The What-Changed Question: ten questions to ask your shop data before you act

The three questions, in the order the video asks them

Most owners who try AI on their shop data start by asking for a number they already have. The what-changed question starts somewhere else:

1. What changed since last week? Not "what is my average repair order", which your shop system already reports, but which number moved.
2. Which other number moved with it? A falling close ratio means more when declined jobs rose in the same week.
3. Which shop or advisor did it come from? A group average hides the one location, or the one person, where the change really happened.

The answer worth acting on shows where each number came from and when that source last updated.

Who this is for

Owners and managers who have a shop management system full of reports, maybe a second or third location, and an AI tool (or a vendor demo) that produces charts nobody acts on. You can use the questions with any AI assistant that can read your exports, with a business intelligence tool, or with a spreadsheet and a coffee on Monday morning.

What your shop system already answers

Before asking AI anything, know what you already have. Tekmetric's report overview, for example, groups its reports into financial, employee, customer activity and parts categories, including efficiency, average repair order, close ratio and declined jobs. The four numbers Tekmetric calls out are worth knowing cold: average repair order, car count, parts margin and effective labor rate.

If you only want to know where you stand against other shops, that is a simpler job. Tekmetric's Shop Index compares owner-entered average repair order, car count, parts margin and effective labor rate against other shops, anonymously. You don't need AI for that.

What reports don't do well is connect a change in one number to a change in another, across systems and across shops. That is where the questions below earn their keep.

The ten follow-up questions

Ask them in this order. The first three are the ones from the video; the next seven keep you from acting on noise.

1. What changed since last week? List the three numbers that moved the most, up or down, with last week's and this week's values.
2. Which other number moved with it? For each change, name the number that moved in the same direction or the opposite one in the same week.
3. Which shop or advisor did it come from? Break the change down by location, then by advisor or technician.
4. Is this a real change or a data gap? Are all of last week's repair orders closed? Did every feed update? When was each source last refreshed?
5. Has it moved for more than one week? One week is a question. Three weeks in a row is a pattern.
6. Is it the season? Compare with the same weeks last year, if you have them.
7. Did car count change, or only the per-car numbers? Fewer cars and smaller tickets need different fixes.
8. What do we already know that explains it? A holiday, a tech out sick, a big fleet job, a price change.
9. Which open items are tied to it? For a falling close ratio, the declined estimates from that advisor that still deserve a follow-up.
10. What is the one decision this week, and who owns it? Write it down with a name and a day.

A worked example (demo shop)

Here is how the questions might run for Redline Auto, a made-up group with two shops, Elm St and Route 9. The numbers are invented for illustration.

• What changed: the close ratio at Route 9 went from 61 to 48 percent in a week. Elm St held steady.
• What moved with it: declined jobs at Route 9 went from 9 to 17.
• Where: most of the declined estimates came from one advisor, Kim.
• Data check: all Route 9 repair orders were closed and the feed had updated that morning, so the drop was real.
• More than one week: no, this was the first week.
• What we know: Kim had started quoting a new brake package that week.
• Open items: eight declined brake estimates, all under two weeks old.
• Decision: the manager reviews the brake package quote with Kim on Tuesday and follows up the eight estimates by Thursday.

Notice what did not happen. Nobody asked for the group's average repair order, and nobody changed prices across both shops because one number moved at one location.

How to ask an AI tool these questions

• Give it the data, not the conclusion. Export the week's numbers by shop and by advisor, plus the last eight weeks, and ask question 1 first.
• Ask one question at a time. Each answer becomes the context for the next.
• Make it show its work. Ask for the source report and the date range behind every number in the answer.
• Ask what it can't see. "Which data would change your answer if it were missing or late?" is a good last prompt.

Where AI helps, and where it gets in the way

AI is useful in the reading: summarizing a week in plain sentences, spotting that two numbers moved together, lining up several shops or systems in one view, and drafting the next step for a named person.

It gets in the way when it writes confident sentences from incomplete data. If one shop forgot to close repair orders or a feed is two days stale, the summary can still read smoothly. That is why trust is a real barrier: in the 2026 Ratchet+Wrench Industry Survey, 23% of shops not using AI said they were unsure what is trustworthy. An answer that shows its sources is easier to trust, and easier to check.

What to watch out for

• Averages across shops. They hide the location where the change happened.
• Correlation dressed as cause. Two numbers moving together is a lead, not a verdict. In Ratchet+Wrench's 2018 survey, shops that tracked KPIs were more likely to report a higher effective labor rate and average repair order, but that is a correlation too.
• Too many questions at once. Ten questions is the ceiling, not the start. Many weeks, the first three are enough.
• No owner. A finding without a name next to it is a chart, not a decision.

How to tell it is working

After a month, you should be able to point to four decisions the questions led to, each with an owner and a result you checked the following week. If the answers keep pointing at data gaps instead of real changes, fix the data first: closed repair orders and fresh feeds make every other question easier.

For the weekly page these questions feed, see auto shop reporting AI and auto repair profitability software. For groups, the RevvOS multi-location view compares connected shops and marks a shop's data as unavailable when a feed is down rather than guessing.