Plan requirement
| Subscription | Any plan |
| Also required | Quality Assurance or Workforce Engagement Management |
| Access | Admin |
Adjusting the automatic categories to your own standard, and switching off the ones that do not fit you.
Adjust them
- Open the automatic categories in the workspace.
- Review what each one checks.
- Turn off any that do not match how you work.
- Adjust the settings on the ones you keep.
- Check the results against twenty real conversations.
Three of them take your own words
Only Spelling and grammar, Greeting and Closing can be tuned with your own content, and what you can put in them is specific.
| Spelling and grammar | A list of words and phrases for AutoQA to ignore as mistakes. Up to 300 exemptions in total, each tagged with the language it applies to. |
| Greeting | Approved greetings that match your standards. Up to 200 in total, shared across every language you support, and they can include dynamic name variables. |
| Closing | Your own sign-offs. Up to 200 in total, again across all languages, also with dynamic name variables. |
You can change the language of several exemptions, greetings or closings in bulk, or delete them in bulk, which matters the first time you paste in a list under the wrong language.
The exemption list is the one that pays for itself fastest: product names, brand names and technical terms flagged as misspellings drag every score down until you add them.
Turn off what does not fit
A category checking for a formal closing is wrong for a team using short chat replies. Left on, it scores everybody down for doing exactly what you asked of them.
Agents notice this immediately, and it discredits the automatic scoring entirely.
Check it per channel
What reads as curt in an email is normal in chat. Where one workspace covers both, a category tuned for one will misjudge the other.
Verify before relying on it
Read twenty scored conversations and compare with your own judgement. Where you disagree consistently, adjust the category rather than accepting the score.
This check takes half an hour and it is what makes the difference between useful and resented.
Language matters, and it is per language by design
Greetings and closings are defined per language, and you have to add the expected greeting for each one you support. Because the 200 is a total rather than an allowance per language, a team working in four languages is working with fifty each, so spend them on the greetings people use. Choosing All languages means you expect that exact greeting in every language detected, which is almost never what a multilingual team wants.
Automatic scoring also works less well in some languages than others, and which languages are covered depends on whether LLM-based AutoQA is on. A multilingual team can find one language consistently scoring lower for reasons that have nothing to do with quality, so check per language before comparing anybody.
Start with a few
Two or three categories you are confident about. Adding more once those are trusted is easy; recovering credibility after a bad start is not.
Tell the team what is automatic
Which categories are scored by machine and which by a person. Agents should know what they are disputing when they disagree.
Comments
0 comments
Article is closed for comments.