Short answer: Call center QA scoring means grading a recorded call against a written form. Each question gets points, some questions carry more weight, and a few are auto-fails. Add it up and you get a score. The score is only fair if the form is clear, the calls you pick are representative, and the people (or the AI) scoring them agree on what each question means.
I've run contact centers for a long time. I've seen QA scores help agents get better. I've also seen them turn into a number nobody trusts. The difference is almost never the agents. It's how the scoring is set up.
Here's how QA scoring works, and how to keep it fair.
What is QA scoring in a call center?
A QA score is a grade for one call. A QA analyst, a supervisor, or software listens to the call and answers the questions on your QA form. Did the agent verify the account? Did they solve the problem? Did they set the right expectations at the close?
Each answer turns into points. The points roll up into a percentage. That percentage gets tracked by agent, by team and over time.
That's it. The math is simple. The hard part is making sure the number means something.
How a QA score is calculated
Most forms use some mix of these:
- Yes / No / N/A questions. The agent did it or didn't. N/A drops the question from the math, so an agent isn't penalized for something that didn't apply to the call.
- Rated questions. A 1-5 or 1-3 scale for things like tone or ownership. These need clear definitions for each level, or two scorers will land in two different places.
- Weights. Not every question matters equally. Resolving the issue should count for more than using the customer's name.
- Auto-fails. A few items, usually compliance or security, zero out the call if they're missed. Keep this list short. If everything is an auto-fail, nothing is.
A simple example. Say your form has 10 questions worth 100 points. The agent earns 85 points on the questions that applied. One question was N/A, worth 10 points. The score is 85 out of 90, or about 94%. If they missed an auto-fail item, the score is 0, no matter what else happened.
Write the math down. Agents should be able to check their own score.
Why QA scores feel unfair to agents
When agents push back on QA, they usually have a point. The common reasons:
- Vague questions. "Showed empathy" means one thing to one supervisor and something else to another. Same call, two scores.
- Too few calls. If an agent gets three or four calls scored a month, one bad call sinks them. That isn't a read on their work. It's a coin flip.
- Cherry-picked calls. If the calls scored are the escalations and the long ones, scores skew low and agents know it.
- Scoring the script, not the outcome. Docking someone for not saying the exact greeting word for word, on a call where the customer left happy, teaches agents to read instead of help.
- No calibration. If the people scoring never compare notes, drift sets in. Scores move because the scorer changed, not because the agent did.
Every one of these is a process problem. You can fix all of them.
How to make QA scoring fair
1. Write questions two people would score the same way
Read each question and ask: would two of my best supervisors give the same answer on the same call? If not, rewrite it. Define what a pass and a fail look like. Add a short example of each.
"Showed empathy" becomes "Acknowledged the customer's problem in their own words before moving to the fix." That's something you can hear on a recording.
2. Weight what matters
Put the most points on the things that drive results: solving the problem, accuracy, compliance. Put fewer on polish. If a perfect greeting is worth as much as fixing the customer's issue, your form is telling agents the wrong thing.
3. Keep auto-fails for real risk
Auto-fails should be things that put the customer or the company at risk. Missed verification. A required disclosure. Not a forgotten closing line.
4. Score enough calls to be fair
One of the biggest sources of unfair scores is sample size. A handful of calls can't tell you how an agent really performs. Score more calls, and pick them at random across call types, times and days. I wrote more on this in how many calls QA should score.
5. Calibrate on a schedule
Get your scorers together, have them score the same calls on their own, then compare. Where they disagree, fix the question wording. Do it regularly, and again any time the form, the script or a policy changes. Here's a step-by-step guide to running a QA calibration.
6. Show agents the why, not just the number
A score with no notes is a grade, not feedback. Every missed question should come with the moment on the call and what good would have sounded like. That's what turns a score into coaching.
7. Let agents ask questions about a score
If an agent thinks a call was scored wrong, give them a simple way to raise it. Sometimes they're right. When they are, it usually points to a question that needs rewriting.
Manual QA scoring vs AI QA scoring
Manual scoring works. It's how most centers have done it for decades. The limit is time. A person can only listen to so many calls in a day, so the sample stays small, and small samples are where unfair scores come from.
AI QA scoring takes your same form and applies it to far more calls than a team can listen to. That fixes the sample-size problem. It also helps with consistency, because the AI reads each question the same way on Monday morning as it does on Friday afternoon.
But AI scoring has to be calibrated just like people do. If you hand an AI a vague question, you'll get vague results, just faster. The fairness rules above still apply. Clear questions, sensible weights, a short auto-fail list and regular calibration.
The people still matter. QA analysts spend less time listening and more time on patterns, calibration and helping supervisors coach. Supervisors still own the coaching conversation.
Where Otto Fits
OttoQA scores calls against your own QA form. You keep your questions, your weights and your auto-fails. We don't hand you a generic scorecard.
- Calibrated to your team. We start with 10-20 calls your team already scored, compare Otto's scores to yours question by question, and tighten the wording until they line up. Otto scores at 96% accuracy against your team's scoring.
- No IT needed. Otto connects to any CCaaS or cloud CRM via API, including Zoom Contact Center.
- Any language.
- Setup takes about 2 weeks.
- Your data stays protected. Hosting is US-only, your calls are never used for training, and OttoQA is PCI compliant and signs HIPAA BAAs.
- Pricing. The one-time $1,000 covers setup and a 200 to 500 call proof of concept. After that, you choose a contract or month-to-month. Details are on the pricing page.
If you want to see your own form scored on real calls, Book a demo.
Ready to see it on your calls? Book a demo and bring your QA form.
Frequently asked questions
What is QA scoring in a call center?
QA scoring is grading a recorded call against a written QA form. Each question earns points, some are weighted more heavily, and a few are auto-fails. The total becomes a score you track by agent and team over time.
How do you calculate a QA score?
Add the points the agent earned on the questions that applied, then divide by the points possible on those questions. N/A questions drop out of the math. If an auto-fail item is missed, the score is zero.
How do you make QA scoring fair?
Write questions two scorers would answer the same way, weight what matters most, keep auto-fails for real risk, score enough randomly picked calls, calibrate on a schedule, and give agents notes and a way to question a score.
Is AI QA scoring accurate?
It can be, if it's calibrated to your form. OttoQA calibrates on 10-20 calls your team already scored and reaches 96% accuracy against your team's scoring.
Can AI QA software use my existing QA form?
Yes. OttoQA scores calls against your own QA form, with your questions, weights and auto-fails. Setup takes about 2 weeks, and no IT is needed.
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.