DEV Community

Cover image for Outcome-Based Pricing for AI Agents: How Pay Per Resolution Actually Works
No Hire Needed
No Hire Needed

Posted on

Outcome-Based Pricing for AI Agents: How Pay Per Resolution Actually Works

Software has been sold per seat for twenty years. AI agents are breaking that. A growing number of customer support tools now charge for outcomes: each time the AI agent resolves a customer's question, you pay a set fee.

Why vendors like it

If an AI agent replaces part of a support person, the vendor wants to be paid like part of a support person. Tying price to resolved conversations does that, and it puts some risk on the vendor: if the AI resolves nothing, you pay nothing for it.

The definition is the product

"Resolution" is not standardized. Depending on the vendor it can mean the customer confirmed the answer helped, the customer did not ask for a human, or simply that the AI took part. Some vendors count outcomes, some count sessions. Before comparing per-unit prices, compare what each unit is.

Estimating your bill

  1. Count last month's repetitive questions (the ones an AI could answer from your docs).
  2. Multiply by an expected resolution rate. 50 to 60 percent is a reasonable first guess for a well-trained agent.
  3. Multiply by the per-resolution price.
  4. Double it for your busiest month.

At low volume this is usually cheaper than a flat AI add-on. At a few hundred resolutions a month, a plan with AI included can win by a wide margin.

What to ask before you sign

  • If a customer reads the answer and leaves, is that billed?
  • If they come back with the same question, are you charged twice?
  • Can you set a hard monthly cap, and does the AI stop or keep billing when you hit it?
  • Is there a log of every billed resolution?

We ran the full math for three small businesses at three volumes here: Pay Per Resolution AI Support: The Hidden Costs

Top comments (0)