AI customer support agents: per-message or per-resolution pricing?
retained text
Hi guys, I've been thinking about the economics and pricing model for AI agents in customer support, and I'm curious what others here think. A lot of AI customer support products I've looked at seem to charge around $1–2 per resolution, although the definition of "resolution" can vary quite a bit. We're building in this space too, and we currently offer two pricing models: **Per message** — pay based on the number of AI messages used. **Per resolution** — pay only when the AI actually resolves the conversation. We're trying to get the per-resolution model to around **$0.50 per successful resolution**. The interesting part is that the actual cost of running an AI support agent can vary a lot. A simple "where is my order?" request might only need a few messages and one API call. A more complicated case might require the agent to read a long conversation, search the knowledge base, look up an order, call multiple APIs, analyze an attachment, follow up with the customer, and potentially hand over to a human. That's why I'm not sure there's one obviously correct pricing model. Per-message pricing is transparent and directly tied to usage, but customers may not want to think about how many messages the agent needs. Per-resolution pricing is easier to understand and aligns the price with the outcome, but then you have the difficult question of what actually counts as a successful resolution. If the AI closes the conversation and the customer comes back an hour later, is that still a resolution? What about a case where the AI does 80% of the work but eventually escalates to a human? Or an agent that processes a refund successfully but the customer has another question afterward? For people building or buying AI agents for customer support: **Would you rather pay per message or per successful resolution?** And if it's per resolution, does \~$0.50 feel like the right price point, or is pricing less important than accuracy, integrations and the percentage of conversations the agent can actually handle end-to-end? Please share your thoughts and thanks in advance!
document fetched 2026-09-12 00:18 · canonical https://reddit.com/r/AI_Agents/comments/1wdmx3s/ai_customer_support_agents_permessage_or