conceptintermediatecurrentAI Product Engineering~1 min readVerified 2026-07-20#ai-product#pricing#cost#unit-economics

Pricing vs compute cost

Mechanism: unit economics → margin → routing decision

price, compute, support = .20, .06, .03
print("margin", price - compute - support)

Run with python3; expected output is margin 0.11000000000000001. Measure successful-task cost, retries, support, and quality loss before lowering price or model quality.

Production lens and exercises

Track gross margin by workflow, customer segment, model route, cache hit, and successful outcome. A low token price can still lose money through retries, reviewer time, support, storage, or abuse; rate limits and budgets are product controls.

  1. Add a 20% retry rate and a human-review cost to the artifact.
  2. Compare a flat price with usage pricing under a stated quality floor.

Sources

AI features often have variable marginal cost. Pricing has to cover model calls, retrieval, tools, retries, human review, abuse controls, and support — while still matching the value users perceive.

Cost is not just tokens

Cost Example
Model inference Input/output tokens or local GPU time
Retrieval Embedding, vector search, reranking
Tool calls External APIs, web fetches, code execution
Review Human approval and correction time
Retries Invalid output, timeout, user regeneration
Observability Trace storage and eval pipelines

Unit economics should be measured per successful task, not per model call.

Pricing patterns

  • Seat-based pricing when usage is predictable.
  • Credits or metered usage when cost varies widely.
  • Tiered limits for model quality, context length, and automation.
  • Enterprise pricing when review, compliance, or data isolation dominates.

Guard against abuse

Rate limits, quotas, model routing, and caching are pricing controls as much as technical controls. Unbounded "AI included" can become margin leakage.

Pitfall

Charging per token can expose cost but obscure value. Users buy completed work, not inference units.

Connects to: cost optimization · product metrics · cost-quality tradeoff