AI unit economics
From Subscriptions to Inference: The New Economics of AI Products
Why the most important number on an AI product roadmap might not be adoption, but the cost of every answer it gives
Written by Vanya Sahi
The SaaS assumption
Traditional SaaS economics are built around a relatively stable relationship between a customer and a product. A subscription buys access; marginal usage is often cheap enough to ignore. AI products complicate that assumption because every request can trigger a meaningful computational cost.
The cost of an answer
Inference makes intelligence variable. The cost of a response depends on model choice, context length, tool calls, retries, and the complexity of the task. AI product teams therefore need to understand unit economics at the level of tokens, workflows, and successful outcomes—not only seats and monthly recurring revenue.
From usage to outcome
A useful metric is cost per successful outcome. An agent that spends more tokens but resolves a difficult task may be healthier than a cheap assistant that creates rework. This reframes product dashboards around value created, confidence, completion, and the computational resources consumed to get there.
The product manager as economic translator
Model routing, thoughtful limits, transparent pricing, and differentiated experiences for free users are product decisions. AI Product Managers now need enough technical and financial fluency to connect model behavior with margin, customer value, and return on intelligence.
Conclusion
The subscription is not disappearing, but it is no longer the whole story. The durable AI products will make intelligence economically legible: measuring what an answer costs, what it enables, and whether the relationship between the two improves over time.