ChatGPT (flat subscription tiers): At $20/month for Plus, users pay for AI capabilities regardless of whether the answers are useful. OpenAI absorbs most compute costs, while usage limits protect its margins. Users bear the risk of inaccurate answers. The simplicity is appealing, but as Kyle Poyar argues, flat subscriptions can become unprofitable when a small group of users consumes disproportionate resources.
GitHub Copilot (per seat): Its Business plan costs $19/user/month, making expenses predictable for companies. However, developers who barely use Copilot cost the same as those relying on it daily. Customers still bear the responsibility of verifying generated code, and additional usage can increase their bills. The model is easy to understand but poorly reflects the actual value delivered.
Intercom Fin (per resolved customer conversation): At $0.99 per resolved conversation, customers pay when Fin delivers a defined result. This aligns pricing more closely with value, while Intercom absorbs the cost of unsuccessful attempts. However, a conversation ending doesn’t necessarily mean the problem was solved correctly. The business still bears the consequences of a bad answer.
For our team’s product, which simulates probable outcomes of product decisions, I’d choose a team subscription with a fixed simulation allowance. Charging per successful decision would be nearly impossible because outcomes are uncertain and difficult to attribute. Per-seat pricing would discourage stakeholder participation, precisely what our product tries to encourage. Instead, a predictable subscription with transparent allowances and optional additional simulations would balance accessibility, customer value, and computational costs.
