Plan requirement
| Subscription | Any plan |
| Also required | Quality Assurance or Workforce Engagement Management |
| Access | Agent |
It reads the text of the conversation and scores observable features. Four of the eight categories exist only when LLM-based AutoQA is switched on.
What it works from
The words in the conversation: what the customer wrote, what the agent replied, and the order they came in. Each enabled category is applied to that text.
The eight categories
Greeting, closing, spelling and grammar, readability, empathy, solution offered, tone and comprehension. What each one judges, and whether it works on text or on voice, is in the article on what AutoQA does.
Four of the eight therefore only exist when LLM-based AutoQA is switched on in the account settings. With it on, those four cover every language; with it off they do not run at all.
The other four are not governed by that switch. Spelling and grammar and readability work the same either way, and readability is English and Spanish only whatever you do. Greeting and closing widen to every language with the switch on, and without it cover English, German, Spanish, French, Italian, Dutch, Polish and Portuguese. So Dutch is covered for the mechanical categories in every configuration.
Where a category cannot be evaluated the score is N/A rather than a low mark, so a blank is not a failure.
It waits for the ticket to close
Nothing is scored while a ticket is still being worked. Categories are evaluated once the ticket reaches a solved or closed status, so an open queue shows no AutoQA scores at all and that is correct rather than broken.
Two more conditions before anything appears: a category has to be marked active, and it has to be assigned to at least one workspace and one active scorecard. An account where AutoQA is on and scoring nothing has usually missed the scorecard.
What it cannot see
- Whether the answer was correct.
- A phone call, an internal note, or advice from a colleague.
- Why an agent deliberately skipped a step.
- Anything in an attachment.
It scores everything, not a sample
Unlike manual reviewing, every conversation in scope is scored. That is its main advantage: patterns across hundreds of conversations become visible.
Its scores are an input
They belong in front of a reviewer deciding what to look at, not in front of an agent as feedback. A machine score with no human comment is not something anybody learns from.
Where it disagrees with you
Read the conversation. If you think the agent did the right thing and the score says otherwise, the category is wrong for your context, and turning it off is a legitimate answer.
Language affects it
Performance differs between languages. Compare within a language rather than across, or the difference you see may be the tooling.
Comments
0 comments
Article is closed for comments.