Monthly AI allowances in a contract drafting tool

Your company sells a contract drafting tool to law firms. Its assistant drafts clauses with a hosted model. Each firm buys a monthly allowance of model tokens with its plan. When the allowance runs out, the assistant stops for that firm until the next month or until an admin buys more.

Firms range from 3 lawyers to 900. In a large firm, up to 250 people use the assistant at the same moment. The app sets a limit of 8,000 tokens on every call, and a draft uses 1,000 to 8,000. The model reports how many tokens a call used only when the call ends. A call takes 5 to 40 seconds. Your contract with the firms says you never serve or bill tokens past the allowance they bought.

What do you check before a request starts, and what do you record when it ends?