Before you sell “unlimited AI”, set the allowance math.
This page is built for SaaS builders asking how to include chat, voice, or agentic API usage in a monthly plan without one heavy user turning a subscription into negative gross margin.
The four rows to model first
1. Normal subscriber
Expected prompts, voice minutes, tool calls, retries, context size, and model mix per month.
2. Heavy but legitimate user
5–10x usage from a real customer who loves the feature and stays inside the UI.
3. Abuse or backdoor path
Automation, shared seats, CLI/API loops, or scripted calls that bypass the intended product rhythm.
4. Failure month
Retries, long-context jobs, failed voice sessions, or agent loops that multiply model calls before anyone notices.
Decision rule: if row 2 or row 3 breaks your target gross margin, the product is not fixed-price anymore. It is a usage-metered service pretending to be a subscription.
Copyable cap-policy worksheet
AI usage cap worksheet Plan price: Target gross margin: Included AI allowance per billing period: Normal-user expected cost: Heavy-user expected cost: Abuse/backdoor expected cost: Failure-month expected cost: Policy choice before launch: [ ] included allowance only [ ] hard stop after allowance [ ] soft cap + warning [ ] paid top-up / overage [ ] higher tier only [ ] bring-your-own-key option [ ] cheaper fallback model after allowance [ ] do not bundle this feature yet What would make this unsafe to sell as fixed-price? What exact limit should the pricing page disclose?
How to turn a public reply into a countable signal
Likes, impressions and generic praise do not count. A reply counts only when someone names a real AI feature/pricing problem, asks for the worksheet, asks for the calculator link, or shows purchase intent.
MarginPulse feedback reply AI feature being priced: Current plan price: Included usage promise you are considering: Most dangerous usage row: normal / heavy / abuse-backdoor / failure-month What cap policy feels acceptable to customers? Would you use a worksheet/template pack? yes/no If yes, acceptable one-time price: CHF 19 / CHF 29 / CHF 49 / other
Manual reply packet for the current SaaS API-usage thread
Use only from an approved account/path that can answer follow-up questions. First reply should be useful and no-link. Share this page only if someone asks for the worksheet/calculator.
The pricing issue here is not just “limit API calls”; it is deciding which usage row your subscription actually includes. Before promising AI chat/voice inside a premium monthly plan, I would model four rows: 1. normal subscriber: expected prompts/minutes/tool calls per month 2. heavy legitimate user: 5–10x usage that still fits the product 3. abuse/backdoor path: automation or shared usage that bypasses the intended UI rhythm 4. failure month: retries, long sessions, or agent/tool loops that multiply calls Then pick the policy before launch: included allowance, hard stop, paid top-up, higher-tier access, BYO key, cheaper fallback model, or don’t bundle it yet. If row 2 or 3 breaks target gross margin, the feature is not really fixed-price; it is usage-metered under subscription branding.