A few months ago I wrote here that dealerships should start with one small, well defined task rather than a full transformation programme. I still think that is right. But starting small only tells you how much to take on. It does not tell you where AI belongs, or what to expect once it is there.
Eurostat reported that 20% of EU companies with ten or more employees used AI in 2025, up from 13.5% the year before. The interest is clearly there. Eurostat also asked the companies that considered AI and then chose not to use it what stopped them. 71% said they lacked the know-how, far more than the 53% unsure where they would stand legally or the 49% worried about data protection, which in Europe usually means working out how GDPR applies to these tools. That same gap exists inside the companies that did go ahead. They just meet it later, when the tool turns out to be something nobody uses.
Closing it is a matter of method, and these are the five steps I work through:
1. Pick the process
The honest test is not which process sounds most suitable for AI, but which one you would fix if you could fix only one.
2. Define the outcome
Name the thing itself, a weekly report, a spreadsheet, a summary of a long document, a drafted message to a customer. Then set the standard it has to meet. If the outcome is a follow up after a service visit, does it always name the work done, what must it never say, and when should the customer receive it.
3. Go through the process one stage at a time
Follow it as it actually runs, not as the manual describes it. With the outcome written down, it is easier to see which stages earn their place. Mark the bottlenecks, the pain points, and the places where the result depends on who did the job that day. Some steps survive only because of a rule that stopped mattering years ago. Remove those rather than automating them.
4. Match what AI does well
AI reads long documents and pulls out what matters, writes in a consistent voice, checks work against a rule you set, and never gets bored on the four hundredth repetition. It is weak at judgement and at anything depending on knowledge nobody wrote down. Set that against the stages you marked. Most of the disappointment comes from asking it to make a decision instead of preparing one.
5. Run it, improve it, repeat
You now have a first version working. Put it to use, then build on it. What your team changes before the work goes out tells you what to fix, whether the tool was missing information or the outcome was defined too loosely. Correct the cause, run it again, repeat. Each round improves the process itself, and what the team learns carries over to the next one. The tool does not.
Which part of the business you start in matters less than people assume. Look for the longest, most manual, most repetitive or most painful processes, wherever they sit, or start where new ideas are already encouraged and a first attempt has room to fail.
So the question I would put to any dealer principal is not which tool to buy. It is whether anyone has gone through one process end to end and written down where it slows, where it hurts and where it goes wrong.
SOURCES
Eurostat, “20% of EU enterprises use AI technologies”, 11 December 2025
https://ec.europa.eu/eurostat/web/products-eurostat-news/w/ddn-20251211-2
Eurostat, “Use of artificial intelligence in enterprises”, Statistics Explained, data extracted December 2025
https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Use_of_artificial_intelligence_in_enterprises