ai-pricing

Start by identifying whether customers can measure a specific business outcome from your product. If outcomes are measurable and attributable, use outcome pricing; if tasks are countable but shared, use workflow pricing; otherwise use consumption pricing per token or API call.

Consumption bills per token or API call, workflow bills per completed task like a document processed, and outcome bills per measurable result like a resolved support ticket. Outcome pricing aligns best with value but requires precise attribution and success definitions.

Offer BYOK when enterprise customers demand it, have existing LLM provider contracts, or need data to stay in their own cloud accounts. Avoid it if your value depends on fine-tuned models, your buyers are non-technical, or your margins require model cost markup.

Apply the margin improvement stack: model routing to cheaper models for simple tasks, prompt caching, batch processing, fine-tuned small models, and response caching. Combined, these levers can improve gross margins by 30-45 points over 12 months.

Per-seat pricing penalizes the buyer for success because agents replace human workers, so fewer seats means less revenue as the product delivers more value. Use outcome or workflow pricing instead to align revenue with completed work.

Follow a six-phase playbook: analyze revenue impact, design the new model, train internal teams, roll out at renewal with grandfathering, put all new customers on new pricing immediately, and fully migrate within 12-18 months with a hard sunset date.