ChatGPT charges for access. Its subscription tiers offer different levels of model access and usage. Users still pay when answers are wrong and spend time checking them. OpenAI bears compute-cost variability within included allowances, while usage limits and paid extras help control that exposure.
GitHub Copilot charges for a person. Its per-seat model licenses individual developers. Employers pay for assigned seats even when suggestions are unused or buggy. GitHub bears included compute costs; customers bear debugging costs and the consequences of faulty code. Additional AI usage can also increase customer spending.
Intercom Fin charges for a measured outcome. Resolution pricing shifts more performance risk to Intercom where unsuccessful attempts consume compute without generating a resolution fee. However, a customer leaving without requesting further help can count as an “assumed resolution.” Businesses still face customer-trust damage from incorrect answers and variable bills as outcomes accumulate.
For our product, Pip, I would start with a monthly subscription per PM, including a clear recording-hour allowance. Pip currently prepares drafts that users review, edit, and send themselves. Charging per “completed task” would make attribution difficult. A subscription gives PMs predictable spending, while recording limits help us manage transcription and generation costs.
