Enterprises and Notion Adopt New Strategies to Control AI Token Costs

Companies are confronting unexpectedly high AI budgets by dissecting token‑spend variance. CIOs recommend a rate‑volume analysis that separates budget versus actual spend across three dimensions – total cost, average token price and token volume – and further breaks down each token type. This method helps pinpoint whether price changes, usage volume or a shift in token mix is driving overspend, enabling targeted corrective actions.

Notion’s AI lead, Sarah Sachs, describes a complementary approach that avoids reliance on token economics alone. Notion routes roughly 75 % of traffic through an auto‑model that dynamically selects providers, a strategy she dubs “AI Switzerland.” The company prioritises cost per capability per second, employs open‑weight models for routine tasks, and runs simple processes on CPUs rather than expensive GPUs, preserving flexibility and preventing vendor lock‑in.