Token Spend Isn't a KPI. It's a Bill.

Every company chasing AI adoption right now is staring at the same dashboard. Token spend. Usage per seat. Leaderboards ranking who's "most AI-native" this quarter.

They picked that number because it was sitting right there in the billing console. Not because it tells you anything about the business. Spend is easy to grab. Whether a customer is happier, whether their problem actually got solved, whether that turned into a sale, that takes real work to define and track. So leadership defaulted to the easy number and called it a strategy.

That's the failure. Not an AI failure, a leadership one. Spend doesn't correlate with outcomes, and any org that's treating it like it does is measuring the wrong thing on purpose, because the wrong thing was easier to find.

🔥 Meta had to shut it downLink to this section

Meta capped internal AI spend after costs started approaching billions. CTO Andrew Bosworth's internal memo didn't dance around it.

"Nobody should be using AI tools just for the sake of using them. All motion is not progress and token usage alone is not a measure of impact of any kind."

Here's what actually happened. Employees started gaming internal usage leaderboards. People inside Meta nicknamed it "tokenmaxxing." When you tell smart people that a number on a leaderboard determines how impactful they look, they will optimize for the number. Not the outcome. The number.

That's what happens every single time you pick the wrong metric. AI just made it visible faster.

💸 Uber burned four months of budget in four monthsLink to this section

Uber ran through its entire 2026 AI budget in four months. When COO Andrew Macdonald was asked whether that spend connected to shipped value, he didn't spin it. "That link is not there yet," he said.

That's a remarkably honest answer from a COO, and it's the whole problem in one sentence. Nobody had built the bridge between dollars spent and value created. So the spend just... happened.

Not everyone let it happen without a fight, though.

🎯 Benioff is closer, but not all the way thereLink to this section

Marc Benioff moved Salesforce off tokens entirely and started tracking "agentic work units" instead. That's real progress. It shifts the question from what went in to what came out.

But output isn't the same as outcome. A work unit can be completed and still not matter to a single customer. The real fix goes one layer deeper: don't just count what got produced, count what it did. Did it make a customer happier. Did it resolve a problem they actually had. Did it drive a sale or protect one that was at risk.

That's the KPI leadership should be orienting around. Not spend, not even output, but the thing spend and output are supposed to serve.

🔁 I've watched this movie beforeLink to this section

I started my career in the 2000s, and I watched engineering orgs measure people by lines of code written. The best engineers really did write more, because they understood the problem cold and the solution just came out of them fast. High output was a byproduct of skill, not the source of it.

Leadership drew the wrong conclusion anyway: if our best people write more code, more code must make people better. So they pushed juniors to write more, faster. It backfired every time. Juniors who hadn't earned that instinct yet needed to slow down and think through the architecture first. Pushed to hit a number instead, they produced buggy code and projects that had to be torn apart and reengineered.

Code volume correlated with skill in the people who already had it. It didn't cause skill in the people who didn't. Token spend is the same trap. Your strongest people spend more because they already know what they're doing. Push everyone else to match that number and you get the AI-era version of the same rushed, half-baked output, minus the thinking that should have happened first.

✅ The good version of this actually existsLink to this section

Slack found it early. In their growth days, they didn't chase signups or daily logins. They found that teams who sent 2,000 messages had a 93% likelihood of sticking around as long-term users. So they built their whole activation strategy around getting new teams past that number.

Slack co-founder Stewart Butterfield explained what that number actually meant:

"Based on experience of which companies stuck with us and which didn't, we decided that any team that has exchanged 2,000 messages in its history has tried Slack, really tried it... regardless of any other factor, after 2,000 messages, 93% of those customers are still using Slack today."

That's the key detail. 2,000 messages wasn't picked because it was a round number or an easy one to track. Someone had to dig through the data and test which early behavior actually predicted the outcome the business cared about. It was the point at which a team had actually put Slack to work. Below that line, they were still just poking at the tool. Above it, Slack had quietly become the place where decisions got made and work got coordinated, and once that happened, teams didn't leave. The number was a proxy for something real: has this product actually solved the problem it was bought to solve.

That's the difference. Lines of code and token spend are numbers you can pull without asking a single question. 2,000 messages is a number you only get to after you've asked the right one: what does a customer actually do right before they become someone who sticks around and pays.

🧭 Point the whole org at the customerLink to this section

None of this means stop tracking cost. You should know what you're spending. The mistake wasn't measuring spend, it was making spend the target instead of the tripwire.

Meta and Uber didn't get hurt because they tracked tokens. They got hurt because hitting the number became the goal, so people spent to look busy instead of spending to solve a customer's problem. Cost went up. Nothing came back for it.

Leadership has to orient the org around customer outcomes first: are customers happier, are their problems getting solved, is that turning into revenue. Spend still gets watched, but as a downstream number that moves in step with those outcomes, not a target chased on its own. The customer is the KPI. Cost is just how you know if you're earning the right to keep spending on them.

#AIStrategy #AIAdoption #Leadership #CustomerOutcomes #AIKPIs #ProductStrategy #TokenSpend