ChatGPT sells access. $0, $8, $20 or $100+ a month buys higher limits and better models. Whether the answer was any good is your problem.
GitHub Copilot sells a person: $10–$100 per user per month. But each seat now carries a monthly budget of AI credits ($15 on Pro), so the seat is quietly becoming a metered allowance.
Intercom Fin sells an outcome: $0.99 per resolved conversation. If Fin can’t answer, or the customer asks for a human, you pay nothing.
| Pays for | AI is wrong | AI is costly to run | |
|---|---|---|---|
| ChatGPT | Access | Buyer | OpenAI, capped by usage limits |
| Copilot | A seat | Buyer (devs review the code) | GitHub, until credits run out |
| Fin | An outcome | Intercom | Intercom |
Fin can only do this because “resolved” is easy to count. Most products can’t define success that cleanly.
What fits Accord (Our product)
Accord should stay per seat, at our $38/user/month pilot price, with a usage cap.
Our outcome is “a decision the team agreed on,” which is hard to verify. Charging per approved decision would also reward PMs for pushing decisions through, which is the opposite of consensus. And Accord’s value grows with the number of reviewers, not the number of AI calls.
The cost risk is real, though. Every AI recommendation burns tokens. Like Copilot, we should include a fair-use allowance per seat instead of promising unlimited AI.
