What is support quality assurance?

Why it matters

QA is the only mechanism that catches quality drift before customers report it. It matters more once AI is answering, because AI errors are systematic rather than random: a human agent makes one mistake, an ungrounded model makes the same mistake five hundred times before anyone notices.

What to know

• Sample deliberately: a mix of random conversations, escalations, low satisfaction scores, and reopened tickets.
• The scorecard should be short. Long rubrics produce inconsistent scoring and reviewer fatigue.
• Calibrate reviewers regularly, or scores measure the reviewer more than the conversation.
• AI answers need the same review as human ones, sampled weekly, checked against source documentation.
• QA findings should feed documentation and process changes, not only individual coaching.

An example

Example: weekly review of twenty AI-resolved conversations catches a pattern where a policy article is being read as applying to all plans. One documentation edit fixes several hundred future answers.

Common mistakes

• Using QA scores punitively, which makes agents optimise for the rubric
• Reviewing only random samples and missing the failure cases
• Running QA on humans and exempting AI answers
• Building a rubric so long that scoring becomes inconsistent

Questions