Companies Turn to Tokens to Measure AI Adoption, but Experts Question Productivity Link

Tokens are the basic units that large language models use to break down and process information, and providers typically charge for both prompts and responses based on token usage. Companies are therefore gradually shifting from per-user licensing metrics to token tracking to measure AI adoption, workload intensity and spending. But token counts reflect only computing volume and do not necessarily indicate output quality or return on investment.

PYMNTS reported on March 19, 2026, that average inference-token usage among OpenAI enterprise customers had increased about 320-fold over the previous 12 months. TechNews reported separately on March 25 that an OpenAI engineer used 210 billion tokens in seven days, while a Claude Code user's monthly bill exceeded $150,000. The figures underscore that heavy usage can also stem from inefficient prompts or leaks in agent workflows.