How to Read Call Quality

Open Dashboard > Call Quality. The page answers one question: is Kai working well? It rolls up health, performance, evaluations, agent tests, and regression suites for the time window you pick.

At the top right, pick a time period — Last 7 days, 14 days, 30 days, or Since last update (only calls after your most recent prompt or voice change). The Refresh button re-fetches everything.

The colored health strip just under the header shows overall status (healthy, warning, critical) with the count of issues — click any segment to jump to the relevant tab.

Default view. Shows recent prompt and voice changes, a trust window, and four nested tabs:

• Latency — how fast Kai picks up and responds. The chart plots first-message latency over time, with target (3s), warning (4s), and critical (5s) bands. The agent table next to it ranks each agent by speed.
• Outcomes — silence-timeout rate, lead capture rate, transfer success, who spoke last, and average turns per call. Key Insights flags any metric outside healthy ranges in plain English.
• Model & Costs — average cost per call in cents, total spend for the period, model and voice usage breakdown, and projected monthly cost.
• Time Patterns — heatmap of call volume by hour and day of week, peak hours, and a business-hours vs. after-hours quality comparison.

If a critical or warning threshold is crossed, an alert banner appears at the top of this tab so you do not miss it.

Every call is automatically scored against your prompt — did Kai follow instructions? The Evals tab shows pass rate, average score, total evals, and a row for every call with its score and pass/fail status. Click a row to see exactly which checks passed and which failed, with an explanation for each.

Run a synthetic call against your agent to see how Kai handles a scripted scenario before a real caller does. Useful for sanity-checking a prompt change without waiting for live traffic. The panel keeps state between tab switches so a long test run will not be interrupted.

Regression evals are canned conversations that gate prompt changes — if you change a prompt and it breaks one of these, you find out before customers do.

Pick an agent from the dropdown. If no scenarios exist yet, Kai auto-provisions a starter set based on your business. Each row shows the scenario name, last run status (pass / fail), and a button to re-run that one scenario. Use Run All to fire the whole suite at once.

Click any scenario to open a drawer with the full run transcript so you can see exactly where Kai went off-script.

• Learning Agent — Kai's suggestions for fixing what the quality dashboard surfaces
• Call History — drill into individual calls referenced in the eval results
• Customize Your Agent — change the prompt or voice and watch the period reset