Call center agent training: onboard, practice and assess your agents
In a contact center, call center agent training teaches new hires to use the tools, follow procedures and handle the conversation with the customer. It starts with a few weeks of classroom training and shadowing, then continues through nesting and every time an offer changes. Training and quality managers measure its effects with a call quality scorecard and a few operational metrics.
What a call center agent needs to master.
A call center agent, or customer service representative, answers customers by phone, email, chat or social media. France Travail, France's public employment service, describes the job as solving problems, recommending suitable offers, handling complaints and following up on cases, and lists self-control and autonomy among the soft skills it requires.
In a contact center, these skills fall into three sets.
| Skill set | What it covers | How to check it's acquired |
|---|---|---|
| Tools | Phone system, CRM, knowledge base, billing or order screens | The agent finds the account and the right information while the customer is talking |
| Procedures | What the agent can decide alone, escalations, authorized credits, required disclosures, data protection | On typical cases, the agent makes the right call without asking a supervisor |
| Conversation | Greeting, questioning, rephrasing, explaining, delivering bad news, closing | On full calls, including with difficult customers, scored with a scorecard |
Tools and procedures, which are specific to each company, are fairly easy to check with a quiz, a case exercise or a review of a handled case. Conversation takes full calls, simulated or live, and someone listening with a scorecard.
In short
An agent needs to master the tools, the company's procedures and the conversation, which is learned on full calls.
How call center agent training usually works.
In France, newcomers to the field can earn a state-recognized remote customer service qualification in about six months at Afpa, the national adult training agency. Training providers also sell short courses on phone etiquette or difficult customers. These don't replace the employer's own training, since offers, tools and procedures differ from one company to the next. In their survey of nearly 2,500 call centers in 17 countries, including the United States, David Holman, Rosemary Batt and Ursula Holtgrewe find that employers train new hires on products, software, processes and customer interaction themselves.
In many centers, new hires start in training classes, groups hired together for the same line of business. They first spend several days in the classroom with a trainer, on offers, tools and procedures, with a few roleplays between trainees. During shadowing, the new agent first listens in on an experienced colleague's calls, then takes calls while a mentor listens and corrects. In the nesting phase that follows, the agent handles simple call types alone, sometimes in a group reserved for new hires, and the scope widens week after week.
According to the Holman, Batt and Holtgrewe survey, published in 2007, new hires typically receive 15 days of initial training, a figure fairly consistent across countries. They then take 11.5 weeks on average to become proficient, from 4 weeks in Israel to 25 in the UK and Ireland. Outsourcers train for less time than in-house centers, 14 days versus 20.
In short
Employer training moves from the classroom to shadowing and nesting. It typically lasts three weeks, then agents take almost three months to become proficient.
Where training breaks down.
In the classroom, conversation often comes after offers, tools and procedures. It is practiced in roleplays between trainees who know each other and find it hard to play an angry customer convincingly. Agents therefore face their first difficult complaints and cancellation threats with real customers.
Xiangmin Liu and Rosemary Batt measured the effect of supervisor feedback by following 2,327 directory assistance operators at a US telecommunications company for five months, all with more than six months on the job. The more coaching an operator received, mostly feedback on monitored calls, the more their handling time dropped afterward. But each one got about one hour of coaching a month, and the time each supervisor spent on it ranged from 5 to 20 hours a week. Many supervisors hold back from coaching, the authors note, for lack of time or confidence.
Part of this feedback can be handed to an AI coach, with the supervisor keeping sensitive cases. The studies on the subject are summarized in our guide AI coach or human coach.
Attrition also forces trainers to restart the same program often. In the Holman, Batt and Holtgrewe survey, the median turnover rate is 20% a year, promotions included, and a third of agents have less than a year of tenure. In France, 52% of hires for remote customer service jobs are on contracts of less than one month, according to Acoss and MSA data published by France Travail.
Once proficient, agents receive only 6 days of training a year on average, according to the same survey. What was covered in the classroom and not revisited afterward fades quickly, as the forgetting curve shows.
In short
In the classroom, conversation often comes after tools and procedures. After that, feedback depends on supervisors' time, and attrition means the same programs keep restarting.
Customer service roleplay scenarios to practice first.
Common call types are learned quickly through shadowing, because they come up every day. The eight customer service roleplay scenarios below, from the angry customer to the cancellation threat, come up less often and are more tense, and a mistake can lead to a complaint or a cancellation.
Complaint from an angry customer
- The scene
- The customer calls about a billing error or a service outage. They are angry and want an answer now.
- Where it goes wrong
- The agent gets defensive, quotes the procedure too early or offers a credit to end the call faster. Answers vary from one agent to the next.
- What to look for
- Listening without interrupting, rephrasing, a specific solution with a timeframe, a credit only if it's justified.
Customer threatening to cancel
- The scene
- After a badly handled incident or a string of disappointments, the customer says they're switching to a competitor.
- Where it goes wrong
- The agent argues or offers a discount when the customer first wants to be heard.
- What to look for
- Acknowledging what went wrong, finding the real reason for leaving, an offer that addresses it.
Delivering bad news
- The scene
- The agent has to announce a refused refund or an outage that will last, when the customer was expecting a different answer.
- Where it goes wrong
- The message is vague or falsely reassuring, and the customer calls back.
- What to look for
- A clear message from the start, the explanation, what the agent can do, the next step with a date.
Frustrated multichannel customer
- The scene
- The customer has already written by chat or email without getting an answer, and calls in irritated.
- Where it goes wrong
- The agent makes them explain everything again and absorbs anger that built up on the other channels.
- What to look for
- Reading the history before asking questions, acknowledging the steps already taken, a commitment to close the case.
The real problem behind the request
- The scene
- The customer wants to change plans or remove an option. Behind it, there is sometimes a bill they don't understand or a cancellation in the making.
- Where it goes wrong
- The agent processes the request without asking any questions, and the customer calls back a few days later.
- What to look for
- Questions about the context, rephrasing the real need, consequences spelled out (contract term, billing).
Incident to assess
- The scene
- A customer reports an outage. The agent has to gauge its impact to set the priority and alert the right teams.
- Where it goes wrong
- The agent sets the priority based on the customer's tone, or promises timeframes they don't control.
- What to look for
- Questions about impact, scope and acceptable delay, commitments the team can keep.
Transferring a case
- The scene
- The case has to go to another department, a specialist or a colleague.
- Where it goes wrong
- The transfer is announced abruptly and feels like a brush-off, then the colleague makes the customer repeat everything.
- What to look for
- Explaining the transfer and what happens next, passing on the context, a handover that shows the case is known.
Upselling after resolution
- The scene
- The customer's request is resolved and they are satisfied. It's a good moment to suggest a plan that better fits their usage.
- Where it goes wrong
- The agent misses the opportunity, or starts selling before the problem is solved.
- What to look for
- Resolution first, a question about usage, an offer that follows from the answer, a no accepted without pushing.
For each scenario, plan several customer profiles, from calm to aggressive. Start from the company's actual complaint and cancellation reasons, as they appear in customer comments and QA reviews.
In short
Eight scenarios, from complaints to upselling, need dedicated practice, built from the company's real call reasons.
The call quality scorecard.
The quality scorecard, or QA form, describes what a good call looks like at your company. The quality team uses it to score recorded calls. If supervisors and trainers use the same scorecard for side-by-side coaching and roleplays, agents know from their first week what they will be evaluated on.
To stay usable, the scorecard fits on one page, and each criterion describes an observable behavior that two evaluators would score the same way. Each criterion can be scored on three levels (met, needs work, not observed). The template below has ten criteria, to adapt to your business.
| Step | Criterion | What to look for |
|---|---|---|
| Greeting | Introduction and verification | Greeting, customer identity verified per procedure, calm tone |
| Discovery | Questions and listening | Open questions about the situation, the customer speaks without being cut off |
| Rephrasing | Understanding the request | The agent rephrases the need and the customer confirms it |
| Answer | Accuracy | Correct information, procedure followed, no long silences while searching |
| Solution | Offer and commitment | Specific solution, realistic timeframe, credit only if justified |
| Emotion | Handling an upset customer | The emotion is acknowledged before any explanation, the agent stays calm and courteous |
| Clarity | Explanations | Words the customer understands, consequences stated (billing, contract term, timing) |
| Closing | Recap and next steps | What was done, what happens next, checking there is no other request |
| Case notes | Documentation | Reason, actions and commitments logged in the CRM, readable by a colleague |
| Compliance | Disclosures and data | Required disclosures, call recording announced, personal data verified and protected |
A technical help desk will give more weight to incident assessment, a retention team to the reason for leaving. To make sure evaluators score the same way, the quality team can run calibration sessions, where several people score the same call and then compare their scores.
Template to download: call quality scorecard and scenario card
The ten criteria with their three scoring levels and a comments column, for training, side-by-side coaching and QA, then a card to write your own call scenarios, as a two-page PDF.
In short
The scorecard describes a good call in observable criteria. The same scorecard serves in training, side-by-side coaching and QA.
The metrics contact centers track.
Three of a contact center's operational metrics matter directly for training. The definitions and benchmarks in the table come from the standards New York State sets for its agencies' call centers; benchmarks vary by business.
| Metric | Definition | What it says about training |
|---|---|---|
| First contact resolution (FCR) | Share of requests resolved on the first call, with no callback. Benchmark cited: 70 to 75%. | A low rate can point to a poorly diagnosed request or an answer the agent can't find in the tools. |
| Average handle time (AHT) | Total time of a call, conversation plus the work that follows (data entry, follow-up). Benchmark cited: 4 to 6 minutes. | It drops as agents master the tools. When very low, it can hide calls cut short. |
| Customer satisfaction (CSAT) | Rating given by the customer in a post-call survey. Benchmark cited: 80 to 90% satisfied customers. | It mostly reflects the conversation, as the customer experienced it. |
Definitions and benchmarks: New York State, “Call Center Standards.”
Many companies also track the Net Promoter Score, which Fred Reichheld introduced in the Harvard Business Review in 2003. Customers say from 0 to 10 how likely they are to recommend the company to a friend or colleague, and the share of detractors (0 to 6) is subtracted from the share of promoters (9 and 10).
On the ICMI website, Scott Sachs observes that pushing agents to cut AHT leads to rushed calls and unanswered questions. He also notes that many companies don't count a transferred call as resolved on first contact, which can lead agents to guess at answers rather than hand the call over. Service level, for example 80% of calls answered within 20 to 30 seconds in the New York standards, depends mostly on scheduling.
To track training itself, add the time each agent takes to handle every call type alone, and their scorecard results, criterion by criterion, which show earlier than AHT what they are missing.
In short
First contact resolution, average handle time and satisfaction should be read together, and the scorecard shows earlier what an agent is missing.
Practicing in simulation before live calls.
In 2008, well before AI-based simulations, Nagesh Murthy and his colleagues compared two kinds of training in the call centers of two Fortune 50 companies, one based on simulation and the other on roleplay. On the live calls that followed, agents trained with simulation handled requests more accurately and faster, and the gap widened as calls became more complex.
In an AI simulation, the agent talks out loud with a customer played by AI. Agents can practice the scenarios above on their own, during onboarding or between calls, and each attempt is scored with the quality scorecard. Supervisors keep side-by-side coaching for real customers and cases outside the scenarios.
In Engramm's AI roleplays, each scenario is built from your own documents, such as procedures, product sheets, call scripts or evaluation rubrics. The character responds in under 500 ms, with its own mood and objections. After the call, the agent gets a scorecard based on your criteria and works on weak points with the AI coach. Trainers follow each agent's progress and the class's weak points, scenario by scenario. Voice simulations work in 70 languages. Contact centers can be billed per minute of practice.
French carriers Orange and Free and contact center outsourcer Konecta train their teams with Engramm. Our telecom page shows typical scenarios for telecom contact centers, from retention calls to technical complaints.
To get started, pick the three call types that generate the most complaints or callbacks, write a scenario for each with two customer profiles, and have your next training class go through them before their first day of shadowing.
In short
In the 2008 study, agents trained with simulation then handled live calls more accurately and faster. With AI, each agent can practice difficult calls alone, scored on the quality scorecard.
Frequently asked questions.
How long does call center agent training take?
In a survey of 17 countries, initial training for a new agent typically lasted 15 days, or three working weeks. Agents then took almost three months to become proficient on every call type.
What skills does a call center agent need?
France Travail, France's public employment service, lists self-control, autonomy, teamwork and initiative. New York State's call center standards stress soft skills such as empathy and active listening, along with product knowledge and regulatory compliance. Clear speaking and comfort with software tools also count, and both can be developed in training.
How do you handle an angry customer on the phone?
Let them explain the problem without interrupting, rephrase what you understood and acknowledge what went wrong before explaining anything. Then offer a specific solution with a timeframe you can keep. If the answer won't please them, say it clearly and explain what happens next.
What is the 80/20 rule in a call center?
It refers to a service level target, answering 80% of calls within 20 seconds. New York State's standards for its agencies give a similar example, 80% of calls answered within 20 to 30 seconds. Service level depends mostly on staffing and scheduling; the quality of the answer is measured with first contact resolution and customer satisfaction.
Sources.
- France Travail, job profile “Conseiller / Conseillère clientèle à distance” (ROME D1408), accessed October 30, 2026 (in French): candidat.francetravail.fr
- Afpa, “Formation qualifiante Conseiller relation client à distance” (in French): afpa.fr
- David Holman, Rosemary Batt and Ursula Holtgrewe, The Global Call Centre Report: International Perspectives on Management and Employment, Global Call Centre Research Network, 2007: ecommons.cornell.edu
- Xiangmin Liu and Rosemary Batt (2010), “How supervisors influence performance: A multilevel study of coaching and group management in technology-mediated services,” Personnel Psychology, 63(2), 265-298: ecommons.cornell.edu
- New York State Digital Service, “Call Center Standards,” accessed October 30, 2026: ny.gov
- Frederick F. Reichheld, “The One Number You Need to Grow,” Harvard Business Review, December 2003: hbr.org; Bain & Company, “Measuring Your Net Promoter Score”: netpromotersystem.com
- Scott Sachs, “Metrics: Do They Accomplish What You Want?,” ICMI, August 18, 2014: icmi.com
- Nagesh N. Murthy, Goutam N. Challagalla, Leslie H. Vincent and Tasadduq A. Shervani (2008), “The Impact of Simulation Training on Call Center Agent Performance: A Field-Based Investigation,” Management Science, 54(2), 384-399: doi.org


