Prerequisites
Analytics needs the right business, time window, and source calls.
Correct business
Confirm the selected business and agent before comparing counts or opening a call.
Completed calls
Performance insights appear after Kai handles calls. An empty dashboard is not a zero performance score.
Known change time
Use Since last update when reviewing a recent prompt or voice change so older calls do not blur the result.
Dashboard path
Start broad in Call Quality, then narrow down what needs attention.
- 1
Open Call Quality
Go to Dashboard → Call Quality and choose Last 7, 14, or 30 days, or Since last update. - 2
Read the health strip
Use it to see which area needs attention, then open the matching tab rather than treating the overall color as the diagnosis. - 3
Use Performance for trends
Review response speed, outcomes, and time patterns. The outcomes panel includes signals such as silence timeouts, lead capture, transfer success, and conversation depth. - 4
Use the other tabs for evidence
Evals check whether Kai followed its instructions. Call Reviews shows conversation findings, Regression reruns saved scenarios, and Agent Testing runs a scripted practice conversation.
Keep the selected time window visible when comparing results. Changing the window changes the calls included in the dashboard.
Source evidence
Open the conversation behind the score before editing the agent.
- Identify the exact warning, failed instruction check, review finding, or saved Regression test you are investigating.
- Open the associated call or use Dashboard → Calls to find it by caller, number, or transcript.
- Read the summary and full transcript, then listen to the recording when it is available.
- Confirm whether the issue came from the instructions, response speed, caller behavior, routing, missing business knowledge, or a missing recording or transcript.
- Compare more than one real call before treating a single unusual conversation as a trend.
- Write down the narrow behavior that should change and the behavior that must remain unchanged.
Coaching loop
Test the proposed fix, then wait for live calls to prove it.
- 1
Make one focused change
Update the relevant instruction, knowledge, voice, route, or tool instead of rewriting unrelated agent behavior. - 2
Run Agent Testing
Use a scripted scenario to catch an obvious failure before a customer reaches the new behavior. - 3
Run Regression
Rerun the saved scenarios and inspect any failed transcript before accepting the change. - 4
Verify real conversations
Switch the dashboard to Since last update, collect completed live calls, and compare their source transcripts and recordings with the intended behavior.
Troubleshooting
When the dashboard does not explain the call
No call data yet
Confirm the selected business and date range. Verify the phone path with a real completed call before diagnosing analytics.
Counts changed unexpectedly
Check the time window and whether Since last update moved after a prompt or voice change, then refresh the page.
A warning has no obvious cause
Open the underlying calls and compare transcripts, recordings, routing, and end reasons instead of changing the threshold first.
A saved Regression test fails
Open that scenario’s run transcript, correct the narrow failure, and rerun it before running the full suite.
A practice test passes, but customers struggle
Treat real calls as the strongest check. Add the missed customer behavior to your saved tests.
A transcript and recording disagree
Use the source audio when available and account for transcription errors before coaching the agent.
Honest limits
A dashboard number is evidence, not a business outcome.
Quality metrics summarize the calls and window that were available to the dashboard. They do not by themselves prove a booked appointment, qualified lead, sale, or revenue. A scripted agent test is not a real call to your business number, and a saved test passing does not replace post-change review of real conversations. Missing recordings or transcripts can also limit what the dashboard can explain.