Skip to main content
Spotlight scores your closed conversations the way a QA manager would and rolls the results up into one Performance score, six quality areas, and a list of the conversations most worth reading. Use it to find out where an agent, or a teammate, loses customers, and to check whether a change you published made things better.
Spotlight is free during beta: the page shows a Free during beta badge, and scoring does not use credits.

​
Before you start

  • Spotlight needs at least 10 evaluated conversations in the view you are looking at. Until then the page shows Getting started with a progress count (“N of 10 evaluated”). If your filters leave fewer than 10 it reads Not enough conversations match these filters, shows how many match, and offers Clear filters.
  • A conversation is evaluated after it is closed, if the customer sent at least two messages. Popcorn looks for newly closed conversations every 15 minutes, so scores appear shortly after a conversation is closed. Conversations older than 120 days are not evaluated.
  • Every participant is scored separately: the AI agent and each teammate who replied. A conversation with both counts once in the conversation totals.

​
Open Spotlight and filter

1

Go to Spotlight

Click Spotlight in the sidebar, under Your agent. The last 7 days are shown; switch to 14 days, 30 days or Custom (pick From and To, then Apply) at the top.
2

Narrow the view

Use Everyone to pick one agent or teammate (each entry shows its number of evaluations), All types for Sales, Support or General, and All sentiments for Positive, Neutral, Negative or Frustrated.
3

Choose who handled the conversations

All, Agent, Agent and team and Team filter by who replied: everyone, only the AI, both, or only your team.
Comparisons such as “vs previous period” always use the period of the same length immediately before your date range. Conversations are placed in a range by the date they started.

​
Performance score

The Performance score is the average of all evaluations in the range, out of 5, with the change against the previous period next to it and the number of conversations it covers. Score over time plots the daily score across the range (hourly when the range is a single day) and appears once there are at least two points; until then it reads “Not enough data to draw a trend yet.” Each evaluation’s score is a weighted average of the six areas below.

​
Score by area

Each area is scored from 1 to 5, and each card shows the average with its change and a small trend line. The scale is deliberately strict:
  • 3 means the job was done adequately. A correct answer with no problems is a 3, not higher.
  • 4 needs clear evidence of doing more than expected; 5 is exceptional and rare.
  • 2 means something went meaningfully wrong; 1 means the customer experience was harmed.
  • Short exchanges (a greeting, a single question and answer) cannot score above 3 on Understanding, Resolution, Communication or Helpfulness.
  • Instructions compares the AI’s replies with the agent’s instructions. Teammates are not judged on instructions and always get a 3 there.
  • A handover is not automatically a failure: handing over something the agent could have answered from its knowledge or tools scores low, handing over something it genuinely could not resolve scores as adequate.

​
Key figures

​
Breakdowns

How conversations ended
  • Confirmed: the customer explicitly said the issue was solved.
  • Assumed: a concrete answer was given and the customer kept engaging, without confirming.
  • Escalated: the conversation was handed over to someone else.
  • Abandoned: the customer stopped replying, or nothing was resolved.
Customer sentiment counts Positive, Neutral, Negative and Frustrated customers, judged from the customer’s messages during that participant’s part of the conversation. Conversation types splits conversations into Sales (browsing or buying), Support (a problem or an existing order) and General (greetings, mixed or unclear intent). Both show a stacked chart over time once the range has at least two points (two days, or two hours when the range is a single day).

​
Conversations to review

The table (“The lowest-scoring conversations, which may need a look”) lists up to ten conversations with the lowest score in the range, one row per conversation. Each row shows the Score, the Main issue (the evaluator’s note on the lowest-scoring area), the Sentiment, the Type and the Date. Click Review to open the conversation in the Inbox, or View inbox to go to the Inbox itself. When nothing scored low, the section reads All conversations look good. In the Inbox, a closed conversation that scored below 3.5 shows a Quality score card in the conversation details with the main issue; Details expands the score of every area and Hide collapses it.

​
Acting on what you see

Filter by Everyone to compare agents, or to see how teammates score after a handover. After publishing a change, compare the score of the following period with the one before it. Per-version outcome figures are also listed in the agent’s Version history; see Test and publish.
  • Check replies before sending: the real-time check that retries weak replies before they are sent. Spotlight scores whole conversations after they close.
  • Reports: volumes, response times and handover rates.