Retail's AI money is in markdowns and returns
The deployments earning their keep are unglamorous: perishable markdown pricing, returns triage, shrink review and catalogue copy. The store-floor robotics story has quietly reversed, and two chains have written off the fleets.
What's genuinely in production
Perishable markdown pricing is the clearest win in the sector. Marchetti Grocers cut fresh waste 18% across 61 stores; the deployment cost less than the waste it removed in the first quarter. The hard part was not the model — it was giving produce managers an override button and then looking at how often they used it.
Returns triage is close behind. Ferndale Home routes 900 returns a day from a photo and a sentence, and the saving is a clean per-unit number that finance can audit. This is the deployment most mid-sized retailers should look at first.
Shrink review turns existing camera footage into an audit function you could not afford to staff. Kirana Junction found ₹3.1 crore in nine months of till video. Expect an uncomfortable conversation with your works council or union before it ships.
What's still pilot theatre
Store-floor scanning robots. Two documented write-offs, no documented success at mid-market scale. The economics only ever worked in the vendor's model of a store.
Personalised generative merchandising. Plenty of demos, plenty of pilots, and nobody willing to show us a revenue figure they would stand behind.
Vendors that keep appearing
| Markdown pricing specialists | in 4 of 4 pricing deployments |
| Helpdesk AI add-ons | the default service entry point |
| Loss-prevention video vendors | often already installed |
| Own data team + model APIs | catalogue copy, almost always |
Regulation coming
Camera-based shrink review is where retail gets into trouble fastest: employee-monitoring rules, DPIA obligations and works-council consultation apply before a single frame is processed. Two of the chains we cover paused deployment for three months on exactly this.