ChatGPT charges for access. Its tiers (Plus at $20, Pro from $100 a month) buy higher usage limits and better models, not results. OpenAI carries the compute risk within each cap, so a heavy user is expensive to serve. The user carries the risk of a wrong answer: they pay the same either way.
GitHub Copilot charges for a person. Business is $19 per seat per month and Enterprise is $39, each with an allowance of premium usage and overage beyond it. The buying company carries the adoption risk, since an unused seat still costs $19. GitHub shifts some compute risk back through the usage allowance.
Intercom Fin charges for an outcome: $0.99 per conversation it resolves. Here the vendor carries most of the risk, because a wrong or unresolved answer earns nothing. The catch is that Intercom defines what counts as resolved, and customers report bills rising when “assumed resolutions” are counted.
Our product, an AI knowledge base that makes it easy for PMs to share and access information across teams, fits a flat subscription best. The value is ongoing access to a trusted summary, and there is no clean outcome to bill against: you cannot count “stayed informed” the way Fin counts a resolved ticket. A per-seat team plan is the natural second step when companies buy it for a whole product team. Outcome pricing would be tempting to pitch but not too possible to measure honestly, and per-use pricing would punish the habit we want to build.
