Open the report
Start with the right business and time period.
You need access to the selected business and completed calls during that period. Some panels stay empty until KaiCalls finishes analyzing the calls.
- 1
Open Call Quality
In the dashboard, choose the correct business and open Call Quality. - 2
Choose a comparison window
Use Last 7 days, 14 days, 30 days, or Since last update. The last option excludes calls before the most recent prompt or voice change. - 3
Read the health strip first
Use healthy, warning, and critical segments to find the panel that needs attention, then inspect the underlying calls. - 4
Refresh after processing finishes
A newly completed call may need analysis before every quality field appears. Refresh rather than assuming a blank panel is a zero.
Performance tab
Separate speed, call behavior, cost context, and time patterns.
Latency
Review pickup and response speed, the first-message latency chart, threshold bands, and agent-level comparisons.
Call outcomes
Review available silence timeouts, lead capture, transfer success, who spoke last, and average turns. Alerts point to values that may need attention.
Call costs
See available cost per call, total spend for the period, model or voice usage, and estimated future spend. Future spend is based on the usage seen so far.
Time patterns
Compare call volume by hour and day, peak periods, and available business-hours versus after-hours quality signals.
Use the summary to spot a pattern, then open the calls behind it before changing your agent. An average can hide differences between callers, call routes, or calls that have not finished processing.
Outcomes & coaching
Know the difference between Kai’s best estimate and a confirmed result.
Kai’s call analysis
Kai can estimate intent, buying signals, outcomes, summaries, and coaching context from a transcript. When available, a confidence level shows how certain the estimate is.
Confirmed outcomes
A badge or reference tied to the booking or sales system is stronger than an AI estimate. Do not call an estimated booking or sale confirmed.
Coaching suggestions
Use cited call moments and repeated patterns to decide what to listen to. A suggestion is a reason to review the call, not automatic proof that your instructions are wrong.
Unavailable analysis
When the required analysis is absent, KaiCalls can show no insights rather than inventing a result. Open the call and check whether processing completed.
Not every completed call has every score or insight. Availability depends on the call data, analysis pipeline, selected period, and whether the relevant evaluation ran.
Scores & practice calls
Use each test for the question it actually answers.
Available call scores
When a score exists, review the score, pass or fail result, and explanation for each check. Some calls will not have a score.
Practice call
Run a scripted conversation to check an agent before live traffic. It shows how that scenario went, not how a real caller or the live phone network will behave.
Repeatable test suite
Select an agent, run one scenario or Run All, and inspect the transcript. Passing these tests lowers risk, but real conversations can still behave differently.
Configuration versus live proof
A saved change is configured. Fresh completed calls show production behavior.
- The prompt or voice change appears as the latest agent update.
- A practice call or repeatable test completes and its transcript matches the scenario tested.
- The Since last update window contains fresh completed real calls.
- The calls behind the summary show the expected behavior and available quality details.
- Any claimed booking, transfer, or other confirmed outcome is clearly separated from an AI estimate.
Fix missing data
A blank or stale panel usually means the wrong business, no calls, or unfinished processing.
No calls in the window
Confirm the selected business and expand the date range. Since last update only includes calls after the latest prompt or voice change.
Calls exist but insights are empty
Open an individual completed call and check whether transcript and analysis data are available, then refresh after processing.
A test passed but real calls did not improve
Compare fresh real calls with the practice transcript. Caller behavior, call routing, audio, and phone network conditions are outside a scripted scenario.
A summary looks inconsistent
Inspect the calls behind it and check their confidence or confirmation details. Missing information can change which calls are included in a measurement.