Companies are pouring money into artificial intelligence and, for the first time, tracking usage with precision—even as many remain unsure of the payoff. Microsoft and Salesforce now let enterprises monitor prompts, agent activity and costs measured in “tokens,” turning every interaction into a budget line. A ModelOp survey finds most firms still rely on estimated time savings rather than auditable financial results, a gap it calls the “AI value illusion.” Some employers are ranking workers by AI consumption, fueling “tokenmaxxing,” though executives concede usage is a poor proxy for productivity. Coinbase is cutting 14% of staff and reorganizing into “AI-native pods,” while Globant bakes token costs into project COGS and restructures teams where agents deliver measurable gains. Salesforce reports 2.4 billion “work units” and rising agent completion rates; McKinsey says 64% see AI driving innovation but only 39% report earnings impact. Meta is testing granular activity capture to train models, intensifying privacy concerns. The bottom line: boards want ROI, and it’s easier to prove when agents complete tasks end-to-end than when humans use AI tools—pushing measurement from individual usage toward workflow outcomes.
Related articles:
— Generative AI at Work: Evidence from a Customer Support Firm
— Copilot for Microsoft 365: Adoption and Measurement Resources
