Skip to content
Ramp Remote

Call Center Quality Monitoring: A Practical Guide

How to run call center quality monitoring that improves calls: scorecards, sampling, calibration, KPIs, AI scoring, recording law, and who should do the reviewing.

By Ramp RemoteUpdated

Call Center Quality Monitoring: A Practical Guide

Call center quality monitoring is the routine of recording customer interactions, reviewing them against a written standard, scoring them, and turning the scores into coaching and process fixes. It works when it is consistent, calibrated and tied to coaching. It fails when it is a pile of scorecards nobody acts on. This guide covers how to set it up, how many calls to review, which metrics matter, where AI helps, and what to check before you record a single call.

Ramp Remote
Build capacity for quality monitoring
Ramp Remote recruits offshore professionals for administrative and operational roles, helping businesses staff quality monitoring while managing HR and payroll centrally.

TL;DR

  • Monitoring is the listening and scoring. Quality assurance (QA) is the wider program that uses those scores to coach agents and fix processes.
  • A working program needs one owner, a short scorecard, a sampling plan, monthly calibration and a coaching loop. Skip any one and the data stops being trusted.
  • AI can now score every interaction, but human reviewers still matter for escalations, low CSAT and repeat contacts.
  • Check recording-consent rules for every state your callers are in before you start.
  • Small teams can staff the reviewer and the analyst roles offshore at a predictable monthly cost. We cover that near the end.

Monitoring, QA and quality management: what each one means

The three terms get used as if they were one thing. They are not.

Call monitoring is the act of reviewing calls. RingCentral’s QA guide describes it as QA specialists or supervisors reviewing a sample of calls against pre-defined criteria: did the agent stay professional, follow the script where one applies, and resolve the issue.

Call center QA is the process around it. It reviews interactions against a written standard, scores them, and turns the results into coaching and process fixes. The review is only the first step.

Quality management is the umbrella: monitoring, scoring, coaching, calibration, reporting and the software that supports them.

If you are building from scratch, think of it as a loop. You define what good looks like, sample interactions, score them, coach on the patterns, and measure whether the scores and customer outcomes move. Then you adjust the standard.

Call quality monitoring loop from standards to adjustment

Build the scorecard before anything else

Most weak programs have a weak scorecard. Either it has 40 line items that nobody can score consistently, or it rewards script-reading over solving the customer’s problem.

A scorecard that works in practice:

  1. Keep it short. Aim for 8 to 15 items you can defend. If a line item cannot be scored the same way by two reviewers, delete or rewrite it.
  2. Group by what the customer experiences. Opening and identification, understanding the issue, accuracy of the answer, handling of the customer’s emotion, resolution, and closing.
  3. Separate compliance from soft skills. A missed required disclosure should be a pass/fail item or an automatic fail, not one line among twenty that can be outweighed by a friendly tone.
  4. Weight what drives outcomes. Accuracy and resolution should carry more points than greeting wording.
  5. Write behavioral anchors. For each item, describe what a full-credit, partial-credit and zero-credit call sounds like, ideally with a real example.

The ICMI white paper on call center KPIs is a useful frame for what the scorecard should sit alongside. It groups metrics into service (such as average speed of answer and service level), efficiency (such as occupancy and average handle time) and quality (accuracy, adherence to protocol and first call resolution). It also cautions against relying only on efficiency measures like AHT, which can push agents toward rushed, poor-quality service.

How many calls should you review?

Historically, QA teams sampled a tiny slice of volume. RingCentral notes that traditional QA relied on manual sampling of 1 to 3 percent of calls, and that current practice leans toward hybrid models where AI scores all interactions and humans focus on complex or high-risk ones.

For a team without AI scoring, a reasonable starting plan looks like this:

Agent tenure Suggested human reviews per agent per month Why
First 30 to 90 days More than your baseline, with at least some reviewed within days New agents form habits quickly, so early feedback pays off
Established agents A smaller, steady number Enough to spot drift without making agents feel watched
Agents on a performance plan Increased temporarily You need evidence for coaching conversations

We deliberately do not put a universal number in that table, because the right figure depends on call length, volume, regulatory exposure and reviewer capacity. Pick a number you can sustain every month. A smaller program that never lapses beats an ambitious one that stops in week six.

Do not rely on pure random sampling

Random checks give you an unbiased picture of the average call, but they waste listening time on routine calls. Practitioners recommend stratifying the random sample and deliberately adding high-risk interactions: escalations, low CSAT results and repeat contacts. A good split is part random, part targeted. The random part tells you how you are doing. The targeted part tells you what to fix.

Calibration: the step most teams skip

Calibration means several evaluators score the same interactions independently, then compare. Without it, one reviewer’s “4 out of 5” is another’s “2”, and agents quickly notice.

NiCE’s best-practice guidance describes monthly calibration sessions as an overlooked success factor. It suggests that if reviewers cannot land within a tight margin of each other, for example within 5 points on a 100-point scale, the evaluation criteria are probably too ambiguous and need refining.

A simple calibration routine:

  1. Pick three to five recorded interactions, including at least one messy one.
  2. Have each reviewer score them alone, before any discussion.
  3. Compare scores line by line. Mark every item where scores diverge.
  4. Discuss the divergent items and rewrite the scorecard wording or anchor examples until the answer is clear.
  5. Record the decisions so new reviewers inherit them.

If you have only one reviewer, calibrate against a supervisor or an outside reader every month. A single scorer with no check drifts without knowing it.

The metrics that tell you whether monitoring is working

QA scores alone are a lagging, internal view. Pair them with outcome measures so you can see whether a higher score means a better customer experience.

  • QA score trend by agent, team and scorecard section.
  • First call resolution (FCR). The ICMI paper lists it among quality KPIs and recommends aligning internal metrics with what customers define as satisfying, rather than leaning on speed alone.
  • CSAT on the same interactions you scored. If QA scores rise and CSAT does not, the scorecard is measuring the wrong things.
  • Average handle time and occupancy as guardrails, not goals.
  • Service level, commonly expressed as a share of contacts answered within a set number of seconds. The ICMI paper gives 90 percent in 20 seconds as a common example.
  • Calibration variance between reviewers.
  • Coaching completion rate. If feedback sessions do not happen, scores are just paperwork.

Using the 80/20 idea

Operations leaders often use the 80/20 rule here: a large share of dissatisfaction or errors tends to come from a small number of call drivers. RingCentral’s QA material frames it as a way to focus QA effort on high-impact drivers rather than spreading reviews thin. In practice, tag every reviewed call with a reason code. After a month, sort by frequency and by score impact. The top two or three reasons usually point to a fixable cause: a confusing help article, an outdated procedure or a broken handoff.

Where AI fits

Automated scoring is not new as a research idea. A paper from IBM researchers described a system combining speech recognition, pattern matching and classification to rank recorded calls by quality. What has changed is scale and cost. Vendor guidance such as Verint’s QA guide now describes modern programs as using AI to evaluate interactions at scale across channels, rather than depending mainly on manual sampling.

What AI does well:

  • Scores every call on objective items: required disclosures, greeting, verification steps, prohibited phrases.
  • Surfaces outliers: long silences, repeated transfers, negative sentiment.
  • Gives you trend data that a 2 percent sample cannot.

What still needs a person:

  • Judging empathy, tone and judgment calls in unusual situations.
  • Deciding whether a flagged call is a real problem.
  • Coaching. A score does not change behavior; a conversation does.
  • Checking the AI itself. Review a sample of its scores against human scores in your calibration sessions.

RingCentral’s guide also describes the QA specialist’s role shifting from auditor to performance coach. With rote flagging automated, the valuable skill is finding root causes, such as an outdated SOP or knowledge base, rather than logging individual agent errors. Plan job descriptions accordingly.

A QA process you can start this month

Here is a sequence that works for a team of 10 to 50 agents.

  1. Name one owner. Not a committee. One person accountable for the scorecard, the calendar and the reports.
  2. Write the standard. Draft the scorecard with two or three of your best agents and a supervisor. Test it on ten old calls.
  3. Choose your sampling plan. Part random per agent, part targeted at escalations, low CSAT and repeat callers.
  4. Run a pilot for two to four weeks. Score, then calibrate. Expect to rewrite at least a third of the line items.
  5. Share the scorecard with agents. Agents should know exactly how they are scored, before they are scored.
  6. Set a coaching rhythm. A short one-to-one after each review batch, with one or two specific behaviors to work on, not a list of ten.
  7. Report monthly. Show score trends, top call drivers, calibration variance and the process fixes you have made.
  8. Review the scorecard quarterly. If customers or products change, the standard should too.

Recording and monitoring: check the law first

Before monitoring, confirm you are allowed to record. In the US, federal law sets a floor and states differ. One summary from Phonexa notes that federal rules require at least one party’s consent, while some states, California and Florida among them, require all parties to be informed and agree. It also notes that when parties are in different states, the stricter rule applies, and that continuing a call after a recording notice is generally treated as consent.

The practical steps:

  • Play a clear recording notice at the start of every call, including calls routed from other states.
  • Default to the strictest rule across your caller footprint.
  • Ask your own counsel to confirm your notice wording, retention periods and rules for regulated data such as payment card or health information. Laws change, so do not rely on a blog summary for a compliance decision.
  • If you use offshore or outsourced agents, confirm where recordings are stored and who can access them. Our offshore data security checklist covers the questions to ask.

Who should do the monitoring?

There are four common setups.

Option Strength Watch out for
Team leads review their own agents Fast feedback, low cost Bias, and leads lose time for floor support
A dedicated in-house QA analyst Consistency and ownership Hiring cost in the US market, and the role is easy to under-resource
An outsourced contact center vendor’s own QA Bundled with volume You get their standard and their reporting, and less control
A dedicated offshore QA reviewer or analyst on your team Full-time focus at lower cost, reporting directly to you You must write the standard and manage the role

The split between a BPO and dedicated staff is a bigger topic than QA. We lay it out in BPO, dedicated staff or employer of record.

Our perspective

We place full-time professionals from the Philippines with businesses in the US, Australia and New Zealand, and customer-facing teams are a regular part of that. What we see is that quality monitoring usually breaks for a staffing reason, not a technical one. The tool is bought, the scorecard is drafted, and then the person who was meant to review calls is a team lead with a full queue. Reviews slip, calibration never happens, and the program quietly dies.

Our perspective — overview diagram

That is why we think of QA as a role to be staffed, not a feature to be switched on. A dedicated reviewer who spends the week listening, scoring and preparing coaching notes will almost always outperform a rotating duty. A data analyst who owns the trend reports and the call-driver tagging turns the scores into decisions. Neither role needs to be large. Both need to be consistent.

We also think owners should be honest about what they are buying. We can recruit, vet, employ and support the people. You still own the standard: what good sounds like for your customers. Write it down, show it to your agents, and let the reviewer enforce it fairly. When that division is clear, offshore QA works well.

— Josh

Staffing the program with Ramp Remote

If you want the reviewing and reporting done by dedicated people, we can recruit and manage them for you. You meet and choose from vetted candidates, and one monthly fee covers salary, compliance, equipment and ongoing support. See the full list of roles and rates and the breakdown of what a quote includes.

Roles that fit a monitoring program:

  • Customer Support Specialist, from $1,400 per month per person. Useful for frontline capacity, so your team leads can step back from handling queues and spend time coaching.
  • Data Analyst, from $1,600 per month per person. Builds the QA dashboards, tags call drivers and reports trends. See the data analyst page.
  • Operations Coordinator, from $1,500 per month per person. Keeps the calibration calendar, coaching follow-ups and process fixes moving.
  • QA tester, if you also need software or workflow testing. See the QA tester page.

We do not publish a separate dedicated call-QA analyst rate, so ask us about your exact need and we will scope it with you.

Common mistakes to avoid

  • Scoring to punish. If agents associate QA only with discipline, they learn to game the scorecard, not serve customers.
  • Chasing average handle time. The ICMI paper warns that efficiency-only targets can encourage rushed service.
  • No coaching loop. Scores that never become conversations are overhead.
  • A scorecard nobody has tested. Pilot it on old calls before you use it for reviews that affect pay.
  • Ignoring systemic causes. If many agents miss the same item, look at the knowledge base, the tools or the procedure first.
  • Trusting AI scores without checking. Compare machine and human scoring regularly.

FAQ

What does QA mean in a call center?

QA stands for quality assurance. It is the process of reviewing customer interactions against written standards, scoring them, and using the results to coach agents and improve processes. Monitoring the calls is one part of it.

What is the 80/20 rule in call centers?

It is a rule of thumb that most problems come from a minority of causes. In QA, that means a small number of call drivers or process failures often account for most dissatisfaction or errors. Tag calls by reason and fix the biggest drivers first.

How do you become a QA analyst in a call center?

Employers typically look for contact center experience, often two or more years, plus strong analytical and communication skills and comfort with tools like Excel. Familiarity with scorecard design and calibration helps, and experience with AI interaction analytics is increasingly preferred. Specific certifications are not always required.

How often should we hold calibration sessions?

Monthly is a sound default, as NiCE recommends. Run them more often while you pilot a new scorecard or onboard a new reviewer.

Do we need software to start?

No. You can start with recordings, a spreadsheet scorecard and a calendar. Software and AI scoring become worth it when volume makes manual sampling too thin to be representative.

Sources

Good people, more possible.