Starbucks dropped AI stock counts after 9 months; now cites daily deliveries
Starbucks ended camera-based stock counts after staff called them unreliable. Its CEO now reports record availability on daily deliveries. No cost given.
At a glance
- Company
- Starbucks · Retail and consumer · 5,000+ employees
- Team and use case
- Supply chain and procurement · Inventory optimisation (in-store stock counts)
- Tools
- Computer-vision inventory counting tool (vendor not named)
- Stage
- Paused or rolled back, since Announced September 2025; ended about nine months later
- Results
- How in-store stock is counted: AI counts for some inventory → One process for all counts
- Time the AI counting tool was in use: About 9 months
- Miscount or mislabel rate: Not disclosed
- N. American company stores with daily deliveries: Nearly 90% (CEO, Q4 call)
- Cost
- Not disclosed. No licence, hardware, rollout or write-off figure, and the annual report filed on 12 November 2026 does not mention the tool.
- People
- Store employees went back to one counting process for all inventory. Starbucks shared employee comments thanking it for trusting them over "unreliable spatial recognition". Who checked the AI counts while it ran: not disclosed.
- Evidence
- Company claim, no method
The company states a result without saying how it was measured.
What this leaves out: Starbucks confirmed the change in a statement to trade press and gave no error rate, cost or store count. The miscounts are Reuters' reporting, which Starbucks did not comment on. The CEO's "all-time high" availability has no figure and no comparison with the AI period.
Starbucks' store inventory availability is at an "all-time high", and nearly 90% of its company-operated coffeehouses in North America now get deliveries every day, chief executive Brian Niccol said on the company's fourth-quarter earnings call, [Supply Chain Dive reported on 18 November 2026](https://www.supplychaindive.com/news/starbucks-inventory-availability-at-all-time-high-with-daily-deliveries/874923/). The AI tool Starbucks announced in September 2025 to fix the same stockouts is no longer part of that: the company ended it about nine months after its debut.
What was tried and what replaced it
The tool used computer vision (software that recognises items in camera images) and 3D spatial intelligence to automate parts of the stock count in each store, with the intention of later automating some restocking orders, according to [Supply Chain Dive's 6 August report](https://www.supplychaindive.com/news/starbucks-targets-24-hour-inventory-replenishment/827030/).
[Supply Chain Dive reported on 4 June 2026](https://www.supplychaindive.com/news/starbucks-ditches-ai-inventory-system-after-just-9-months/821104/) that Starbucks was ending it in favour of traditional stock-keeping. In a statement shared with the trade title Restaurant Dive, Starbucks said it had "moved to a single, consistent process across all inventory counts. This approach supports accuracy and product availability in our coffeehouses." It added that it was being disciplined about where automation adds value. Reuters reported that the tool occasionally miscounted or mislabelled items; Starbucks did not comment on that report.
What Starbucks did afterwards was operational. On its 29 July earnings call it described testing a 24-hour model: more frequent inbound deliveries and restocking within a day of a store's signal. Niccol now says the replenishment goals will be met in the next six to 12 months, and that the range will be cut to 6,000 or 6,500 items. Tests of demand forecasting and ordering accuracy are under way; the report does not say whether they use AI.
The people and the catch
The verdict came from the people doing the counts. Starbucks shared images of what it said were internal employee comments. "Thank you for trusting the partners over unreliable spatial recognition to handle these counts," one read (Starbucks calls its employees partners). Another: "Very grateful our thoughts about AI count were heard." Who checked the AI's counts while it ran, and how much recounting it caused, is not disclosed.
Nor is anything a buyer could benchmark: no cost, no error rate, no number of stores, no vendor named in these reports. The "all-time high" has no figure or baseline, and the report gives no comparison with the months the tool was live. [Starbucks' annual report](https://www.sec.gov/Archives/edgar/data/829224/000082922426000162/sbux-20260927.htm), filed on 12 November 2026, does not mention the tool; it gives the workforce as approximately 306,000 people.
Reading this at work? Get one story like it explained every morning.
A five-minute read every morning at 7am ET: one story explained, two in brief, what else happened, one number. Unsubscribe in one click.
What to do if you run inventory or store operations
- Ask any vendor of camera-based counting for its miscount and mislabel rate on your own items, measured against hand counts on the same shelves, before you sign.
- In a pilot, keep one counting process as the record and give store or warehouse staff a named route to report wrong counts, with a date on which their verdict decides whether the tool stays.
- Check how much of your stockout problem is delivery frequency and range size before automating the count. Those are the two levers Starbucks is pulling now.
Take this to work
Forward this
Starbucks ended its camera-based AI inventory counting about nine months after announcing it, after store employees called it unreliable; its CEO now reports record availability with daily deliveries to nearly 90% of company-operated North American stores (Supply Chain Dive, 4 June and 18 November 2026). Any counting tool we trial should run beside hand counts and need a staff verdict before it becomes the record.
Questions to ask
- What is the tool's miscount and mislabel rate on our items, in our storage conditions, measured against hand counts on the same shelves?
- Who corrects a wrong count, how long does that take per count, and does the corrected figure or the tool's figure feed the reorder?
- What does it cost per site per month including devices, and what do we owe if we stop after nine months?
- How much of our stockout rate comes from counting errors, and how much from delivery frequency and range size?
Questions this article answers
Why did Starbucks stop using AI to count inventory?
Starbucks said it had "moved to a single, consistent process across all inventory counts" to support accuracy and availability, Supply Chain Dive reported on 4 June 2026. Employee comments the company shared called the tool's spatial recognition "unreliable", and Reuters reported it sometimes miscounted or mislabelled items, which Starbucks did not comment on.
How long did Starbucks use its AI inventory counting tool?
About nine months. It was announced in September 2025 and Supply Chain Dive reported its end on 4 June 2026.
What is Starbucks doing about stockouts instead?
Daily deliveries, which chief executive Brian Niccol said reach nearly 90% of company-operated North American coffeehouses, a 24-hour replenishment model due in the next six to 12 months, and a smaller range of 6,000 to 6,500 items, according to Supply Chain Dive on 18 November 2026.
What did Starbucks' AI inventory tool cost?
Not disclosed. Starbucks has given no cost, error rate or store count for the tool, and its annual report filed on 12 November 2026 does not mention it.
Indeed: AI users claim 2.1 hours saved a day, down from 2.3, as reviews sour
Indeed and Glassdoor report self-reported AI time savings slipping and 45% of AI mentions in staff reviews positive. HR should measure before the next rollout.
Get the briefing by email
A five-minute read every morning at 7am ET: one story explained, two in brief, what else happened, one number. Unsubscribe in one click.