Call QA on every call instead of a small sample

Sampled QA has a known weakness: the calls that matter most are rare, and a sample of a few calls per agent per month will usually miss them. A team that takes thousands of calls a week can only listen to a small share, so a missed identity check or an ignored complaint can sit unnoticed for weeks.

Checking every call changes the job of the QA team. Instead of choosing which calls to hear, reviewers start from the calls that already failed a check, with the quote that failed it. This post walks through a Call Compliance dashboard built with Agent Serve Dashboards that does that, panel by panel.

Getting transcripts onto the dashboard

The dashboard is created from the Call Compliance template, one of the templates in Dashboards. It includes a Transcripts panel where a team uploads call transcripts directly. It accepts txt, vtt, srt, json, md, docx, pdf and csv files, which covers what most call platforms export.

It also takes call audio: mp3, m4a, wav, ogg, opus, webm, flac, aac, and video files such as mp4, mov or mkv. You choose the transcription model, and speaker labels are optional, so the agent and customer can be told apart. A workflow with the Speech-to-Text block remains an option if recordings should be transcribed automatically as they arrive.

Each transcript is checked against a checklist you write in plain English. You can start from a preset (Financial services sales calls, Collections, Healthcare or Support QA) and edit it. Every check has a severity of low, medium, high or critical, and each call gets a pass, fail or not applicable per check, with a verbatim quote as evidence where one exists.

What each panel shows

The dashboard above is an illustrative example. Read it from top to bottom.

• KPI tiles. Calls checked, checklist pass rate, critical breaches and average QA score, each with a comparison to the previous period.
• Pass rate by week, team and agent. The chart answers whether things are getting better, and where. Drill from team to agent to see who needs coaching and who could run it.
• Severity and sentiment. A breakdown of failures by severity sits next to sentiment and escalation risk, so a rise in critical failures on angry calls stands out.
• Evidence table. The latest failed checks, each with the call, the check, the severity and the quote. This is the panel reviewers live in.

The same results also carry analytics fields: topics, intent, objections, next steps, talk ratio and a summary. These help with coaching even when no check failed.

The rule and metric behind the alerts

The template ships with two governed definitions so that every chart and alert counts the same way:

• A critical-breach rule that flags any call with a failed check of critical severity. Rules here are plain conditions over each row, so the definition is visible to anyone who asks.
• A breach-rate metric: the share of checked calls with at least one failure. Because it is a named metric, the weekly chart, the team tile and the monitor all read the same number.

Edit either one to match your QA policy. If your firm treats some high-severity checks as critical, add them to the rule rather than building a second one.

Monitors that put a person on the call

Monitors turn the dashboard from something people check into something that tells them. Two come with the template:

1. Any critical breach. Fires as soon as one call matches the critical-breach rule.
2. Breach rate above a threshold. Fires when the breach-rate metric passes 5% by default. Set your own figure.

Alerts show in the app and can go by email, or to a Slack or Teams channel through a webhook. Each alert opens when the condition holds and resolves when it clears, so the list of open alerts is also a work queue.

The Relay that drafts the review