Pricing

Why agent product pricing cannot inherit from SaaS — service cost is an endogenous variable. Three base models, tier design principles, price drift handling.

Two failure modes when applying SaaS pricing

A fixed-seat monthly subscription transplanted from SaaS to an agent product triggers two simultaneous failures:

  • Heavy-user margin collapse — at the same $40/month subscription, a heavy user’s token spend can hit $75, yielding a −88% margin (see metrics/unit-economics)
  • Vendor absorbs model price increases — when upstream prices rise, every tier’s effective capacity shrinks; monthly fees are locked at fixed numbers, the cost rise is absorbed entirely by the vendor

The structural difference: agent product service cost is an endogenous variable — every inference is billed, tightly coupled to per-customer usage intensity, and cannot be priced independently of behavior. The new objective is to encode “cost grows with usage” into the pricing structure rather than bet on user behavior staying cost-friendly.

Three base models

Subscription

Fixed monthly fee, usage included.

  • Suitable when: user behavior is predictable, task complexity is uniform, market has clear price anchors
  • Risk: heavy users become loss sources; under-pricing fails to cover cost
  • Required guardrail: tier cap (see economics/controls-and-roi) + degradation or suspension on overage

Usage-based

Billing by token count, task count, or labor-hours.

  • Suitable when: user usage frequency varies widely; customers prefer “pay for what you use” financial model
  • Risk: customers are sensitive to bill uncertainty; enterprise budget approval is difficult
  • Required guardrail: daily / monthly caps + usage alerts + transparent real-time billing

Hybrid

Base subscription covers medium usage + per-token charge above limit. Most enterprise-targeted agent products adopt this model; trade-offs in subscription-vs-usage.

Tier design

Tier differences are not simply “more features” — at their core they are layered model budgets + layered task volume caps. See tier-design.

Price elasticity and drift

Model unit price is variable. OpenAI and Anthropic have lowered average input prices by approximately 60% and output prices by approximately 50% over the past 18 months when releasing new models. Agent product pricing must have a response plan for this drift:

  • Contract terms specify “current public model unit price × usage” rather than fixed amount — cost transparent, risk shared
  • Or reserve a “cost buffer” — monthly fee sufficient to cover the most expensive model case, with extra margin when models become cheaper
  • Do not bet that “models will only become cheaper” — historically price movement is bidirectional

Cross-section connections

Was this page helpful?
Edit on GitHub