PSP 7, Three Ways to Price AI

The main difference arises from what is actually being charged: outcome, person, or access.

ChatGPT: By using a monthly subscription model, they are basically charging for access. This makes billing relatively easy, but there’s the downside of risking spending a lot of money on a model for a project only to get a broken output. OpenAI has the burden of setting the monthly/weekly plans optimally so that service is still beneficial and financially sound cost-wise. Outcome quality is on the customer to make sure.

 

Copilot: Charging per seat allows customer companies to budget by considering the headcount (again very predictable for them), but also incur the risk of paying for underused or, in some cases, unused seats, paying more than the value it provides. Still, outcome quality is on the customer to make sure.

 

Fin: A less common way that Intercom chose is charging per outcome, including resolved conversations, meaning here Intercom takes the cost of failed attempts. While on a higher level, there is less burden on the customer to make sure they are getting their money’s worth, a customer can leave a conversation without making the decision to ask for further help (quit, not force), and this could count as a resolution, and hence the consumer still has a burden of risk.

 

For our product, the plan that makes the most sense, in my opinion, is similar to Copilot’s. However, I believe we should have “viewers” like employees, with free access and a team price per editing team (decision makers who log their decisions to the platform and make use of our features like suggestions). We couldn’t use Intercom’s system because we don’t have an easily verifiable outcome (nothing that “resolves”), and until analysis becomes very costly, charging per usage wouldn’t be worth the revenue/billing unpredictability.

Avatar

About the author

Leave a Reply