All posts
Research#Forecasting#Data Science
Forecasting AI spend with cohort modeling
Linear extrapolation is why your end-of-month bill keeps surprising you. Cohort-based forecasts predict the next invoice within ±3%.
PN
Priya Nair
Data Science Lead, SpendTensor
May 28, 2026 7 min read
Forecasting
If you forecast AI spend by extrapolating last week's daily run rate, you will be wrong every month. Usage doesn't grow linearly — it grows by feature cohort, and each cohort has its own adoption curve.
Our forecasting model treats every feature × tenant pair as a cohort. We fit a logistic curve to early usage, weight by recent traffic, and roll up to a portfolio-level estimate. Across our customer base, the median forecast error for the next 30-day invoice is 2.8%.