Every operator I talk to right now is feeling the same pressure. The board asks about AI. A competitor announces something. A vendor demos a tool that looks like magic. The temptation is to pilot the tool. I understand the temptation, and I am telling you to resist it.
The better move is to audit your data architecture first. Without clean data pipelines, every AI initiative stalls within six months, and I have watched it happen at companies of every size. The model gets blamed. The vendor gets blamed. The real problem is that the AI was asked to reason over data that lived in eight systems, three spreadsheets, and one estimator's head.
When I automated core business processes at a glazing contractor, we eliminated more than 2,000 annual labor hours in scheduling, reporting, and documentation. People hear that and assume the AI was the hard part. It was not. The hard part came first: consolidating eight legacy systems into one ERP so there was a single source of truth for the AI to work from.
That is the risk nobody puts in the demo. The magic in the sales meeting is real, but it was performed on clean sample data. Your job order data, with its duplicate customer records and free-text notes and fields nobody fills in, is not clean sample data.
Here is the order I run with clients, and it will look familiar if you know my framework. Start with the data audit: where does each number live, who enters it, and where does the same fact exist twice with two values? Then fix the pipeline: one system of record per fact, with process discipline around entry. Then, and only then, pick the AI use cases, starting with the boring ones. Scheduling, reporting, documentation. The unglamorous automations pay for everything else.
There is one more piece, and it is the one operators skip. Only 31 percent of workers report receiving any AI-related training while 86 percent of employers expect AI to transform their business. That gap is where initiatives die. Your team does not need to fear the tool or worship it. They need to know what it is for, what it is bad at, and when their judgment overrides it.
Data first. Process second. Model last. Get the sequence right and AI becomes the cheapest employee you ever hired. Get it backwards and it becomes the most expensive pilot you ever wrote off.