Prepaid token bundles, drawdown models, credit expiry, and designing credit economies for AI products.
Credits are everywhere, but most credit systems are designed casually. Five decisions you need to make deliberately: exchange rates, tiered pricing, refresh cycles, pooling, and currency design.
Hard credit expiry looks like found revenue on your P&L and destroys NPS in your customers' heads. The math on why rollover credits almost always win, and how to design them.
AI-native companies figured out how to sell usage-based pricing to CFOs without them realizing it. It's called credits, and it's brilliant.
Credits are prepaid units that customers draw down as they use a product. They translate variable usage into a predictable purchase — making usage-based pricing palatable to CFOs who want budget certainty. OpenAI, Anthropic, and many AI companies sell credits.
Five key decisions: credit-to-usage exchange rate, expiry policy, rollover rules, top-up mechanics, and volume discount tiers. Get the exchange rate wrong and you're either leaving money on the table or pricing out customers.
Rolling credits almost always win on NPS and retention. Hard expiry creates found revenue on paper but destroys trust — customers feel punished for buying ahead. If you must expire credits, use 12-month windows with clear notifications.
Tokens are the raw compute unit (input/output tokens for LLMs), while credits are an abstraction layer above tokens. Credits let you decouple your pricing from the underlying model cost, giving you room to change models without repricing.