AI resolved, self-served, needs a human: how conversations are counted

Every conversation in your analytics lands in exactly one of four outcomes. Understanding what each one means — and the order they're decided in — is the key to reading the whole dashboard, because the same four buckets drive your AI-handled rate and the "Where every ticket went" bar.

The four outcomes

Needed a human. A real teammate got involved. This is counted whenever a person sent a reply, the conversation was assigned to someone, or the customer explicitly asked to talk to a human.
AI resolved. The conversation was resolved and the AI had sent at least one reply — and no human signal was present. The AI answered, and that was that.
Self-served. The customer was shown a suggested help article and never needed to escalate to a full conversation. They found the answer themselves.
Still open. None of the above has happened yet — the conversation is in flight.

Order matters: first match wins

The outcomes are checked in a fixed order, and the first one that fits is the one that sticks:

Needed a human is checked first. So if you assign a conversation to a teammate — or a teammate replies — it counts as human-handled, even if the AI also drafted or sent something earlier. Human involvement always takes precedence.
If there's no human signal, and the conversation is resolved with an AI reply, it's AI resolved.
If it isn't resolved but the customer was shown an article and never escalated, it's self-served.
Otherwise it's still open.

This is why simply assigning a conversation moves it out of the AI buckets — assignment is treated as "a human has this." It's signal-based, never guessed from the channel.

How the AI-handled rate is built

Your headline AI-handled rate is straightforward once you know the buckets:

(AI resolved + self-served) ÷ all conversations, rounded to a whole percent.

In other words, it's the share of conversations that reached a good ending without a human writing a reply.

One important exclusion

On the Overview page, conversations created through the API are left out of these counts, along with spam. That keeps your headline rate focused on real, human-facing channels — email, contact form, and chat. The Channels page is the exception: it includes API traffic on purpose, so you can see every source. If the two pages ever look like they disagree, this is usually why.

Reading it in practice

A high self-served slice means your help articles are doing real work — customers are answering their own questions. See Turn your analytics into a better knowledge base.
A high needed a human slice isn't bad — it's your team doing what only people can. But if it's higher than you'd like, it often points to gaps the AI could fill with better knowledge-base coverage.