Classifying existing conversations

Short answer: automatic classification only looks back seven days. At the bottom of Settings → Topics two buttons classify your history, so your report covers a real period. Classify history takes the closed conversations of the last ninety days. Classify conversations without a topic again gives the ones still on Not classified a fresh chance against your full list.

The two buttons

Classify history takes the closed conversations of the last ninety days, newest first, up to a maximum of two thousand. Conversations that already have a classification are left alone.

Classify conversations without a topic again first removes the mentions sitting on Not classified and then starts the same round. That's the button you use after adding topics: conversations judged while your list was still short get a fresh chance.

Mentions with a topic are touched by neither button.

What happens in the meantime

The round runs in the background. On a full history that takes a few hours. You can close the page, and the counter is simply there again when you come back.

Three things you won't see on screen, and they explain what you notice in the meantime:

• While the round runs, ordinary automatic classification for your workspace is paused. Fresh conversations sit without a classification for a while. They get picked up on their own once the round finishes, so nothing is lost.
• If the round gets interrupted, for example by a restart, it resumes where it left off. It doesn't start over and doesn't classify anything twice. The counter is recalculated on resume, so it won't lie to you.
• On a large history the oldest part is missing. The limit of two thousand cuts from the back, so your recent period is complete and your oldest one isn't. The report tells you from which date the data is available.

There's no way to stop it. A round that started runs out, or you wait for it.

What it costs

Every classified conversation is an AI call and counts towards your usage. So a full round is two thousand calls in one go. It doesn't count as a conversation and doesn't touch your conversation limit.

If you work with your own key at the AI provider, that runs on your credit. In that case the warning on the screen isn't a formality. See The AI agent: billing mode, limits, and usage.

When to do this

There are two good moments. The first is after filling your list of topics for the first time, see Managing topics. The second is after approving or assigning a proposal, see Having topics suggested.

Beyond that, preferably not. It costs usage every time, and ordinary classification keeps up on its own afterwards.

Good to know

A conversation whose classification somebody corrected by hand is touched by neither button, not even when that mention sits on Not classified. That's exactly the intention, and it explains why some conversations stubbornly stay without a topic. See Adjusting a conversation's classification.

The second button really deletes. The mentions without a topic disappear straight away, even when the new round doesn't find a topic for them either.

And classified conversations land in the report on their own date, not on today. So a round today fills your chart of the past months instead of giving you one spike.

Also searched as

"classify old conversations after all", "my report is empty for previous months", "can I categorise retroactively", "how long does classifying my history take", "have conversations without a topic judged again"

See also

• Managing topics and switching classification on
• Having topics suggested
• Topics and question types: what your customers write about