An AI cost number in isolation is just a number. The question that matters is whether it is earning its keep, and that requires tracking AI cost against revenue. For any product with AI features, this is the difference between a healthy margin and a business that scales itself into a loss.
AI cost is a cost of goods sold
For a product with AI features, AI cost is not overhead; it is a direct cost of serving customers, which means it belongs in your gross margin calculation. Tracking it against revenue turns an infrastructure line into a margin input, which is the number that determines whether the product is actually profitable at scale.
Track cost as a percentage of revenue
The core metric is AI cost as a share of the revenue attributable to AI features. As long as that ratio stays within a healthy range, the feature is sustainable. When AI cost consumes too large a share of the revenue it supports, the unit economics need attention through optimization or pricing, and only the ratio makes that visible.
Attribute cost to revenue-generating features
To compare against revenue, AI cost has to be attributed to the features that drive revenue. Tag AI calls by feature so you can line up each revenue-generating feature's cost against its contribution. A feature that generates strong revenue can justify high AI cost; one that generates little cannot, regardless of how cheap it looks.
Watch cost and revenue scale together
The risk is AI cost scaling faster than the revenue it supports. Track both curves over time. Revenue growing faster than AI cost means improving economics; AI cost outpacing revenue is the early signal of a margin problem, and catching it early is the point of tracking the two together.
Feed it into pricing
Tracking cost against revenue is what grounds AI pricing in reality. It tells you which features can absorb their cost at current pricing, which need usage-based tiers, and which are underpriced relative to what they cost to serve. Pricing set without this tracking is a guess.
AI cost is only meaningful next to the revenue it produces. Track it as a share of attributable revenue, feature by feature, and watch the two curves together, and you protect the margin instead of discovering a problem in the quarterly review.