Bain & Company has published "AI Pricing: A Reality Check on Effort, Usage, and Outcomes," a short piece that challenges the seat-based default still anchoring most enterprise SaaS contracts. For product managers shipping AI features this quarter, the framing matters: AI cost has split into three distinct billable dimensions, and procurement teams are starting to price each one separately.

What does "effort" pricing actually mean?

Effort-based pricing ties the invoice to the work done to build and run a feature — model training, fine-tuning, retrieval-pipeline maintenance, and ongoing evaluation. A team shipping a RAG-powered support assistant pays for indexing, prompt iteration, and the engineering hours behind regression suites. The trade-off: predictable for vendors, punishing for buyers. PMs who route this through professional services or fixed-fee integration usually discover the marginal cost of every additional use case only after the SOW is signed.

Why is usage the default for API-first vendors?

Usage pricing converts consumption into dollars — tokens, API calls, image generations, minutes of inference. OpenAI, Google Cloud's Vertex AI, and Anthropic all publish metered rate cards tied to model size and context length. The advantage for buyers: pay only for what ships. The failure mode: a single looping agent or runaway prompt can burn a quarterly budget in a weekend, and a usage spike from a successful launch looks identical to an outage. PMs need hard spend caps and per-tenant alarms before any production cutover.

Can outcomes be priced at all?

Outcome-based pricing — pay per resolved ticket, per qualified lead, per document processed — is the model vendors pitch when usage looks risky to the buyer. The math is appealing: cost tracks value. In practice, defining the outcome, attributing it cleanly to the AI component, and disputing edge cases eats the savings before the first invoice lands. Bain's framing pushes PMs to ask which metric the vendor will sign a contract around, and which they will quietly refuse to put in writing.

What should a PM do on Monday?

Treat AI pricing as a portfolio decision, not a line item. Negotiate at least two of the three dimensions in the same contract — for example, a usage cap paired with an outcome rebate — and write the dispute-resolution clause before signing. Track internal cost per resolved unit alongside the vendor's invoice; the gap between the two numbers is where margin either expands or disappears by year-end.

The direction of travel is set: as foundation models commoditize and reasoning costs fall, usage rates will keep dropping while outcome guarantees rise. PMs who lock in usage-only contracts today will renegotiate from weakness in 18 months.