TL;DR
AI projects are never “done” in the traditional sense — the pace of innovation outpaces any implementation. Instead of one-time projects, companies need continuous evolution: honest process documentation, a culture of experimentation, and a partner who can contextualize new developments. The question is no longer “When are we done?” but “How do we get better every day?”
Contents
- Why is the pace of AI a problem?
- What does this mean for ongoing projects?
- Where should companies start?
- What’s the first practical step?
- What does the new way of working look like?
- FAQ
Why is the pace of AI a problem?
In short: AI innovation is so fast that traditional project planning no longer works. While you’re implementing, the technology is already outdated.
With the steam engine or the combustion engine, entrepreneurs had decades. A loom from 1820 still worked in 1850. A database from 1995 was still state of the art in 2005.
AI is different. The major players — OpenAI, Anthropic, Google, Meta — are in a race where new capabilities emerge every few months. Imagine: you invest time and budget in an AI solution, roll it out, train your employees. Then the next generation arrives — and can do everything out-of-the-box that you spent months building.
This isn’t an exaggeration. It’s the current normal.
What does this mean for ongoing projects?
In short: Many companies continue working with suboptimal solutions because they’ve already invested. That’s understandable — but expensive.
A colleague described the situation aptly: It’s like a bar fight. You can’t step outside, have a beer, and wait until it’s over. Because it won’t be over for years.
The classic reactions are understandable:
- “We’ve invested so much, we have to keep going” (Sunk Cost Fallacy)
- “We can’t change everything again” (Change Fatigue)
Both lead companies to work with solutions that are better than nothing — but far below what’s possible today.
Where should companies start?
In short: Not with technology, but with understanding how the company actually works.
In conversations with decision-makers, I often hear: “Where do you think AI could help us?” The most common answer: “No idea — you tell me.”
That’s not failure. It’s honest. And the right starting point.
The use cases that worked elsewhere — knowledge management, process automation, intelligent documentation — don’t automatically fit every company. Even if they theoretically fit: maybe they’re not the current bottleneck. A “nice to have” instead of a real lever.
What’s the first practical step?
In short: Document how your company actually works — not how it’s supposed to work according to the manual.
Before deploying AI, you need clarity about your lived processes:
- Not the official documentation from quality management
- Not the role descriptions from HR
- But the informal agreements, the handoff points that “just work that way,” the arrangements that aren’t written down anywhere
In many companies, business processes are surprisingly poorly documented — yet AI agents need to understand and follow these processes. That’s not due to laziness. People are good at using implicit knowledge: The sales rep knows that customer X is easier to reach on Thursdays. Accounting knows which invoices have priority. This knowledge exists — but it’s nowhere to be found.
Concrete action: Sit down with key people and ask: “How does this really work?” — not “How should it work?”
What does the new way of working look like?
In short: Permanent evolution instead of one-time projects. “Done” is a warning sign, not a goal.
What you need instead:
1. Acceptance of chaos You will optimize permanently. That’s not a bug — that’s the new way of working. “Done” used to be desirable. Today, “done” means you’ve stopped learning.
2. Continuous accompaniment Someone who accompanies the process permanently, with an outside perspective. An external consultant or an internal champion — whose explicit job is to recognize new possibilities.
3. Culture of experimentation The willingness to introduce things, question them, keep going. The knowledge that “done” doesn’t exist — only “current state.”
FAQ
How long does an AI implementation take?
There is no “done.” Expect a first usable result in 4-8 weeks, but plan for continuous development. The question shouldn’t be “When are we done?” but “How quickly do we see initial value?”
Is AI investment worth it if everything becomes outdated so quickly?
Yes — but differently than before. Don’t invest in “finished solutions,” but in learning capability. Build competence, document processes, create data foundations — as shown in our case study on sales automation. These assets remain valuable even when the tools change.
We just introduced an AI tool. Switch again already?
Not necessarily. Ask: Does it solve the problem? Do employees use it? If yes, keep going. But: Stay open to improvements. “We’ve invested” is no reason to stick with an inferior solution.
Do we need an internal AI expert?
Not necessarily a full-time expert, but a “champion” — someone who feels responsible for evaluating new possibilities. This can also be an external partner who checks in regularly.
What does it cost to do nothing?
Your competitors who start now are building an advantage. Not because they have perfect AI — but because they’re learning while you wait. The learning curve starts with the first step.
Conclusion
AI projects will never be “done” in the traditional sense. That’s not bad news — it’s an invitation to think differently.
Away from one-time projects with clear beginnings and ends. Toward permanent evolution, where improvement becomes part of daily work. With a partner who accompanies the journey and helps set the right priorities.
The question is no longer “When are we done?” — but “How do we get a little better every day?”
This article is based on a conversation between Manuel Zorzi and Michael Kirchberger about the reality of AI implementation in the German Mittelstand. Watch the full podcast on YouTube →