I'm researching how outcome-based AI agents work in real production environments.

I'm talking about #AiAgents like #Intercom #Fin, #Zendesk #AI, #Salesforce AI, #Ada, #Decagon, #Sierra, etc.

On paper, the model makes a lot of sense:

AI interaction → successful outcome → customer gets value → you pay

But I'm curious about the messy real-world cases.

For example:

- AI answers the question, but the customer comes back N hours later because it wasn't actually solved while no Fallback human is available or inactive.

- Customer stops replying, so the conversation gets considered “resolved.”

- AI completes most of a workflow but fails at the final step.

- AI gives an answer that technically satisfies the configured rules but doesn't actually solve the customer's problem.

- A conversation gets billed as a resolution and the support team disagrees.

- Multiple AI actions happen in one conversation — how do you decide what the actual outcome was?

If you're using an outcome-based AI agent in production:

Do you trust the provider's definition of “resolved”?

And more importantly:

Have you ever challenged or disagreed with a billable outcome?

I'd be particularly interested in hearing from people using Fin or Zendesk AI, but experiences with any outcome-based AI agent are useful.

Not looking for vendor marketing — I'm trying to understand what actually happens in production.

What is the biggest pain point you've experienced?