Measuring Aura AI's performance

The Performance page under Aura AI is how the agent is doing: conversations, resolutions, escalations and the questions it could not answer. The same panel also appears on Home, so you do not have to come here to see it.

Along the top: Last 7 days, Last 30 days, Last 90 days, Last 12 months and All time, with the range actually in effect written out beside them. On all time it reads "Everything since your agent was created".

Conversations

The first thing on the page is conversation volume over the range, as a bar per day. It is context for everything below — a resolution rate over eleven conversations is not a measurement.

The ten headline measures

They are grouped into three bands, and the bands are the argument.

Did the agent finish the job? — how conversations ended: resolved by the agent, handed to a person, or answered with nothing.

• Resolved by the agent — the share the agent closed without a person stepping in.
• Handed to a person — the share that raised a ticket or asked for a human.
• Answered with nothing — the share of questions the agent had no answer for.

The third is the one to act on. Escalation is a working outcome; a customer who asked something and got nothing is not.

What did people make of it? — only rated conversations count here, so read the volume beside the score.

• Satisfaction score — the average rating people gave, out of five.
• Ratings received — how many answers were rated at all.
• Ratings per day — rated conversations averaged over the days in the range.

A 4.8 over six ratings and a 4.2 over nine hundred are not comparable numbers, which is exactly why the volume sits next to the score rather than behind a click.

What did it take? — how much back-and-forth each conversation needed, and how long it ran.

• Questions per conversation — how many times someone had to ask before they were done.
• Time to resolve — how long a conversation the agent resolved took, start to finish.
• Conversation length — how long the average conversation ran, resolved or not.
• Trained answers used — times the agent answered from something you taught it directly.

Reading the week-on-week figure

Each card carries a change on the previous week, or says No week-on-week figure when there is none.

One honest limitation: a reported change of zero means either genuinely unchanged or no prior week to compare against, and the two are indistinguishable in the data. On a new agent, read zeroes as "not yet meaningful" rather than "flat".

If Aura AI ever reports a measure this page does not know, it shows it with its name tidied up and a note that the measure is newer than the page. Nothing is dropped for being unfamiliar.

The four lists

Below the cards, four lists, and this is the part that tells you what to do next.

What people asked about — the subjects that came up most.

What got handed on — the subjects that ended with a ticket or a person. Fix the top of this list and the escalation rate follows.

Where the answers came from — which of your sources the agent drew on, as a share, a count and a change. A source that never appears is one the agent is not finding, which is a different problem from a source that is wrong.

Questions with no answer — what people asked that the agent could not handle, with how many times each was asked. Every row has a Teach the agent link straight to Training data. This is the list to work from.

When there is nothing to show

Three distinct empty states, and it is worth knowing which you are looking at:

• "Your agent hasn't handled any conversations yet." — nothing has happened yet. Once it starts answering, its numbers appear here.
• "No conversations in this range." — try a wider range, or all time.
• "No metrics came back." — Aura AI answered and the reply carried no measures at all. Nothing is being hidden; there was nothing in it to show.