Short answer: Average QA score is the least useful number on most QA reports. Leaders should track coverage (how much was scored), compliance miss rate, behavior-level miss rates, evaluator agreement, time from interaction to feedback, coaching follow-through, behavior change after coaching and dispute overturn rate. Together they tell you whether QA is accurate, fast and actually changing how agents work.
Most QA reports I see lead with one number: the average QA score. Maybe it's 87%. Maybe it went up two points.
So what?
An 87% average can hide a compliance problem on a handful of calls, a form question nobody scores the same way, and coaching that never happened. The number looks fine right up until a client or a regulator finds what it was hiding.
Here are the QA metrics I'd want in front of me as a contact center leader, and the questions each one answers.
The Three Questions QA Metrics Should Answer
Every number on a QA report should help answer one of these:
- Are we scoring enough, and scoring it right? (Is the data trustworthy?)
- Where are agents falling short? (What do we fix?)
- Is anything getting better? (Is QA worth what we spend on it?)
If a metric doesn't help with one of those, cut it from the leadership report.
1. Coverage
Formula: interactions scored ÷ total interactions handled, by channel and by agent.
This is the first number I look at, because it tells me how much to trust everything else.
If an agent took 1,000 calls and 4 were scored, their QA score is mostly noise. The sample is too small to say much of anything. The NIST engineering statistics handbook shows how sample size drives the margin of error on a proportion, and our post on how many QA evaluations you need per agent walks through it for QA.
Report coverage next to every score. A score with no sample size is a guess dressed up as a fact. Not sure how much coverage you need? Start by building a sampling plan around your goals.
2. Compliance Miss Rate
Formula: interactions with at least one compliance or auto-fail miss ÷ interactions scored on those items.
Keep this separate from the overall score. Compliance misses are rare, expensive and easy to bury in an average.
Report it two ways:
- Rate, so you can see the trend
- The list, every miss with the agent, date and item, so someone acts on each one
If your compliance items are only scored on a small sample, say so on the report. Rare misses don't show up in small samples.
3. Behavior-Level Miss Rates
Formula: for each form question, misses ÷ times the question applied.
This is the most useful metric on the page, and most reports skip it.
The average score tells you something is off. The question-level view tells you what. "Verification missed on 9% of calls" is something a supervisor can coach on Monday. "Average score 84%" isn't.
Look at it three ways:
- By agent, for coaching
- By team, for supervisor focus
- By call type, because a miss that shows up on one call type everywhere is usually a script, training or process problem, not a people problem
4. Evaluator Agreement
Formula: for a set of interactions scored by two or more evaluators (or an evaluator and AI), how often they gave the same answer, question by question.
If your evaluators don't agree with each other, every other number on the report is shaky. Agents notice before leaders do.
Percent agreement is the simple version. For questions that are almost always "yes," Cohen's kappa adjusts for agreement you'd get by chance. This NIH-hosted article explains both in plain terms. We walk through running calibration between people and AI in a separate post.
You don't need this on the weekly report. Put it on the monthly one, by question, and watch for any question that's sliding.
5. Time to Feedback
Formula: average time from the interaction to the coaching conversation about it.
This one gets ignored, and it matters a lot.
If a call gets scored eight days after it happened and coached four days after that, the agent has repeated the same habit on a lot of calls in between. Fast feedback is cheap improvement.
Break it into two parts so you know where the delay is:
- Interaction to score
- Score to coaching
6. Coaching Follow-Through
Formula: coaching sessions completed ÷ coaching sessions planned, by supervisor.
QA without coaching is just grading. This number shows whether scores are turning into conversations.
Low follow-through usually isn't laziness. Supervisors get pulled onto the floor when volume spikes. If that happens every week, rerun the how many agents do I need math with an Erlang C calculator and build coaching time into shrinkage. Track it anyway. It tells you when QA has quietly stopped doing its job.
7. Behavior Change After Coaching
Formula: miss rate on the coached behavior in the period after coaching, compared with the period before.
This is the number that proves QA is worth the money.
Don't measure coaching by how many sessions happened. Measure it by whether the behavior you coached changed on the agent's next scored interactions. If it didn't, change the approach, or look at the script, the training or the form question itself. Here's the coaching routine I'd use.
8. Dispute Rate and Overturn Rate
Formula: disputed evaluations ÷ evaluations, and disputes upheld for the agent ÷ disputes.
A few disputes are healthy. It means agents read their scores and care about them.
The overturn rate is the interesting part. When agents win disputes on the same question again and again, the question is unclear or the evaluators aren't calibrated. Fix the form, then re-check agreement.
How QA Connects to Customer and Business Outcomes
Leaders eventually ask: "Does a better QA score actually mean happier customers?"
Fair question. Here's how I'd check it without overselling it:
- Pick one behavior you think matters, like resolving the issue without a transfer.
- Compare that behavior's pass rate against an outcome you already track, like repeat contacts or survey results, by agent or by team.
- Look for a relationship over a few months, not a few days.
That gives you a reason to weight that question more heavily, or to stop scoring something that doesn't matter. Treat it as a signal, not proof. Lots of things move customer outcomes.
A Simple Leadership QA Report
Here's a layout I'd use. Every value is yours to fill in. This is a structure, not a benchmark.
| Metric | This period | Last period | Owner |
|---|---|---|---|
| Coverage by channel | QA lead | ||
| Compliance miss rate (plus the list) | Compliance | ||
| Top 3 missed behaviors | Operations | ||
| Evaluator agreement (monthly) | QA lead | ||
| Time to feedback | Supervisors | ||
| Coaching follow-through | Supervisors | ||
| Behavior change on coached items | Operations | ||
| Dispute overturn rate | QA lead |
One page. Owners on every line. If a number moves, someone already knows it's theirs.
Metrics I'd Drop From the Leadership Report
- Average QA score on its own. Keep it in the detail, but never without coverage and the behavior breakdown.
- Number of evaluations completed. It measures effort, not quality. Coverage is the better version.
- Agent rankings from tiny samples. They swing week to week and create disputes, not improvement.
Where OttoQA Fits
Most of these metrics fall apart for the same reason: not enough interactions get scored, and scoring is slow.
OttoQA scores calls against your own QA forms, so coverage goes up and your behavior-level numbers come from the questions your team already uses. No seat licenses, no IT, and setup takes about 2 weeks. There's a $1,000 one-time setup fee, and you can start with a 200 to 500 call proof of concept: we score 200 to 500 of your calls against your form.
Bottom line... stop reporting one average. Report what was scored, what was missed, and whether coaching changed anything.
FAQ
What are the most important contact center QA metrics?
Coverage, compliance miss rate, behavior-level miss rates, evaluator agreement, time to feedback, coaching follow-through, behavior change after coaching, and dispute overturn rate. Together they show whether your QA data is trustworthy and whether it's improving performance.
Is average QA score a good KPI?
Not on its own. An average can hide rare compliance misses and unclear form questions. Report it with coverage and question-level miss rates so leaders can see what's behind it.
How do you measure whether QA coaching is working?
Compare the miss rate on the coached behavior before and after the coaching session, using the agent's next scored interactions. If the behavior doesn't change, adjust the approach or look at the script, training or form.
What is QA coverage in a call center?
It's the share of interactions that were actually scored, usually reported by channel and by agent. Low coverage means agent-level scores are based on very few interactions and can swing a lot.
How often should QA metrics be reported to leadership?
A short weekly view for compliance misses and top missed behaviors, and a monthly report that adds evaluator agreement, coaching follow-through and behavior change. Keep it to one page with an owner on every line.
About the author
Tom Laird is the founder of OttoQA and CEO of Expivia, with about 30 years in contact centers. He hosts the Advice from a Call Center Geek podcast. Connect with Tom on LinkedIn.