1. Describe behaviour, not personality.
“Professional attitude” can mean different things to different reviewers. “Acknowledges the customer’s request before proposing a solution” describes a behaviour that can be found in a conversation.
Start with three to five criteria. For each one, write what good performance looks like and what evidence the reviewer should find. Keep the customer’s objective visible: a polished script is not enough if the underlying request remains unresolved.
2. Set weights around the purpose of the call.
In the illustrative grid, needs discovery counts for 30%, response quality for 40% and agreeing the next step for 30%. The weights add up to 100%. With scores of 8, 7 and 8 out of 10, the result is 8 × 0.30 + 7 × 0.40 + 8 × 0.30 = 7.6.
This is a starting point for discussion, not a universal standard. A sales team may need to evaluate qualification and relevance of the proposal. A service team may focus on diagnosis, ownership and clarity of resolution.
3. Define scoring anchors and exceptions.
Describe what a low, intermediate and strong score mean for each criterion. For “next step”, a weak result might be no agreed action; an intermediate result, an action without a clear owner; a strong result, an action with an owner and timeframe.
Agree what to do when a criterion does not apply or the evidence is insufficient. Document the exception policy rather than silently penalising an agent for an action that the situation did not require.
4. Calibrate before you compare agents.
Ask two reviewers to assess the same small set of calls independently. Compare the reasons for their scores, return to the disputed passages and refine the wording of the criteria. The discussion matters as much as the numeric difference.
Repeat this exercise when you change the grid, introduce a new call type or see recurring disagreement. Avoid comparing averages across teams whose criteria, weights or call mix are materially different.
5. Turn the result into one next action.
Choose one observable behaviour for the next coaching cycle. Attach an example, agree what the agent will try and set a review date. Use a new, comparable call sample to assess whether the behaviour has changed.
AUDIO supports weighted call evaluations out of 10, explanations by criterion and manager debriefs. Its AI training scores are out of 100: they describe practice performance and should remain distinct from real-call results.
Questions managers ask.
How many criteria should a scorecard contain?
Start with a small number you can explain and review consistently. Three to five is a practical starting point for this exercise; expand only when a distinct behaviour needs its own assessment.
Should every criterion have the same weight?
Only if they carry the same importance for that workflow. Weighting makes priorities explicit, but it needs to be agreed and tested by the team.
Can we use this example in AUDIO?
Yes. Use it as a discussion starting point for your own evaluation fields and weights. Review the configuration against representative calls before deployment.