What Uber, Starbucks, and Microsoft Got Wrong About AI

Three of the world's most recognizable brands just had very public AI fails. The lesson isn't that AI doesn't work — it's that poor strategy and rushed implementation break even the best tools.

Three of the most recognizable brands in the world just had very public AI fails. Uber admitted that they burned through its entire 2026 AI budget in four months. Starbucks scrapped an AI inventory tool after nine months of miscounts. Microsoft canceled thousands of AI coding licenses it had been actively pushing employees to use just six months earlier.

The headlines are writing themselves: AI is too expensive. AI doesn't work. AI is a lost cause.

That's the wrong takeaway. If you're a business leader watching this unfold and doubting your own AI initiatives, here's what you actually need to hear:

The Problem Was Never AI

These cases prove what we’ve been saying: AI failures are rarely a failure of the technology and more so that of poor strategy and rushed implementation.

Starbucks rolled out an AI-powered inventory tool called “Automated Counting” across all North American stores in September 2025. They instructed their baristas to use tablet cameras and LiDAR to automatically scan shelves and flag shortages. This was a brilliant idea in theory. As they rolled out, they found the system couldn't tell oat milk from 2%, it missed items sitting in plain sight, and employees would have to manually fix these discrepancies. After nine months, Starbucks pulled the plug entirely and went back to counting by hand.

The tool was not tested at scale, and real-life data is much more inconsistent than computer data. Varied product labeling across thousands of locations and no standardized shelf organization led to inaccurate results. You can have the most sophisticated AI on the planet, but if you're feeding it garbage, you'll get garbage back every time (another thing we warned about months ago).

Uber's situation is different but the lesson is the same. The company set up internal leaderboards to rank teams by how much AI they were using. This quickly showcased that engineers were “tokenmaxxing” — consuming AI compute as fast as possible to climb the rankings. By April, they'd burned through their entire year's AI coding budget. When COO Andrew Macdonald sat down with senior engineering leaders to understand what the company actually got for that spend, he found he couldn't draw a clear line between all those AI tokens and meaningful consumer features actually shipped. “That link is not there yet, right?” he said publicly.

That's not an AI problem. That's a measurement and integration problem. Uber gamified usage without defining outcomes. They optimized for the metric instead of the mission.

Microsoft canceled a significant portion of its Claude Code licenses across its Experiences and Devices division, a tool its own engineers had come to rely on and adopted quickly. The cost of scaling had become the bottleneck, not the value of the tool itself. As one industry analyst put it: when agentic AI models require far more compute per task than standard tools, and enterprise consumption explodes, even falling per-unit costs can't keep up with a runaway bill.

The Pattern Is Clear

All three companies made the same core mistake: they deployed AI before they were ready for AI.

They skipped the foundational work. They didn't establish clean, structured, governable data. They didn't define what “success” looked like before they started spending. They didn't integrate AI into their workflows thoughtfully, they bolted it on top of existing chaos and hoped for the best.

This is what we mean when we say dirty data is the silent killer of every AI initiative. AI is only as smart as the information you give it. If your data is siloed, inconsistent, or unstructured, an AI tool doesn't fix that, it only amplifies it.

And without a clear integration strategy tied to business outcomes? You get Uber's situation: massive spend, maximum adoption, with zero ability to prove ROI.

What They Should Have Done

Before deploying any AI tool at scale, the groundwork has to be laid. That means three things:

1. Get your data in order first: Audit what data you have, where it lives, who owns it, and whether it's clean and consistent. For Starbucks, this would have meant standardizing product data across every store before asking AI to recognize it. No amount of sophisticated scanning technology overcomes a labeling system that varies by region.

2. Define success before you spend: AI ROI doesn't measure itself. You need to establish baselines (no pun intended) and decide upfront what “better” looks like. This could be in dollars, in hours saved, in features shipped, in customer experience. If you can't measure it, you can't manage it, and you definitely can't justify it to your board.

3. Integrate strategically: The companies winning with AI right now aren't the ones moving fastest but rather with intention. They're identifying the specific workflows where AI adds the most leverage, deploying in contained environments first, measuring results, and scaling what works. They're not gamifying token consumption.

So Is AI Worth It?

Absolutely, but not like this.

The companies bleeding money on AI right now failed in their approach. AI implemented on top of clean, well-governed data with a clear integration strategy and defined success metrics? That's where we've seen clients reduce operational costs by up to 50%. That's real, measurable business value.

This is exactly why we built Baseline the way we did. We're not here to sell you the hottest AI tool. We’re most passionate about making sure you're actually ready for it. This means your data is clean, your strategy is sound, and any technology you invest in has a clear line back to ROI. We always diagnose first, prescribe second.

The lesson from Uber, Starbucks, and Microsoft isn't “AI is too expensive.” It's: don't build on a broken foundation.