Written by the CustomerEagle team
AI support tools are converging on the same pricing idea: pay per conversation the AI resolves. It sounds fair — you pay for outcomes, not seats idling overnight. Whether it actually is fair depends entirely on one question: what counts as "resolved"?
The three definitions you'll encounter
- Generous (vendor-friendly): any conversation the AI replied to that the customer didn't escalate. Silence counts as success — you pay for customers who gave up.
- Moderate: conversations the AI answered where the customer didn't reply negatively within some window.
- Strict (buyer-friendly): zero human replies in the conversation, positive signal required for abandoned chats, and automatic reversal if the conversation reopens.
The differences are not academic. On the same ticket stream, a generous definition can report a far higher resolution share than a strict one — which means a larger bill for the same actual outcome.
Questions to ask any vendor
- Does a conversation still count if my agent sent even one message in it?
- Does a customer who silently abandons the chat count as resolved?
- What happens if the customer comes back the next day — is the resolution reversed?
- Can I see and audit every counted resolution with its full transcript?
We publish our own answers on how we measure AI resolutions — zero human replies, unengaged conversations closed without counting, and a three-day settling window that automatically reverses a resolution if the conversation reopens. Hold any tool you evaluate, including ours, to that standard, and check it against what each plan includes. For what the other vendors publish, see how five vendors define an AI resolution. And because a resolution is not the same event as a deflected contact, read deflection rate versus resolution rate before you accept either number in a vendor report.
Frequently asked questions
An AI resolution is a support conversation that an AI agent handled from start to finish without a human agent replying. Vendors bill per resolution instead of per seat, so the exact definition — especially how abandoned conversations and reopened conversations are treated — determines what you actually pay.
Under a strict definition it does not. In CustomerEagle, conversations the customer never engaged with are closed by timeout without being counted, because silence is not evidence that the answer was correct.
It should be, and in CustomerEagle it is. Every resolution has a three-day settling window: if the customer reopens the conversation, a human takes over, or a negative rating arrives, the resolution is removed from usage and credited back if it was already invoiced.
Not automatically. Per-resolution pricing shifts cost from headcount to volume, so it favours teams whose ticket mix is highly repetitive and penalises teams whose volume spikes without a matching knowledge base. Model both against your own ticket volume before switching.
