Plan requirement
| Subscription | Any plan |
| Also required | Quality Assurance or Workforce Engagement Management |
| Access | Admin |
Automatic scoring of the categories a machine can judge. Frees reviewer attention for the ones it cannot.
The eight categories
| Category | What it judges | Where |
| Greeting | Whether the agent greeted the customer | Text and voice |
| Closing | Whether the conversation was closed properly, including offering further help and thanking the customer | Text and voice |
| Spelling and grammar | Grammar mistakes, misspellings and style errors | Text only |
| Readability | How easily the text can be understood, from word complexity and sentence length | Text only |
| Empathy | Whether the agent was empathetic toward the customer and their concerns | Text and voice, needs LLM-based AutoQA |
| Tone | Recognises tones and scores them by positive or negative weight | Text only, needs LLM-based AutoQA |
| Solution offered | Whether the agent proposed a solution during the conversation | Text and voice, needs LLM-based AutoQA |
| Comprehension | Whether the agent understood the customer's issue | Text and voice, needs LLM-based AutoQA |
Four of them need LLM-based AutoQA turned on in the account settings, so an account without it is running half the list. Text and voice are reviewed separately.
Language decides as much as the channel does. Greeting, closing, and spelling and grammar are scored in Dutch either way. The four that need LLM-based AutoQA are then scored in any language, and in none without it. Readability is the exception in the other direction: English and Spanish only, whatever else you switch on.
All of it runs across every conversation rather than a sample, but only once a ticket reaches a solved or closed status. Open work carries no scores, which is correct and not a fault.
Where the line falls
Read that table carefully, because two entries look like more than they are. Solution offered asks whether a solution was proposed, not whether it was the right one. Comprehension asks whether the agent understood the issue, not whether they then handled it correctly.
So what remains outside is: whether the answer was correct, whether the right process was followed, and whether an exception was justified. Those need somebody who knows the subject, and they are the categories that matter most.
The right division
Let it handle the mechanical categories on everything, and have reviewers spend their time on correctness and judgement in a sample.
That combination gives you broad coverage of the basics and deep coverage of what matters.
Full coverage changes what you can see
Manual reviewing samples a few conversations per agent. Automatic scoring runs on all of them, so a pattern affecting one channel or one time of day becomes visible.
Do not put it in front of agents alone
An automatic score with no human comment is a machine telling somebody they did badly. It is the version of feedback nobody learns from, and it damages the programme's credibility.
Check it against your own judgement
Read twenty conversations it scored and see whether you agree. Where you do not, the category needs adjusting or removing.
Comments
0 comments
Article is closed for comments.