# Contact centre QA: a week of call review compressed into hours of checking

Source: https://getstartedwithai.net/case-studies/contact-centre-qa-week-of-work-to-hours
Site: Get Started With AI
Tagline: Build AI automations that actually run your business.
Audience: solo professionals, freelancers, consultants and AI implementers in training
Contact: hello@getstartedwithai.co.uk

This Markdown companion is generated from the prerendered page for AI crawlers.
# Contact centre QA: a week of call review compressed into hours of checking

A 120-agent contact centre QA team reviewed roughly two percent of calls. The build moved review to a hundred percent, with humans focused on the cases the automation flagged. Eighteen months in, the build is still running under the original named owner.

## The before

QA reviewed two percent of calls a week. Each review took an analyst about 12 minutes including notes. The pattern caught some issues; what it missed was anyone whose calls were rarely sampled. Coverage was uneven by design, which everyone knew.

## The build

A transcript pipeline was set up with Whisper for transcription and a Claude Project for scoring against the QA rubric. The rubric was the existing one the team had used for years; we did not change the standard, only the coverage. The output landed in a triage queue ranked by score.

- Every call transcribed and scored automatically.
- A ranked queue of cases for human review, lowest score first.
- Sampling kept on the high-score cases, so the team still saw a slice of good work.
- The original QA rubric used unchanged.

## The governance layer

The build was embedded with a written governance document, a named QA lead and a monthly review. The Four Breaks were checked every month for the first six months, then quarterly. Eighteen months in, the build is still running.

## The outcome

- Coverage moved from 2 percent to 100 percent of calls.
- Human review time dropped from a week to hours, focused on the cases the automation flagged.
- Two coaching patterns the team had not seen before were surfaced inside the first month.
- No drift over 18 months under monthly governance.

## See this kind of build run live

The free webinar walks through the WIRE Method end to end on a real audience workflow. 60 minutes, no upsell. Bring one repeatable task; leave with a mapped automation.

### Live build session

The next live build session covers the same shape as the case studies on this page.

[Join the next session](/webinar)

## Questions people ask before they start

### Did agents trust the AI scoring?

Trust came from running the AI scoring side by side with the QA team for the first month. After comparing scores on the same calls, the agreement rate was high enough that the QA team felt comfortable signing off on the build.

### Was the original rubric kept?

Yes. The build does not change the QA standard, only the coverage. The team did not need to learn anything new; the AI just applied the rubric they already used.

### Where did the freed QA hours go?

Coaching. The team spent the reclaimed hours on agent development instead of sampling calls. That pattern is the Talent Redirect: hours freed by automation flow back into higher-value work, not into reduced headcount.