Most businesses are measuring AI wrong, and it’s costing them

Amazon has discontinued its internal AI leaderboard that tracked token usage, as the gamified system encouraged more AI tasks but yielded fewer valuable outcomes. An Amazon SVP instructed staff to avoid using AI solely for the sake of adoption. This trend is not isolated; Uber exceeded its 2026 AI coding budget within four months, and Google's CEO, Sundar Pichai, reported a sevenfold increase in token usage over the past year. Companies like Meta, Microsoft, and Salesforce are reportedly implementing measures to curb token consumption. This focus on metrics like tokens per query, cost per inference, and GPU hours, while appearing sophisticated, represents a misdirection. The prevalent "tokens are the new oil" slogan highlights a shift towards measuring AI adoption and productivity through token usage, but it fails to address whether these investments are genuinely increasing revenue, accelerating decision-making, reducing friction, or providing a tangible competitive advantage. While companies can quantify AI expenditure, they often lack insight into the actual utility of the intelligence generated. The current economic model for AI is becoming unsustainable, with improved dashboards not translating into better returns, and proliferation of metrics like word counts and lines of code not impacting the bottom line. This obsession with cost overlooks the strategic value AI can deliver.