Tracking AI spend as it happens is valuable, and its real payoff is forward-looking: a trend you can track is a forecast you can make. The catch is that AI spend does not extrapolate linearly, so tracking the total and drawing a straight line produces forecasts that are reliably wrong.
Why linear extrapolation fails
AI spend grows on two compounding drivers. Adoption breadth, as more people and use cases come online, and adoption depth, as existing users consume more intensively and agentic workloads scale with activity. A straight-line projection captures neither, which is why forecasts built on last quarter's total miss badly during a growth phase.
Track the drivers, not just the dollars
A forecast that holds up tracks three things beyond the total. User adoption rate by team, per-user spend intensity over time, and the pipeline of AI features heading to production. These are the levers that actually move next quarter's spend, and tracking them turns a guess into a model with inputs you can defend.
Build the range, not the point
Because the drivers involve judgment, the honest output is a range. A low-growth case where adoption slows and intensity stays flat, a base case on current trends, and a high-growth case where a major new use case ships early. Tracking the drivers lets you attach real assumptions to each scenario instead of presenting a single false-precision number.
Recalibrate every month
A forecast is a living thing. Each month, update the driver inputs from the latest tracked data and check whether last month's projection held. When projections are consistently off in one direction, the growth assumptions need adjusting. That monthly loop is what keeps the forecast honest as the program evolves.
Tracking AI spend over time earns its keep when it improves the forecast. Track the drivers, build a range, and recalibrate monthly, and next quarter's number stops being a surprise you explain and becomes one you called.