Starbucks Ties Tech Bonuses to AI Usage as NomadGo Retires

Starbucks tied a quarter of its technology employees’ bonuses to artificial intelligence usage this week, according to an internal document viewed by Bloomberg News, joining a fast-growing list of companies that now measure how often staff reach for AI tools alongside what they actually ship. The Seattle coffee chain wants its developers to use an AI assistant multiple times a week to stay in compliance with department adoption targets, and it is counting how many initiatives under chief executive Brian Niccol’s “Back to Starbucks” turnaround plan run on AI under the hood.

The timing carries an awkward edge. Five days before the bonus story broke, Starbucks pulled NomadGo, the AI-powered inventory counter it had spent nine months running across roughly 11,000 North American stores, after baristas complained the system mislabeled bottles and grew less accurate the longer it ran.

The 25 Percent Slice and What Triggers It

Roughly a quarter of the tech bonus pool now tracks against department-wide AI goals, per the internal Starbucks document Bloomberg reviewed. Half of each tech worker’s bonus still keys to overall company financials, with the remaining slice tied to individual performance reviews. The new chunk is the one with the rule book.

That rule book has two components. Software developers must demonstrate they are calling on an AI assistant several times a week to count as compliant. Department leaders, meanwhile, have to show that a rising share of priority initiatives within the turnaround plan are being built with AI infrastructure under them.

The bonus design also pulls in operational metrics that have nothing to do with AI directly. Performance of the mobile order-and-pay function inside the Starbucks app sits in the same department-level basket as the AI targets, according to the same reporting. The combination tells engineers the bar is two-track: ship working features for the consumer app, and prove the work was AI-assisted along the way.

The Same Week NomadGo Came Out of 11,000 Stores

The NomadGo exit was confirmed in late May. The computer-vision inventory system, announced in September 2025 as a centerpiece of the “Back to Starbucks” plan, was rolled across roughly 11,000 North American locations and then quietly retired nine months later. Workers told Fortune the tool “started off not particularly accurate and got less accurate over time,” forcing teams to rearrange back-of-house shelves so cameras could read them and to recount items the model missed.

The mechanics of the failure matter for the bonus debate. Undercounting meant stores did not receive products they needed; overcounting meant cases of bottled drinks piled up at locations that already had stock. Carl Addison, a nine-year shift supervisor at a Shoreline, Washington store, told reporters the recount work landed back on the human staff. The pilot had been positioned as proof that AI could remove a routine labor burden; in practice it added one.

Niccol’s framing of the broader AI strategy has since shifted away from the efficiency-replacement language he used at launch.

Our vision has shifted from just automating for efficiency to augmenting the craft of our baristas.

That was the chief executive speaking at Starbucks’s Las Vegas leadership meeting, attended by more than 14,000 North American store managers, weeks before the inventory tool was pulled.

Green Dot Assist and the Tools Still Standing

Three Starbucks AI projects remain active going into fiscal Q3. Green Dot Assist, a generative AI helper built on Microsoft Azure’s OpenAI platform, began testing at 35 stores in June 2025 and is on track to expand across the United States and Canada through fiscal 2026. Baristas use a tablet to ask the assistant questions about drink recipes or equipment problems via voice or text. Chief Technology Officer Deb Hall Lefevre has said the helper should eventually auto-file IT tickets and surface staffing suggestions.

Deep Brew, Starbucks’s in-house AI platform that pre-dates Niccol’s tenure, still drives personalized marketing, demand forecasting and supply chain modeling. The Siren System, an automation push from 2022 designed to manage drink prep and order fulfillment, was scaled back to limited use by early 2025 after the rollout slowed queues at stores rather than speeding them up.

The pattern across the four projects is telling. The two tools facing customers and store partners directly hit walls when the model’s accuracy and the staff’s workflow diverged. The two that augment rather than replace human work keep expanding. The bonus structure now being rolled out is engineered to push toward more of the latter, but the metric it picks rewards adoption regardless of which kind a developer ships.

Meta, Amazon, Uber Run the Same Play

Starbucks joins a roster of companies running this play. Meta tied performance reviews and bonus eligibility to AI use earlier this year through an internal tracker called Checkpoint, which counts (among more than 200 other data points) how many lines of code at the company are generated with AI assistance. The company also rolled out a new “Meta Award” with a 300 percent bonus multiplier for top performers, with AI-aided output as a qualifier.

Amazon publicly encouraged engineers to “tokenmaxx,” internal shorthand for burning as many AI tokens per task as feasible. At Meta, an engineer built a leaderboard nicknamed Claudeonomics that tracks which colleagues draw the most usage from Anthropic’s models. Uber’s chief technology officer told staff in May that the rideshare company had spent its entire 2026 AI coding budget in four months, after running internal leaderboards that ranked teams by how heavily they used coding assistants.

The patterns these programs share are consistent across employers:

• Department-level adoption goals counted against headcount, not against measurable productivity output
• Bonus or rating consequences for engineers who do not show enough AI tool activity
• Public or semi-public leaderboards that turn usage into a status signal
• Approval rather than caution from the executive layer when teams overshoot their AI-spend budget

The cumulative effect is that “did you use it” has become a separate management metric from “did it work.” The 25 percent slice extends the same logic into a sector where the customer touch happens at a hot bar in a store rather than inside a development environment.

Adoption Counters Versus Outcome Counters

The design problem inside an adoption-keyed bonus is straightforward. A developer who calls an AI assistant five times a week to draft scaffolding she would have written by hand in twenty minutes still triggers the compliance check, and so does a developer who hands the assistant a complex refactor and ships better code in a third of the time. Both employees register the same data point.

Inventory teams hit a similar wall during the rollout that ended this month. The computer-vision tool was deployed across thousands of stores fast enough to look like a textbook adoption success, until the per-store accuracy numbers started turning the wrong way. A program that rewards rate of rollout rather than rate of error correction can carry a failed pilot a long way past the point where someone should have pulled it.

Niccol’s corrective language, the shift toward “augmenting” rather than “automating,” reads as a direct response to the failed inventory episode. Whether that corrective extends to how the bonus is engineered will show up in which projects developers select inside the AI-funded portion of next fiscal year’s roadmap. If the tools they pick lean back toward outcome-shaped reviews (barista hours saved per week, percentage of customer interactions resolved without escalation), the program will read as different from the Meta and Uber template. If it tracks closer to a usage-counter system, it will read as the same.

The Q3 Earnings Window

Starbucks’s fiscal Q2 numbers, posted April 29, gave the turnaround its first clear quarterly win since Niccol took the corner office. Net revenues climbed 9 percent to $9.5 billion, comparable store sales rose 6.2 percent globally, and the company lifted full-year same-store guidance to “at least 5 percent” from a prior 3 percent, according to the company’s fiscal Q2 2026 earnings announcement. The investor pitch behind the bonus push depends on tech spending translating into more of those quarters, not fewer.

Fiscal Q3 is the first reporting period that will sit fully under the new compensation rule. Two questions point at the same earnings call. Did the Azure OpenAI helper’s wider rollout produce faster service times at the locations that received it? And does the count of AI-powered initiatives correlate with the comparable-sales trajectory, or run independently of it? The supporting data will surface inside Starbucks’s next 8-K quarterly filing on EDGAR.

If both lines move together, the bonus design will graduate from an HR story into a template the next consumer-services company copies. If they do not, the structure will look like the Meta and Uber playbook in a green apron, with the same exposure to the gap between adoption volume and operational result. The answer arrives at the next earnings release.