Plan requirement
| Subscription | Any plan |
| Also required | Quality Assurance or Workforce Engagement Management |
| Access | Admin |
Five failure modes to expect before you put automatic scores in front of anybody. Judgement calls score badly, and judgement is what you want.
One: the deliberate exception
An agent who skips the standard greeting because the customer is furious and wants an answer has done the right thing. Automatic scoring marks it down.
Judgement calls score badly, and judgement is what you want people to exercise.
Two: brevity read as coldness
A short reply to a simple question is good service. Tone categories often read brevity as lacking empathy, which pushes agents towards padding.
Three: language
Performance differs by language, and directness that is normal in one reads as blunt in another. A multilingual team can show differences that are entirely artefacts of the tooling.
There is a harder limit underneath that, and it differs per category. Without LLM-based AutoQA, empathy, tone, comprehension and solution offered do not run at all. Greeting and closing cover eight languages including Dutch, spelling and grammar covers nine locales either way, and readability is English and Spanish whatever you do. A conversation the tool cannot judge is not scored badly; it is scored N/A, which looks like a gap in somebody's record rather than a gap in the tool.
Four: context it cannot see
A phone call, a colleague's advice, a previous ticket, an internal note. The conversation looks incomplete and the score reflects that.
Five: it cannot tell right from wrong
A well-structured, warm, perfectly spelled reply containing incorrect information scores highly. That is the failure that matters, and it is inherent rather than fixable by tuning.
What follows from this
Automatic scores are an input to a reviewer, not feedback to an agent. Where they go directly to people, the exceptions above become arguments, and the programme loses credibility over cases where the agent was right.
Check it yourself first
Twenty conversations, your judgement against its scores. Where you disagree, turn the category off rather than explaining it away.
Comments
0 comments
Article is closed for comments.