How do I generate an AI performance summary?
Required Feature Flags
The following feature flag is required to use this feature:
Feature Flag | Technical Name | Description |
AI Coaching |
| Enables AI-generated performance summaries on 1-to-1 meetings and agent profiles |
Required Permissions
Access depends on where you generate the summary from:
From a 1-to-1 meeting — only the facilitator assigned to that meeting can generate, edit, or delete the AI summary for that session. The meeting participant can view the AI Performance Summary tab but cannot take action on it.
From an agent's profile — you can generate a summary for any agent in your reporting hierarchy, or for yourself. Bots and agents outside your reporting line don't show the button. Agents only see profile summaries that have been shared with them, plus any they've generated for themselves.
Overview
The AI Performance Summary pulls together evaluation results, feedback, conversations and prior 1-to-1 notes from a chosen period, then drafts a written summary to help you prepare for a coaching conversation. It surfaces highlights, trends, examples and suggested talking points so you walk into the conversation with a clear structure.
You can generate a summary in two places:
Inside a 1-to-1 meeting — from the AI Performance Summary tab on the meeting itself. Best when you're prepping for a scheduled session.
From an agent's profile — use the Generate AI Performance Summary button in the profile header to create a detached summary that isn't tied to a 1-to-1. Detached summaries are saved to the AI Performance Reports tab on the profile (10 per page, newest first). This is the right path if your team doesn't run formal 1-to-1s, or if you want a snapshot of recent performance outside the meeting cadence.
Both paths use the same modal and produce identical output.
Generate a summary from a 1-to-1 meeting
Open the 1-to-1 meeting from your meetings list or calendar.
Select the AI Performance Summary tab.
Click Generate summary. The Generate AI Performance Summary modal opens.
Choose a Data Period:
- Since last 1-to-1 meeting — uses the date of the participant's previous completed session as the start date. Disabled if there's no previous session, and not available on profile summaries. - Last week — the last 7 days. - Last 2 weeks — the last 14 days. - Last 4 weeks — the last 28 days. - Last quarter — the most recently completed calendar quarter. - Custom range — pick any start and end date. The modal blocks generation if the range is incomplete or the end date is before the start date.
Review the Data Sources panel. It shows what's available for the selected period:
- Evaluation Results — up to 300 evaluations are analysed. If more are available, the modal shows "(max 300 will be analysed)". - Feedback, Actions & 1-to-1s — comments, actions and prior 1-to-1 records. - Conversations — up to 500 conversations are analysed. If more are available, the modal shows "(max 500 will be analysed)". - Previous 1-to-1 — goals and action items from the participant's last meeting.
Optionally narrow the data by Scorecard Type:
- All scorecards — the default. - Manual — evaluations completed by a human reviewer against a fully manual scorecard. - Blended — evaluations against scorecards that mix AI-scored and human-scored line items. - Automatic — evaluations scored end-to-end by AI.
Switching the scorecard type refreshes the evaluation count in the Data Sources panel so you can see how many evaluations the summary will be built from before you generate.
Click Generate. While the summary is being prepared, you'll see an "Analyzing Performance Data" message.
Generate a summary from an agent's profile
Open the agent's profile.
Click Generate AI Performance Summary in the profile header. The same modal opens — pick a Data Period and review the Data Sources as above.
Click Generate.
The finished summary opens in a slide-over and is saved to the AI Performance Reports tab on the profile. Open that tab to find previous detached summaries — click any row to reopen the report.
Detached summaries don't reference a previous 1-to-1, so the Previous Session block doesn't appear when you generate from the profile.
What the summary includes
Each summary contains:
Key performance highlights and areas for improvement
Trend analysis and progress tracking
Specific examples and metrics
Suggested talking points and action items
Conversation starter questions
Line item pass rates
Strongest and weakest line items show the agent's pass rate inline next to the line item name — for example, Active Listening & Responsiveness — 92%. The percentage matches what you'd see for that line item in the Line Item Performance KPI report.
A few rules of thumb:
If the agent's evaluations span more than one scorecard, the scorecard name appears next to the line item so two scorecards sharing a line item name aren't blended into a single misleading percentage.
Pass rates are only shown when the sample is large enough to be meaningful — line items with a low applicable count don't render a percentage, just the name.
The pass rate excludes any evaluations where the line item was marked N/A.
Conversation Insights metric definitions
The Conversation Insights section shows pills for metrics such as xNPS, Agent Sentiment, Customer Sentiment, Resolution Rate, Repeat Rate and Average Handling Time. Hover (or tab to and focus) any pill to see a plain-English definition of what the metric measures — useful when you're walking the agent through the numbers in the meeting.
Actions from the previous 1-to-1
When the participant has a previous completed or acknowledged 1-to-1, the summary includes a Previous Session block. Inside that block, Actions from last 1-to-1 lists each action the agent left the last session with, paired with a status chip:
Completed — the action has been closed.
Overdue — the due date has passed and the action is still open.
On track — the action is still open but within its due date.
Improved — recent evaluation data suggests the agent is performing better on the behaviour the action targets.
Still developing — the action's target behaviour is still showing up as a development area in recent evaluations.
Each chip is paired with a one-line note explaining the evidence behind the call. If the prior session had no actions, the Actions from last 1-to-1 block is omitted.
Evaluation reference chips
Where the summary calls out a specific behaviour, the supporting evaluations appear as reference chips underneath the relevant section. Chips appear in:
Areas for development
Positive patterns
Each individual recommendation
Each chip is the contact reference for an evaluation that backs up the point. Hover the chip to see a short note explaining why that evaluation was chosen as evidence. Click the chip to open the evaluation in a new tab so you can review the conversation, the scoring, and the agent's feedback in context.
A few things to note:
Chips reference real evaluations from the data period you chose — they're filtered server-side, so a chip will never point at an evaluation that wasn't actually analysed.
Each section shows at most two chips, picked by the model as the strongest evidence for that point.
If the model couldn't find supporting evaluations for a section, the chips simply don't render — the narrative text still appears.
Chips work for voice and non-voice interactions in the same way.
Turn a recommendation into an action
Every recommendation card in the summary has a + Action button. Click it to open a pre-filled Create Action modal — the description is built from the recommendation (area, action and rationale) and the participant is locked in as the assignee, so you can't accidentally assign the action to someone else.
A few details worth knowing:
The modal title becomes Assign action to {participant name} and the assignee picker is hidden — this is by design for the recommendation flow.
The action's Origin is set to AI Coaching Report, which makes it easy to track in the Actions tab which actions came out of an AI summary.
If the summary was generated from a 1-to-1, the new action is attached to that session. If it was generated from the profile (detached), the action is attached to the AI Performance Report itself.
The + Action button is hidden while the summary is in edit mode and is only visible to the facilitator. Participants don't see it.
Actions you create this way feed back into the next summary — the Actions from last 1-to-1 block will progress-check them in the participant's next AI Performance Summary.
After the summary is generated
Once a summary is ready, use the menu in the top right of the summary panel to:
Print / Export as PDF — open the system print dialog to save or print a copy.
Edit — open the summary in edit mode, make changes, then save.
Delete — remove the summary.
- If the summary has actions attached to it (created from recommendation cards), the first delete attempt asks you to confirm the cascade — confirming removes the summary and the linked actions in one go. - If nothing is linked, the summary is removed straight away. - Deletes can't be undone.
What the participant sees
The meeting participant can open the AI Performance Summary tab but cannot generate, edit, or delete the summary. If the facilitator hasn't generated one yet, the participant sees the message "The facilitator can generate an AI-powered performance summary for this meeting."
On their own profile, agents only see profile-level summaries that have been shared with them or that they generated themselves — drafts written by managers stay hidden until shared.