Frameworks, core principles and top case studies for SaaS pricing, learnt and refined over 28+ years of SaaS-monetization experience.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
.png)
Agentic AI breaks this model in two ways that are each independently serious and in combination are structurally destabilizing.
In traditional SaaS, the marginal cost of an additional user is zero. The software runs on multi-tenant cloud infrastructure. Whether ten people use the product or ten thousand, the cost to the vendor is roughly the same. This is what makes SaaS margins so attractive: gross margins of 75 to 85 percent are standard because revenue scales while costs do not.
AI features invert this equation. Every inference costs compute. Every API call burns GPU cycles. Every agent task that runs for an hour consumes tokens, calls external tools, and racks up infrastructure costs that scale with usage, not with seats.
The inversion is easy to underestimate because the most expensive work is often invisible. Reasoning models generate thousands of internal "thinking" tokens that never appear in the output the customer sees but count against the bill all the same. A response that shows five hundred words can process ten thousand tokens or more under the surface. The intuition that a lower per-token price means a lower total cost quietly stops holding.
To make this concrete, consider an agentic marketing platform. A user clicks a single button: "Generate Campaign." Behind that click, the agent drafts a plan, writes five emails, generates three images, and produces landing-page copy. The workflow consumes roughly fifteen thousand input tokens, four thousand output tokens, three image-generation calls, and twenty vector-database lookups. One click. One line in the product.
And the revenue question is the hard one: how should this be charged for? Each option distributes risk differently between vendor and customer: passing through token cost with a markup leaves the customer unable to predict the bill; a flat per-campaign fee places the variance on the vendor, particularly when a campaign is regenerated repeatedly; a platform fee smooths both but weakens the link between price and cost. Resolving that tradeoff, and metering finely enough to support whichever model is chosen, is the core design problem.
One customer paying the same monthly fee as another can consume fifty times the computational resources. Heavy users, the engaged customers a vendor should want most, can generate ten to a hundred times the cost of light ones. Margins that look healthy in aggregate can be sharply negative on specific accounts. As one infrastructure lead at a Series C AI startup described it:
"We realized we were running a utility company but billing like a magazine subscription."
If an agentic AI product replaces human workflows (customer support agents, junior analysts, content writers, code reviewers) then the number of human users decreases as the product succeeds. Per-seat pricing punishes the vendor for delivering value. The more effective the agent, the fewer seats the customer needs, and the less revenue the vendor collects. This is not a hypothetical concern. It is the exact dynamic that Microsoft confronted when it realized that Copilot, priced as a per-seat add-on at thirty dollars per user per month, was reducing the number of seats customers needed. Microsoft responded by raising the effective price per remaining seat: bundling more value into higher-priced enterprise tiers so that the seats AI eliminated did not simply vanish from the revenue line.
But this is a stopgap rather than a structural solution. It papers over the mismatch between how the product creates value and how it captures value.
Join companies like Zoom, DocuSign, and Twilio using our systematic pricing approach to increase revenue by 12-40% year-over-year.