Week 7 PSP: Who Carries the Risk When You Price AI?
Kyle Poyar’s Your Pricing Is (Probably) Broken got me thinking about what customers are really paying for when they buy an AI product. The more I looked at real examples, the more I noticed that every pricing model makes a quiet decision about who gets stuck with the bill when things go wrong.
ChatGPT:
ChatGPT sells flat subscription tiers. You pay for access, not for answers. I like that my monthly cost never surprises me, but I also pay the same amount on days when it confidently gives me something wrong. On the other side, OpenAI eats the cost when heavy users push right up to their limits every month.
GitHub Copilot:
Copilot charges per seat, so a company pays for every developer who has access. The catch is that seats and actual usage rarely match. A team might pay for someone who barely opens the tool, and engineers still have to review whatever code it writes. GitHub covers part of the operating cost, though usage limits keep that from getting out of hand.
Intercom Fin:
Fin only charges for qualifying resolved conversations. This one stood out to me because it’s the closest to paying for actual work instead of potential work. But it raises an awkward question: who decides what counts as “resolved”? Intercom takes on more risk when Fin can’t close a conversation, while customers could end up paying for resolutions their own users weren’t happy with.
Our Product:
For our team’s meeting productivity tool, I’d go with per-seat pricing plus a cheaper tier for occasional users. The whole point of our product is helping people run better meetings and actually follow through afterward. If we charged per meeting or per completed task, we’d basically be taxing people for using it, which works against what we’re trying to build. Poyar’s idea of lite seats shaped this choice. Someone who shows up to one meeting a week doesn’t need the same access as the person running the team, and pricing should reflect that without making the bill hard to predict.
The thing I’m taking away from this week: good pricing follows how customers feel the value, not just how much AI they happen to consume.
Sources:
Kyle Poyar, “Your Pricing Is (Probably) Broken”
OpenAI, “ChatGPT Pricing”: https://openai.com/chatgpt/pricing/
GitHub, “Copilot Plans and Pricing”: https://github.com/features/copilot/plans
Intercom, “Fin AI Agent Pricing”: https://www.intercom.com/pricing
