Picture Monday's team meeting on a renewals floor. Two supervisors have listened to the same call. One passed it. The other failed it because the telecaller never mentioned the grace period. The telecaller wants to know which of them is right, and nobody can say, because the standard lives in two heads.
Call QA can read every call you ask it to, but it can only grade against what you wrote down. So the scorecard is the real work. Most of the arguments a floor has about grades are arguments about a badly worded dimension.
Start with five things you would fail someone for
You can start from a ready-made scorecard or write your own dimensions and a pass bar. Either way, resist the long list. Sit down with your two strictest supervisors and ask what would make them mark a call as bad, then write those down as plain questions.
- Did the caller confirm they were speaking to the right person before discussing the account?
- Was the amount and date stated the way your record shows them?
- Did the caller offer the next step and say when it would happen?
- When the customer objected, did the caller answer the objection or talk over it?
Each one should be something two people would mark the same way after hearing the same call. “Showed empathy” fails that test. “Let the customer finish before replying” passes it.
Use the Policy pack for what is never negotiable
A Policy pack sits beside the scorecard and holds required phrases, banned phrases, required actions and rules. This is where a promise you never want made belongs, and the line your legal team wants said. Phrases are matched in what the graded speaker said, and never in what the customer said, so a customer who shouts a banned word at your telecaller does not fail the call.
Keep this list short as well. A long banned list makes the Risky words flag on Live Calls fire so often that supervisors stop looking at it.
Decide how much runs by itself
There are four settings in Call Settings: Automatic, Risky calls only, On demand and Off. A new workspace starts on On demand, so nothing is graded or charged until someone asks. Automatic grades every answered call and charges for each. Risky calls only grades calls where a banned phrase was said or a required phrase was missed, plus a share of the rest that you set.
While you are still arguing about the wording, stay on On demand and grade a batch you have also listened to yourself. Compare the two. Once the AI's marks and your own mostly agree, widen it. The rate is on the pricing page, and grading calls together shows the total before anything runs.
Keep the AI and your people apart
A handed-over call has two speakers doing two jobs. The AI's part and the person's part are graded on their own, and staff calls can have a scorecard of their own. Use that. A telecaller who takes an angry customer after a hand-over should not be marked on how the AI opened the call. One hand-over can mean two charges, one for each part.
Let people answer back
A grade is the AI's reading of a transcript, and it can be wrong. A misheard word can fail someone. So the product gives people a route: the person opens the call and presses Dispute on the QA card, once. A reviewer answers, the person sees the answer in their notifications, and nobody reviews their own grade.
Tell your floor this exists on the first day. A floor that has been told “the system said so” will quietly game it. A floor that can challenge a grade tends to read the feedback.
Check your reviewers too
Review rules choose which grades a person looks at. They never grade a call themselves. Calibration has several reviewers score the same calls and shows where they disagree. Run it monthly, and when two supervisors split on a dimension, reword the dimension before you talk to either of them.
What can catch you out
- Using a template or resetting to the default replaces that scorecard's dimensions and Policy pack in one step, with no undo. Copy your wording somewhere first. The ready-made packs for collections, healthcare and admissions only add items.
- Changes apply to calls graded from now on. Grading a part again replaces the old grade and clears its review.
- The numbers describe the graded calls only. The page says how many of how many, so read that line before the average.
- Test calls are never graded, so a scorecard needs real calls to be tried on.
If you want the longer route from a graded call to a better script, read make your calls better. Call QA depends on your plan, which the pricing page shows.