ChatGPT: access
ChatGPT sells an allowance of model use. Plans run from $20 for Plus to $500 for Pro. OpenAI carries compute risk, because a heavy user costs more to serve at the same price. The customer carries quality risk and pays the same for wrong answers. The predictable bill is clever, but price has no link to value, so light users overpay and heavy users drain margin.
GitHub Copilot: a person
A Business seat costs $19 a month, including AI Credits that pool across the organization. The buyer pays for idle seats and wrong code and shares compute risk through overage. Pooling is clever, because heavy and light users offset each other. The flaw is that seats track headcount, while AI that works well reduces headcount.
Intercom Fin: an outcome
Fin charges $0.99 per resolution, plus $29 per helpdesk seat. Unresolved conversations are free, so Intercom absorbs the cost of failures and earns more only when Fin works. The flaw is that Fin counts customers who go quiet as resolved. This is the kind of ethical assumption Torres says teams skip: the vendor profits from silence that may be a customer giving up.
What we would do for Roadmap Decision Brief
Outcome pricing does not fit, because our brief stops short of a decision and leaves no outcome to count. We would charge per PM seat with a pooled brief allowance, as Copilot does. Our interviews show the value lands on one person, the PM who saves about 2.5 hours per review, and buyers budget by PM headcount. The pool caps our cost when a PM uploads 40 documents instead of three. Following Torres, we treat price as a viability assumption and test it with signed pilots, not survey answers.
