ACXPA · Coaching & Performance Hub
The QA Sample Size Calculator that tells you the truth
Most contact centres pick their QA sample size out of thin air. Work out how many calls you actually need to review to trust your quality scores — or check how much your current sample is really telling you.
Scoring individual agents needs a much bigger sample than checking how the team's doing overall.
The number of agents whose calls you review.
Every call (or contact) you could review — say, over a month. Drag the slider to model a different volume.
How far off the true score you can live with. ±5% is a common target — drag the slider to see the effect.
95% is the industry standard — the usual balance of being sure and being precise. Right for almost everyone.
Your fail rate is the share of calls that don't meet your QA standard. The more lopsided it is — mostly pass, or mostly fail — the fewer calls you need to review. A 50/50 split needs the most, so 50% is the safe setting when you're not sure.
Fine-tuning for a known fail rate is part of the Advanced QA Calculator.
Calls to review
Verdict
Coverage
of all calls
Per agent
calls each
Likely fails
at 50% fail rate
How much you're reviewing
Worth knowing
Your call volume barely matters
What sets your accuracy is how many calls you review, not how many your centre handles. It's the same reason a national opinion poll can read a whole country from a couple of thousand people: once your pool is big enough, making it bigger barely changes the answer. Drag the call-volume slider above from a few hundred up to tens of thousands and watch the required number sit almost still.
Bigger won't fix biased
That same poll falls apart if it only reaches one kind of person — one suburb, one age group, landlines only. QA is no different: review only the easy calls, a single team, or one slot of the day and the result is skewed however many you check. A lopsided sample doesn't straighten out by getting bigger — it just makes a wrong read look more convincing. The maths here assumes every call has an equal chance of being picked.
Method & assumptions
QA sample size uses the standard proportion formula with a finite population correction: n = N·z²·p(1−p) / [ e²(N−1) + z²·p(1−p) ]. Accuracy reverses it: e = z·√(p(1−p)/n)·√((N−n)/(N−1)). Assumes calls are picked at random (not cherry-picked) and scored pass/fail. Percentage scorecards use a different (variability) method — that's an ACXPA Member feature.
Included with your membership
You've got the full version
, the Advanced QA Calculator and the QA Sampling Benchmark are part of 's ACXPA Business Membership — fail-rate fine-tuning, the cost & time overlay, PDF export and saved setups.
The Advanced QA Calculator is exclusive to Business Memberships
, the advanced toolkit is reserved for ACXPA Business Members and isn't part of 's Vendor Membership. Talk to us about access.
How to use the QA Sample Size Calculator
The calculator runs in two directions — pick the mode that matches the question you're actually asking.
How many should I review?
Enter your call volume, how accurate you need the score to be (±%) and your confidence level. It returns the number of calls to QA — and what that works out to per agent and per week.
How good is my current sample?
Already review a set number? Enter how many and it tells you how tight your score really is — whether it's solid enough to act on, or just a rough signal.
Choose The team overall for a read on how the whole team is performing, or Individual agents to check whether your sample is big enough to compare people fairly. It rarely is — and the tool tells you straight when it isn't.
A worked example
Say you run a 25-agent team handling about 8,000 calls a month, and you want a quality score you can trust to within ±5% at 95% confidence.
Enter the basics
8,000 calls in the period, ±5% accuracy, 95% confidence.
Read the number
You get roughly 370 calls to review — about 15 per agent across the month, or three or four a week each.
Sanity-check the ambition
Switch to Individual agents and you'll see 15 calls each is nowhere near enough to rank people fairly — fine for coaching and a team read, not for a league table.
The takeaway
About 370 well-chosen calls reads the whole team's quality for the month. Reviewing thousands more would barely add anything — but spreading those same calls across 25 people to score them individually would tell you almost nothing reliable.
Calculator FAQ
How often should I run this?
Each QA cycle — usually monthly. Your team size and call volume drift over time, so the right sample drifts with them. It takes seconds to re-check.
Should I calculate for the team or per agent?
Start with the team overall — that's what most QA programmes actually need. Only switch to Individual agents if you intend to score people individually, and use the per-agent check to see whether that's even defensible on your review budget. Usually it isn't.
What if my call volume swings month to month?
Use a typical month. Because total volume barely moves the required number once it's well above your sample, ordinary swings make almost no difference to how many calls you need to review.
Does this work for chat, email and other channels?
Yes. “Calls” is just the label — the maths is the same for any interaction you score pass/fail, whether it's a call, chat, email or social message. Run each channel separately if you hold them to different standards.
Why does it default the fail rate to 50%?
50% is the most cautious assumption: it produces the largest, safest sample whatever your real pass/fail split turns out to be. Setting your actual rate can trim the sample when your split is lopsided — that's an Advanced QA Calculator feature.
Can I use the result to score individual agents?
Usually not, and the tool will warn you. A team-sized sample split across agents leaves only a handful of calls each — far too few to separate a genuinely better agent from a worse one. Use it to coach and to read the team's quality, not to rank people.
New to the concept?
For the plain-English background — what it means, what drives it and the traps to avoid — read the QA Sample Size explainer in the ACXPA glossary.
