Summary: Successful AI adoption doesn't start with "which tool?" but with "which process costs us the most time?" A focused pilot with a measurable result beats any strategy presentation - and builds the buy-in for everything that follows.
Almost every company is dealing with artificial intelligence by now. Most in the same way: a workshop, a tool list, and then an odd silence. In AI consulting at Luftschloss, we see what it hinges on – and what works instead.
The most common mistake: starting with the tools
The question "should we use ChatGPT, Claude, or something else?" is the wrong first question. Tools change every quarter. Processes don't. The right first question is: which task costs our team the most time, follows clear patterns, and has a verifiable outcome? That's the candidate to start with – not whatever's being demoed at conferences right now.
Start small, mean it
A good AI pilot has three properties:
- Tightly scoped: one process, one team, one measurable goal. Not "AI in sales," but "first drafts of proposal text in half the time."
- Four to eight weeks: long enough for real results, short enough to learn without losing face.
- With the people who do the work: AI adoption rarely fails on the technology, often on acceptance. Whoever helps shape the tool also uses it.
Tool, workflow, or your own software?
After the pilot comes the scaling question. Roughly: standard tools for standard tasks (text, research, images), configured workflows for recurring processes – and your own AI agents or custom software where the process is company-specific and means a competitive advantage. That line keeps moving down: what was custom development last year is often configurable today.
The factor everyone underestimates: data and context
An AI tool is only as good as the knowledge it can access. The reason the same AI delights one company and disappoints another rarely lies in the model – it lies in whether it has access to the right information: proposal templates, product data, past projects, tone of voice. Before thinking about bigger build-out stages, it's worth the sober question: is our relevant knowledge even available in a form an AI can work with – or is it stuck in people's heads, PDFs, and email threads? Often, cleaning up that foundation is the real first step, not the AI itself.
Setting expectations right
The second most common cause of failed AI projects isn't technology but miscalibrated expectations. Whoever believes AI will replace entire departments starting next week will be disappointed and switch it off. Whoever understands that AI first saves 20 to 30 percent on a concrete task, and that this saving grows with every iteration, scales sustainably. Realistic goals aren't a sign of timidity but the precondition for a pilot to run long enough to have an effect at all.
How to recognize credible consulting
Good AI consulting doesn't want to sell the biggest project, it wants to sell the right one. Warning signs: buzzword density, tool sales disguised as consulting, and three-year roadmaps in a field that reinvents itself every three months. Good signs: process questions, pilot proposals with metrics, and the willingness to say where AI (still) doesn't help.
Conclusion
AI adoption is process work with a new tool, not magic. Whoever starts with a tightly scoped, measurable pilot and turns the people affected into participants has achieved more after eight weeks than with any strategy presentation.