ChatGPT charges for access. Flat subscription tiers unlock different capabilities and usage allowances: Free is $0/month, Plus is $20/month, and Pro is $500/month. My takeaway is that customers bear the risk of poor results, while OpenAI bears compute costs within the included allowance, protecting itself through limits.
GitHub Copilot charges for a person. Its per-user subscription ties revenue to developers equipped with the tool, but it doesn’t necessarily care about the correctness of the code. Employers bear the risk of unused seats and time spent fixing mistakes. GitHub bears compute costs within included usage, while credits and additional purchases limit its exposure to heavy workloads.
Intercom Fin charges for an outcome. A resolution costs $0.99, shifting more performance risk to Intercom: unsuccessful attempts still cost compute. But “resolved” can mean the customer left without requesting more help. Customers still bear reputational damage from bad answers and uncertainty about monthly spending.
For our product, I think we should use seat-based pricing. While Kyle Poyar warns that seat-based pricing is becoming less effective over time due to AI, I think that it is a realistic choice for us, given that we want to integrate our product with Slack, which uses the same pricing type. Slack also charges different tiers per seat, depending on how much a person uses it, which feels appropriate for our product.
