Why do AI pilots get stuck, and how does AI get into daily business?
Pilots get stuck because they start as a trial next to daily work: without a business goal, without an accountable person, without a metric and without a plan for the day after. AI gets into daily business when it is clear from the start which process changes, who is accountable for the result and how success is measured. That is when an experiment becomes operation.
Tool, project, operation: only the third stage counts
AI goes through three stages in a company:
- Tool: individuals use ChatGPT, Copilot or Claude at their own desk. That is usage.
- Project: a team tests a use case. That is an experiment.
- Operation: AI runs as a fixed part of a process, someone is accountable for the result, and it is measured. That is a position in the market.
“We use ChatGPT” is not adoption. The difference is the one between having Excel and having controlling. And an AI project that is no longer running was an experiment. Valuable, but not a position in the market.
Five reasons why pilots get stuck
- No business goal: the test is whether the technology works, not whether something in the business gets better.
- Nobody accountable: the pilot belongs to “IT” or “the project team”, but to no single person who is accountable for the result.
- No fixed process: the pilot runs alongside the actual work. For the team it is extra work, and extra work is the first thing to go.
- Approvals clarified too late: data protection, the works council or IT security only join at the end, and then everything stalls.
- No metric: after eight weeks nobody can say what the pilot achieved. So nobody decides what happens next.
From pilot to operation in four steps
- Build the pilot in the real process: with real cases from daily work, not with test data in a sandbox. Before you start, it is clear which work changes and who is accountable.
- Clear internal approvals: for example with data protection and the works council, at the beginning, not at the end.
- Measure in daily work: compare the metric with the baseline after a few weeks of real use.
- Embed it: describe the new process, onboard new colleagues into it and keep the accountability with one person for good. Only then has the pilot arrived in operation.
How to tell that your pilot is stuck
- Nobody can say who used it this week.
- When asked what it achieved, the room goes quiet.
- There is no date on which the next step will be decided.
- The person who started it is busy with something else.
If several of these apply, an honest look is worth it before the next pilot starts.
Frequently asked questions
Should we stop a stalled pilot?
Decide deliberately: either you set it up again, with a goal, an accountable person and a metric, or you end it and record what you have learned. Both are better than letting it run on.
How long should a pilot run?
Before you start, set a date on which you decide: into operation or end it. A pilot without that date runs into nothing.
Who should own a pilot?
One person from leadership or the business unit, not IT. IT takes care of operations and security, leadership is accountable for the result. AI is not an IT ticket. It's a business decision.
Do we need new people for this?
Not necessarily. If your team wants to do it themselves, external guidance is often enough. If there is no internal capacity, an interim manager can take over the introduction of AI.
More answers
- Where does AI deliver measurable results in sales first, in mid-sized companies?
- How do you choose your first AI use cases?
- Where does AI pay off in marketing for mid-sized companies, and where not?
About the authorSelina Gaertner, AI for Leadership, Sales and Marketing. TÜV Rheinland certified AI Strategy Consultant, more than ten years in sales and marketing in medical technology, co-author of the Amazon bestseller “Fokus Mensch”.
Your next step: AI with Results, the 90-day programme
For the person leading the introduction of AI: twelve weekly steps and live briefings, and at the end your first AI initiative runs in daily work.