Why SMEs Fail Due to the Order of AI
6 min read
41 percent of German companies are actively using AI-up from 17 percent just a year ago. Bitkom’s 2026 figures suggest momentum. But they also gloss over the fact that 59 percent of firms with 20 or more employees are still hesitating. The reason rarely lies in the technology itself, but in the order of implementation.
Key Takeaways
- Tool-first is the costliest mistake. Starting with software instead of the use case means buying a subscription and hoping for the best. The Bitkom survey reveals hesitation isn’t about tech-it’s a structural issue.
- Order matters. Data first, then a concrete use case, then people, and finally the tool. Skipping any step before that will come back to haunt you in operations.
- Data protection is solvable, not a roadblock. 50 percent cite data security as a concern. Most of these worries dissolve with one clean upfront question: Which data is even allowed into the model?
Related:Generative AI in SMEs: Why the 78-percent figure is misleading / When AI tools suddenly eat into margins
Why a tool-first approach fails in SMEs
What is AI enablement? AI enablement refers to the organizational groundwork that precedes tool implementation: a clear data foundation, a concrete use case, trained employees, and a process to validate output. Without this foundation, even the best software remains an expensive experiment.
The sequence in most SME projects is reversed. Someone sees a demo, the tool impresses, and the subscription is signed. Only then does the search for a problem the software could solve begin. This is the costliest approach because the license is already running before it’s clear whether it will deliver value.
The 2026 DIHK survey of nearly 5,000 companies is unequivocal on this point. Firms want to use AI but cite data protection, lack of skilled staff, and missing structures as the biggest hurdles. None of these are software features. All three must be addressed before the purchase-or not at all.
The Order That Stands the Test of Operations
A successful rollout flips conventional logic on its head. It doesn’t start with the tool, but with the groundwork that never comes up in sales pitches.
First, the data. Before any tool, the question is which data can even be used-and what condition it’s in. A sales team storing proposals across four different repositories won’t get clean answers from any AI. This inventory takes days, not weeks, and it determines everything that follows.
Second, the use case. A single, tightly defined application beats platform ambitions every time. Double data entry in procurement, approval bottlenecks in the back office, delayed quotes in sales-these pain points are measurable and can be adjusted in two to four weeks. They deliver the numbers that justify the next project.
Third, the people. The tools cost between 5,000 and 25,000 euros per year for 50 employees. What decides success isn’t the price, but the training. A team that’s never learned how to give an AI a precise task will only produce faster nonsense with the most expensive model.
Fourth, the tool. Only now does the tool decision come into play-and it’s an easy one. Those who’ve clarified their data, use case, and team can instantly tell in a demo whether the software fits. This order removes the risk from the purchase.
Where Data Protection Really Hits a Snag
Half of all companies cite data security as a concern-and rightly so. But often, the worry is too vague when a single concrete question would suffice. Which data can go into which model, under what conditions, and with what deletion period? Answering this cleanly for each use case replaces a diffuse unease with a workable checklist.
What derails a rollout
- Tool purchased before the use case was defined
- Data scattered across four repositories
- Training treated as an afterthought
What makes it work
- A narrow, measurable use case
- Data questions answered per use case
- Training before going live
The reflex to think big costs the most here. A company-wide data protection concept can stall progress for months. A data clearance for a single use case can be resolved in days-and builds the experience needed for the next one.
What This Means for the Next Step
The jump from 17 to 41 percent shows the barrier to entry is dropping. Those still hesitating aren’t waiting for better technology-they’re waiting for a decision no one’s making. The pragmatic way forward is small: one use case, one clarified data foundation, one trained team. It’s unspectacular, and that’s exactly why it works. Companies that truly leverage AI rarely have the best tools. They’ve followed the right order.
Frequently Asked Questions
Where should a mid-sized company start with AI?
Not with the tool, but with a tightly defined use case and the question of which data is already clean and available. Only then does software selection make sense.
How long does it take to implement a first meaningful AI use case?
A narrowly scoped case-like eliminating duplicate data entry or clearing an approval backlog-can typically be adapted in two to four weeks and delivers measurable results.
Is data protection a real obstacle to AI adoption in mid-sized companies?
Rarely a complete roadblock, but often an unresolved question. Defining which data can be used for each specific use case replaces vague unease with a workable checklist.
Why do AI projects fail despite having good tools?
Because the groundwork is missing. Without clear data, a defined use case, and trained staff, even the best model only speeds up existing chaos.
Editor’s Picks
- Three AI setbacks in one week: lessons for mid-sized companies
- Stanford AI Index 2026: Inaccuracy overtakes cybersecurity as top risk
- GPAI becomes a documentation trap for mid-sized businesses
More from the MBF Media Network
Image source: Cover and article images AI-generated (May 2026), C2PA certificate embedded in image