I’d probably feel differently about a bad AI answer depending on whether I was paying for the tool or paying for it to get something done.
ChatGPT charges for access through subscription tiers. The monthly bill is predictable, but I pay the same whether its answers save me hours or send me in circles. OpenAI covers the computing costs within my plan’s limits; I’m left checking answers and fixing mistakes.
GitHub Copilot charges for each person who has access. A company pays for a developer’s seat whether they use it constantly or barely touch it. The company also absorbs the time spent reviewing and correcting generated code. GitHub covers the included computing usage, though its current plans also use AI credits and allow additional purchases.
Intercom Fin charges for an outcome: $0.99 per resolved customer conversation. That gives Intercom more reason to make the AI useful, since unsuccessful resolution attempts still cost it money. But there’s a catch: a customer leaving without asking for more help can count as a resolution. Someone might simply give up, leaving the business with both a charge and an unhappy customer. Bills also rise as more conversations are resolved.
That’s what stood out to me in Poyar’s reading: charging for outcomes sounds fair until people disagree about what an outcome means.
For our team’s product, I’d start with a subscription that includes a clear usage allowance. Customers could budget for it, and we could manage computing costs. I’d consider outcome pricing later, once we can reliably measure success. Before charging someone for a finished job, I’d want to know they actually consider it finished.
