Automated quality assurance is software that scores every customer call or chat against a company's quality checklist, instead of a supervisor scoring a small sample by hand. In a contact centre, quality assurance, or QA, means checking whether agents handle customers correctly. It is one part of conversation intelligence.
Every contact centre has a scorecard. A scorecard is a list of questions that a reviewer answers for each call. Did the agent greet the customer and confirm their identity? Did the agent read out the required disclosure? Did the agent solve the problem? Was the agent polite? Each question carries points, and the total is the call's score.
Earlier, a supervisor listened to calls and filled in the scorecard by hand. One review took 15 to 20 minutes. So each agent got two or three calls reviewed a month, out of several hundred. An agent could be scored on a lucky day or an unlucky one. And a serious mistake on an unreviewed call was never found.
Here is how the automated version works. Each call is first turned into text. A language model then reads the transcript and answers each scorecard question. It also notes the line in the call that supports each answer. The QA team checks a sample of the model's scores and corrects the questions it gets wrong, so that the model improves.
Take an outsourcing company in Mohali that runs customer care for a telecom operator. It has 300 agents, who handle about 45,000 calls a month. Earlier, its QA team of six reviewers scored 600 of these calls a month, about 1 in 75. Now every one of the 45,000 calls is scored on a 10-question scorecard. Say one agent's average across 160 calls this month is 71 out of 100, against the team's 84. The software shows that she skips the identity check on calls that come in after 10 p.m. Her team leader plays her two of those calls in a coaching session. A reviewer could easily have missed this pattern with two calls a month.
The scores also matter outside the contact centre. Banks, lenders and insurers must follow regulators' rules on some kinds of calls, such as insurance sales calls and loan recovery calls. They must be able to show that their staff followed those rules. The same checks feed employee compliance monitoring, and the recordings are kept by communications archiving systems. Companies now use the same scorecards to test AI voice agents before launch, through voice agent testing.