How to QA 100% of Calls in a Call Center, Not a 2% Sample

To QA 100% of calls, stop treating QA as a listening job and treat it as a review job. Connect your dialer so every recording is transcribed and scored automatically against your own scorecard, then send your QA analysts only to the calls that failed a critical item or scored low. Software does the first listen on every call. People still make the judgment calls and do the coaching.

That is the whole idea. The rest of this guide covers why sampling falls short, how to set up full coverage step by step, and what changes for your QA team once every call is scored.

Why most call centers only review a small sample of calls

Manual QA is limited by listening time. A reviewer has to play the call, often at normal speed, and then fill in the scorecard. On a busy floor the math stops working fast.

Here is a worked example with hypothetical numbers:

• 40 agents, each handling 80 calls a day, is 3,200 calls a day.
• If listening and scoring takes an average of 10 minutes per call, reviewing all of them takes about 533 hours a day.
• At 8 productive hours per analyst, that is roughly 67 full time reviewers.

No floor staffs QA that way, so teams sample. A common setup is a few calls per agent per week. If an agent makes 400 calls in a week and QA reviews 4 of them, coverage is 1 percent. Change the inputs to your own numbers and you will usually land in the low single digits.

What sampling misses

Sampling is fine for estimating an average. It is weak at catching rare, expensive events, and those are the calls that hurt you.

Rare compliance misses go unheard

Say an agent skips the recording disclosure on 8 of their 400 weekly calls. If QA picks 4 calls at random, the chance that none of the 4 is a missed disclosure is about 92 percent. In most weeks, QA never hears the problem, and the agent never gets told.

Scores are based on too few calls

Four calls is not enough to judge an agent fairly. One rough call can drag a weekly score down, and one easy call can hide a habit. Agents know this, which is why sampled scores lead to disputes.

Feedback arrives late

By the time a sampled call is reviewed, the agent has made hundreds more calls the same way. Coaching a habit a week after it formed is slower than coaching it the same day.

Picking calls introduces bias

Reviewers tend to pick calls of a convenient length, from agents already on their radar, or from the start of a shift. That is human and understandable, but it means the sample is not really random.

What the research says about the calls a sample misses

Small samples are the norm, not the exception. ISG Research wrote in 2022 that contact centers typically score as little as 1% to 3% of their interactions. The calls left unheard are the ones that cost money:

• Bad experiences cut spending. In its 2024 analysis of a survey of more than 28,000 consumers in 26 countries, Qualtrics XM Institute found consumers cut spending after 38% of bad experiences and stopped spending with the company entirely after another 13%. It estimated $3.7 trillion of 2024 sales were at risk from bad experiences.
• One bad interaction can be enough. In Zendesk's CX Trends 2023 research, more than half of customers said they would switch to a competitor after one unsatisfactory interaction, and 73% said they would after multiple bad experiences.
• Unresolved calls come back. SQM Group reports that customer satisfaction drops 15% on average every time a customer has to contact the company again, and that repeat calls make up 23% of the average contact center's operating budget.

A 1% to 3% sample can estimate an average. It cannot tell you which of today's calls were the bad ones, and those are the calls you can still do something about. For more numbers like these, see our call center customer experience statistics.

What "100% call QA" actually means

100% call QA means every recorded call is transcribed, scored against your scorecard and checked for risk, with the evidence behind each mark stored on the call. It does not mean a person listens to every call. It means no call goes unchecked, and human attention goes where the scores say it is needed.

In practice, each call gets:

• A score for each criterion on your scorecard: pass, partial, fail or not applicable.
• The quote from the transcript that supports each mark.
• Measured delivery metrics such as dead air, talk ratio and interruptions.
• A flag when a critical item fails, so it lands in a review queue.

How to QA every call, step by step

1. Fix your scorecard first. Automated scoring is only as clear as your criteria. Rewrite each criterion so a pass is described in one sentence that two reviewers would agree on. If you need a starting point, use our free QA scorecard template.
2. Mark your critical items. Decide which misses make the whole call fail, such as a required disclosure, identity verification or a misleading claim. These become auto-fails and drive your review queue.
3. Write a scorecard per campaign. A collections line, a Medicare line and a support line have different rules. One generic scorecard for all of them produces vague scores.
4. Connect the dialer. Recordings should flow in automatically, not through manual uploads. Most cloud dialers expose recordings through a webhook or an API.
5. Calibrate before you trust the numbers. Have two or three analysts score the same set of calls by hand and compare their marks with the automated scores. Where they disagree, the criterion wording is usually the cause. Tighten it and rerun.
6. Build the review queue. Decide what a human must look at: every auto-fail, every call under a set score, and any criterion your compliance team wants confirmed by a person.
7. Change the QA job description. Analysts stop picking random calls and start confirming flagged calls, running calibration and coaching agents with the quotes in hand.
8. Report on trends, not just calls. With every call scored you can see pass rates per criterion, per agent and per campaign, week over week.

Manual sampling vs 100% automated QA

For a longer side by side, see Sillana vs manual QA.

What changes for the QA team

Analysts become reviewers and coaches

The job moves from "find a call and listen" to "confirm this flag and decide what to do." A flagged call with the quote attached takes far less time to judge than a random call played from the start. That frees analyst hours for calibration and coaching, which is the part of QA that actually changes behavior.

Agents get feedback on every call

When every call has a score and quotes, agents can see their own patterns: which criterion they miss, on which kind of call, and what they said. Coaching gets specific. "On Tuesday's call you said X, try Y" lands better than "your discovery needs work."

BPOs can show clients evidence

If you run campaigns for clients, full coverage changes the conversation. Instead of reporting a sampled average, you can show pass rates for the client's own required items across every call, with examples. That matters most on regulated lines such as debt collection and insurance and Medicare.

Common mistakes when moving to 100% QA

• Vague criteria. "Agent was professional" gets inconsistent scores from people and from software. Describe what a pass sounds like.
• Too many criteria. A 40 line scorecard hides the items that matter. Keep it to what moves revenue, risk or customer outcome.
• No "not applicable" rule. If a criterion cannot apply to a call, it should not count as a miss. Otherwise short calls look worse than they are.
• Treating every automated mark as final. For compliance findings that could lead to discipline or a client report, have a person confirm the call first.
• Ignoring audio quality. One mono channel with heavy noise makes any transcript worse. If your dialer can record agent and customer on separate channels, use it.

How Sillana helps

Sillana connects to dialers including Aircall, Dialpad, RingCentral, Convoso, Salesfinity, ReadyMode and CallTools, and accepts calls from other systems through its API. It transcribes each recorded call and scores it against the scorecard you write, with the quote behind every mark. Critical criteria act as auto-fails, and speech metrics like dead air, talk ratio, pace and interruptions are measured from the audio. You can run a different scorecard per campaign. Pricing is based on audio minutes scored with unlimited seats; see pricing.

It does not replace your QA team. It does the first listen on every call so your team can spend its time on the calls that need a person.

Key takeaways

• Sampling a few calls per agent per week usually covers only a small share of calls and misses most rare, high risk events.
• 100% QA means every call is scored and checked, not that a person hears every call.
• Clear criteria and defined auto-fails come before any tooling.
• Calibrate automated scores against your analysts before relying on them.
• Move analyst time from listening to reviewing flagged calls and coaching.

Questions people ask