transformation looks like in practice
Not as a tool you switch on. As a method you work through. At Struon that method is called Shadow Rebuild: you rebuild the back office as owned software plus AI agents, while the old system keeps running (the shadow), and you only cut over once the new version is proven better. No big bang, no risk of standstill.
- Diagnose. Where the duplicate work sits, and what it costs per year.
- Knowledge extraction. Capturing how the work actually flows, including what only lives in people's heads.
- Rebuild. Owned software plus AI agents, on data that has been cleaned up.
- Shadow cutover. The new runs alongside the old until it is proven better.
- Run. The old system is switched off.
The outcome is called Streamlined: an organisation that runs calmer and faster, with 30 to 50% lower back office costs. The cost saving is not the goal you chase, it is the consequence of an organisation that no longer retypes the same thing four times.
Bolt-on AI versus real AI transformation
The difference along the axes that matter in practice:
| Dimension | Bolt-on AI (copilot on top) | Real AI transformation (Shadow Rebuild) |
|---|---|---|
| Where the AI sits | On top of the existing organisation, as a layer | Built into a rebuilt process, in the foundation |
| Data | The old, messy data stays as it was | Data is cleaned and restructured first |
| Process | Manual steps remain, AI makes one of them faster | The process is redesigned around what AI can do |
| Risk | Pilot stalls, 95% reach no result on the books | Shadow cutover: the old version runs on until the new one is proven better |
| Cost | An extra licence on top of the existing stack | 30 to 50% out of the back office, because duplicate work disappears |
| Result | An expensive pilot, little that is measurable | A streamlined organisation that moves faster |
| Who owns it | The software vendor, so a new dependency | The company itself, free of half-used SaaS |
The short version of that table: bolt-on AI makes your old organisation slightly faster and more expensive. Real transformation rebuilds your organisation.
This is also why a copilot subscription never becomes your competitive advantage: your competitor can buy exactly the same subscription. What they cannot simply copy is a back office rebuilt around your work.
It is not the AI, it is where you put it
The 95% that fail did not pick the wrong tools. They put the right tools in the wrong place: on top of an organisation still running on retyping and disconnected systems. The 5% that do get a result rebuilt the work underneath. That is the entire difference, and it is exactly the choice now open to every mid-sized company that wants to get further than a pilot.
In a no-obligation call we look together at where the real lever sits in your organisation, before anything gets stuck on top of it. We map out where the duplicate work and the silent costs are, and what a Cost-Stack X-ray puts against that in concrete euros. No commitment, just an honest picture of where you stand.
The core
95% of AI pilots do not fail because of the AI, but because it gets stuck on top of an old organisation. Real transformation rebuilds the organisation underneath.
Veelgestelde vragen
Why do AI projects in mid-sized companies fail?
Because the AI gets stuck onto an organisation that is not set up for it: messy data, manual processes, disconnected systems. The MIT NANDA study (2025) found that 95% of pilots produce no measurable result on the books. The cause is structural, not technical. The AI only works as well as the foundation beneath it.
What is the difference between real AI transformation and bolting AI on top?
Bolting AI on top leaves your organisation as it is and adds intelligence over it. Real AI transformation rebuilds the organisation itself: data, processes and software again, with AI in the foundation. The first gives you a pilot, the second a streamlined company. McKinsey (2025) calls redesigning workflows the strongest predictor of profit impact.
Is AI only for large companies?
No, but large companies are ahead. Statistics Netherlands (2025) measures 66.2% AI use among large companies against 29.8% in the mid-market. That gap is exactly what makes the timing relevant: the mid-sized company that gets the transformation right now builds a lead while most competitors are still trying tools.
Is it the AI model that makes projects fail?
Almost never. The model is a commodity, everyone can reach the same models. The lever sits in data and process. According to McKinsey, companies that fundamentally redesign their workflows are 3.6x more likely to be the ones getting real profit out of it.
What exactly is an AI transformation?
Real AI transformation is rebuilding the organisation with AI in the foundation, not adding AI tools to the organisation you already have. At Struon that happens through Shadow Rebuild: diagnose, extract knowledge, rebuild as owned software plus agents, shadow cutover, run. The outcome is a streamlined organisation with 30 to 50% lower back office costs.