Starbucks investors get tough-luck news on AI inventory bet

Corporate America has spent the past two years acting like every problem has an artificial intelligence (AI) solution attached to it. The pitch is always the same. Faster than people. Cheaper than people. More accurate than people. Boardrooms have funded it, consultants have sold it, and shareholders have rewarded it.

Then the software meets the actual store.

A demo that looks airtight on a conference room screen has a habit of falling apart somewhere between the milk fridge and the morning rush. Computer vision that can tell a stop sign from a yield sign in a self-driving simulation can still get tripped up by two cartons sitting inches apart on the same shelf. The gap between “this works in a pilot” and “this works in 11,000 stores at 7 a.m. on a Monday” is the gap where a lot of corporate AI projects quietly die.

That gap just claimed a high-profile one.

Starbucks (SBUX) has retired the AI-powered automated inventory tool it rolled out to its North American coffeehouses last September, returning to the manual counts the technology was supposed to replace. For a turnaround story that has leaned heavily on technology fixes, the reversal is more than a footnote.

It’s a tell about how the next 12 months might go.

Starbucks retires AI-powered automated inventory tool after eight months

DKart / Getty Images

The tool Starbucks just retired

The tool, called Automated Counting internally, used handheld tablets running computer vision software to scan refrigerators, shelves, and display cases and tally items like milk jugs, syrup bottles, and coffee bags. NomadGo, the Redmond-based vendor that built it, marketed the system as delivering up to eight times faster results than manual methods with 99% accuracy, according to a BusinessWire report at launch.

The reality was a system that reportedly struggled to distinguish similar milk types, missed bottles altogether, and, in one promotional video, failed to recognize a basic peppermint syrup, according to Reuters.

More Restaurants 

30 year old restaurant has closed all restaurantsAfter bankruptcy, Hooters closes restaurants, fights for survivalIconic Las Vegas Strip restaurant closes without warning

“Starting today, Automated Counting will be retired,” read an internal Starbucks newsletter dated Monday, May 18, and verified by Reuters with two store employees. Beverage components and milk will revert to manual counting, the memo said.

The reversal came from a decision to “standardize how inventory is counted across coffeehouses as we continue to focus on consistency and execution at scale,” Starbucks told Reuters in a statement.

That is corporate-speak for one thing. The tool was not working.

Related: Starbucks delivers tough update on regional offices, cuts 100s of jobs

Why the AI inventory tool failed Starbucks

The failure mode here is worth understanding because it shows up in nearly every enterprise AI rollout that misses the mark.

NomadGo’s system, built on what the company calls Spatial Vision technology, combines computer vision, 3D spatial intelligence, and augmented reality to identify products on a shelf. In a clean warehouse demo, the system is convincing. In a Starbucks back-of-house at peak hours, where oat milk, 2% milk, almond milk, and breve sit inches apart in nearly identical jugs, the math gets harder fast.

When I ran the numbers on what these systems actually have to do at scale, the problem became clear. The AI has to read a label, classify contents, and update an inventory count in real time without slowing a barista who already has 15 customers in line. Get the call wrong, and a store that thinks it has plenty of oat milk runs out at 8:15 a.m. and turns customers away.

NomadGo said the tool is “continuously learning from customer and user feedback,” according to a statement reported by Ground News.

That is the polite version of the story. The harder version is that Starbucks rolled this out to more than 11,000 stores after a pilot, and within nine months it pulled the plug.

What this means for Brian Niccol’s turnaround math

For Starbucks investors, the AI walk-back lands at a sensitive moment. Brian Niccol’s “Back to Starbucks” plan is finally producing the comp sales numbers shareholders wanted, but the cost side has not caught up.

Here is the picture in numbers:

Global comparable store sales grew 6.2% in fiscal Q2 2026, the strongest quarter in two and a half years, according to a Starbucks earnings release.North America comparable transactions rose 4.4% year over year in the same quarter, according to the same Starbucks 8-K filing.North American operating margins fell to 9.9% in the most recent quarter, down from 18% two years earlier before Niccol took the helm, according to The Globe and Mail.Starbucks shares have risen roughly 24% year to date in 2026, according to IBTimes.

The translation for a 401(k) holder is simple. Revenue is growing again. Margins are still bleeding. Every dollar Niccol spends fixing operations has to clear the bar of actually working, because every bet that misses is margin that does not come back.

This is where the AI inventory tool stings. It was supposed to give Niccol live visibility into the product shortages he has blamed for hurting sales. Instead, the company spent nine months running a system that frequently could not tell one white liquid from another, and is now back to manual counts.

My analysis of the broader trend is that Starbucks is not alone here. MIT’s NANDA initiative found that 95% of enterprise generative AI pilots delivered no measurable impact on the P&L, despite roughly $30 billion to $40 billion in industry spend, according to The Next Web’s coverage of the study. The Starbucks tool was not generative AI specifically, but it lives in the same category of expensive, board-blessed pilots that struggle when they hit a real store at scale.

Niccol has not abandoned the AI thesis. The company is still building tools for order sequencing and barista assistance, with a new generation of AI barista support reportedly in development.

The question for Starbucks shareholders is no longer whether the company will use AI. It is whether the next AI bet ships better than the last one. Investors have given Niccol the benefit of the doubt for 18 months. They have not given him forever.

For now, the manual count is back. So is the human pair of eyes that has been doing this job for decades. Sometimes the lower-tech answer is just the right one.

Related: Starbucks drops another summer surprise as competition heats up

#Starbucks #investors #toughluck #news #inventory #bet