• This thread is just the tip of the iceberg.The people ahead of the curve aren't Googling for answers — they're already in here, having the conversations you haven't found yet. DealerRefresh is free.Get the full picture →

What dealership task has AI actually improved—not just made flashier?

Alex Snyder

President Skroob
Staff member
May 1, 2006
4,119
2,991
Awards
13
First Name
Alex
There’s no shortage of new AI tools hitting the car biz right now, but I’m more interested in what has produced a measurable improvement inside real dealerships.
  • Faster lead response?
  • Better VDP descriptions, merchandising, or photo/video workflow?
  • More effective service scheduling?
  • Less time spent on reporting?
  • Better website content, SEO, or social?
Bonus question: what sounded promising but turned out to add work, create risk, or produce no meaningful ROI?
 
Full disclosure, I build Marketplace posting software for dealers, so weigh accordingly.
The one I can put a number on is photos, and the part that moved was not the part that looks like AI.
Across 196 rooftops we have run about 94,000 listing photos through AI background replacement, in 3,237 processing jobs. The generated showroom background is what gets demoed. It is not what produced the improvement. That came from the classifier sitting in front of it, the one that decides whether a photo is a usable exterior shot at all before anything gets generated. Feed photos are a mess: interiors, dashboards, window stickers, a thumb over the lens, another site's watermark baked into the corner. Before that step we were producing beautiful backgrounds behind unusable subjects.
So the pattern I would offer on your first question: the AI win is usually the boring decision in front of the flashy generation. Classify, route, decide what not to do. Those have right answers you can measure against. "Is this photo usable" does. "Write me a better VDP description" does not.
Which is your bonus question. Generated vehicle descriptions were our worst spend of time. They read fine, they changed nothing we could count, and they carry a real risk, which is a model confidently listing a feature the car does not have. We now write nothing about a vehicle that did not come out of the feed, and we will not let the model repaint a color. A buyer standing next to a car that is not the color in the ad costs more than any lift in clicks pays for.
Where my own number is weak: 94,000 is a volume count, not a lift. I can tell you the work got done and that stores keep it turned on. I cannot hand you a clean units-sold delta, because nobody controls the other twenty things that changed on that lot in the same quarter. Anyone who hands you that attribution cleanly is selling you something.
The practical version for anyone evaluating: pick one task with a countable before and after, and instrument it before the tool goes on. Photo reject rate, minutes to first response, listings live versus units in stock. If the vendor cannot name the single number that should move, that is your answer.
 

✨ AI Highlights

Alex Snyder asks dealers and vendor partners to share which AI applications have delivered real, measurable results inside dealerships—covering lead response, VDP content, service scheduling, reporting, and marketing—while also calling out tools that overpromised and underdelivered. The thread invites honest, experience-based input rather than hype, making it a practical resource for anyone evaluating where AI actually earns its keep in a dealership operation.

Replies Views 1 110 Started Last Reply