Inference Cost Envelope

Inference Cost Envelope is the maximum monthly model spend a client engagement can absorb before gross margin falls below the agency's floor. It is set by three variables: model tier, reasoning effort, and request volume. Most agencies price AI work from a single demo run, then discover that production traffic at high effort tiers multiplies token consumption several times over. The envelope forces a pre-sale calculation: map each deliverable to a reasoning tier, estimate tokens per task, and cap the monthly ceiling in the statement of work. A practical example is routing structured extraction through Cloudflare's Clairic on Workers AI instead of a frontier model, which cuts per-document cost and latency for high-volume client pipelines. Agencies that publish the envelope alongside the retainer convert cost overruns into a change-order conversation rather than a margin write-off.