How is AI simulation reinventing contact center agent training? (feat. ElevenLabs)

Episode topics

• 00:00 - What's the opportunity in modern agent onboarding?
• 01:32 - How do you design the future of learning and development?
• 03:09 - How do you boost agent tenure in the contact center?
• 05:57 - How does employee growth connect to customer loyalty?
• 06:52 - What is Fuel iX Agent Trainer?
• 07:57 - What does faster learning with AI look like in practice?
• 11:45 - How does ElevenLabs voice AI make training more realistic?
• 14:00 - What does an AI learning scenario look like in practice?
• 17:29 - How can you use AI to master emotional intelligence?
• 22:17 - How can AI-driven insights improve coaching?
• 25:15 - What does the future of high-performance hybrid workforces look like?

Transcript

[00:00:00] Robert Zirk: Every day, contact centers around the world onboard new agents. And the quality of training during those first days and weeks is a major factor in whether they complete that onboarding process.

[00:00:13] BambooHR research indicates that 70% of new hires have already decided if a role is right for them by the end of their first month — and that includes the 29% who make this decision within their first week.

[00:00:27] Employees need to feel adequately prepared and supported as they begin a new role and undergo training. When they don't, the costs can be significant — not just in attrition, but in lost productivity, team morale and the overall customer experience.

[00:00:44] Today on Questions for now, we'll explore how CX agent training has evolved in a way that's cultivating more skilled agents, building confidence faster and creating better customer experiences from day one. i'm joined by Mitch Lieberman, vice president of product for Fuel iX™ at TELUS Digital, and Jack Piunti, a go-to-market leader at ElevenLabs, as we ask: How is AI simulation reinventing contact center agent training?

[00:01:20] Welcome to Questions for now, a podcast from TELUS Digital where we ask today's big questions in digital customer experience. I'm Robert Zirk.

[00:01:32] For decades, contact center training for customer service agents has involved a combination of classroom training, in-person role plays and listening to calls.

[00:01:42] But lots has changed and the role of an agent has evolved.

[00:01:47] I spoke with Mitch Lieberman, vice president of product for Fuel iX at TELUS Digital. Fuel iX is an award-winning generative AI engine designed to help enterprises scale AI applications safely and responsibly.

[00:02:01] Mitch highlighted the high standards agents are held to.

[00:02:04] Mitch Lieberman: There's multiple dimensions to what a customer service agent needs to do. There's the ability to have foundational knowledge about the product or service that they're supporting. That's just table stakes. Whether it's a phone company or an internet service or banking or healthcare, you need to understand and have the foundation of what it is that you're supporting.

[00:02:26] You have to learn some skills like how to probe for information and get to the question the customer's actually asking. Communication skills and process-type questions. There's also script adherence. There are very specific questions that you need to ask: authorization, authentication, who you are, what kind of information is required.

[00:02:49] And compliance. When you send a text message, as an example, your rates may apply. Things you have to say: refund policies, return policies, cancellation policies. And so it's multidimensional.

[00:03:01] And finally, digital competency. You're gonna be using systems and applications and you need to understand how to use those systems and applications.

[00:03:09] Robert Zirk: Between product or service knowledge, process knowledge, compliance, and digital competency, there's a lot for new agents to learn, understand and master. This underscores why it's important to continually improve agent training processes.

[00:03:27] The cost of training that doesn't meet the need of new agents shows up strongly in one particular metric: agent churn. And Mitch notes that it's one of the most expensive problems in customer experience management.

[00:03:40] Mitch Lieberman: Churn is basically what percentage of your overall agent population leaves each year. And it's a very expensive proposition because onboarding takes time. Industry-wide, and it varies across domains or industries, but in general, it can be 30 to 50% per year. It's not tenured agents. Once you get past 18 months or two years, your likelihood to churn goes down. It's that first year and a half that's really tough.

[00:04:08] Robert Zirk:   That's why the human agent experience has been a top priority for Mitch and the Fuel iX™ team at TELUS Digital.

[00:04:14] Mitch Lieberman: So we're really working hard to help people to feel comfortable in that first year. So good looks like reducing that 30 to 50 to 20 to 30, being realistic.

[00:04:25] If you can make the onboarding experience positive and people feel like they're ready to take on these tough customer conversations, the knowledge, the compliance, the scripts that we just spoke about, we can reduce the churn percentage, right? So we can help people to feel more comfortable in the situation.

[00:04:43] Robert Zirk: Up until recently, one of the biggest gaps between training and an agent handling a live customer interaction has been that, as much as classroom settings provide important knowledge, the real world is different. It's more complex, more emotional and more dynamic.

[00:05:00] Mitch Lieberman: I have friends who spend time on the golf course. You can have an idea that, "Okay, when I'm here, this is the putter and this is the driver." You can't learn to golf in a classroom.

[00:05:10] Robert Zirk: When agents feel unprepared or rushed through training, they step onto the floor without the confidence they need. And that lack of confidence often leads to higher churn rates as agents leave before they've had a chance to truly succeed in the role. High churn means lost productivity, increased hiring costs, and ultimately, a negative impact on customer experience.

[00:05:34] The higher your agent churn rates, the more likely customers are to reach inexperienced agents, which can mean customers experience longer wait times, slower problem resolution and less empathetic interactions. Those factors contribute to customer churn. In that way, agent churn can directly damage customer loyalty.

[00:05:57] But Mitch pointed out what happens when you optimize training processes to increase agent engagement.

[00:06:02] Mitch Lieberman: There's a direct connection between satisfied customers and their willingness to stay on with you as a customer. So customer lifetime value. Positive employee experiences are also known to lead to positive customer experiences. So to connect that dot, to connect that string, positive employee experience, positive customer experience, increased loyalty to the brand, long-term customer value. And everybody's happy.

[00:06:32] Robert Zirk: Building confidence, especially when handling more complex issues, is key to feeling ready to handle real customer calls. The challenge is that traditional role-playing doesn't scale well — and even if it did, it wouldn't meet the same level of realism that an agent would encounter on a day-to-day basis.

[00:06:52] So how can you create an environment where agents can fail safely and feel confident by the time they answer their first real call?

[00:07:00] That's what Mitch and his team at TELUS Digital have developed through Fuel iX Agent Trainer, a training simulation that uses AI to role play realistic customer conversations, powered by voice technology from ElevenLabs.

[00:07:14] Mitch Lieberman: Agent Trainer is a role play platform which gives people a comfortable place to get started. For some, the content of Flight Simulator feels good and for others, sandbox or dress rehearsal feels good, but they both state the same sort of purpose, which is where we use generative AI to allow an AI to pretend to be a customer and, using voice, giving new associates or tenured associates or agents the ability to practice tough conversations. We're moving into helping leaders to coach. How do you have those tough conversations in a safe environment where you can get something wrong and get feedback on that?

[00:07:57] Robert Zirk: So how do the AI training simulations provided through Agent Trainer translate to faster agent readiness and better customer outcomes?

[00:08:06] Consider the case study of TELUS Communications, which, like TELUS Digital, is a subsidiary of TELUS Corporation. TELUS Communications is a world-leading technology company with more than 15 million customer connections. When customers contact TELUS, they expect an expert who can resolve their issue on the first interaction across wireless services, high speed internet, home security, television and more.

[00:08:33] In the context of this case study, traditional training methods required up to 10 weeks to prepare agents to support that range of services. So TELUS Communications partnered with TELUS Digital to accelerate that process with Fuel iX™ Agent Trainer by creating voice and chat simulations based on actual TELUS support scenarios.

[00:08:54] As a result, roleplay training productivity increased by 75 to 85%. Two to three full scenarios, which used to take several hours of training, could be covered in just 45 minutes. And coaching efficiency also increased by 26%, allowing supervisors to guide more agents more often, scaling their support and expertise.

[00:09:19] Mitch shared another advantage from the agent perspective based on feedback his team has received.

[00:09:25] Mitch Lieberman: When customer service agents are interacting with an AI, they know it's an AI, but that's okay because they also know that there's not somebody in the other end of the phone line grading their performance. We will get to that step to make sure that we can coach them and get better. But in the beginning, it's all about just working through the conversation and understanding it. And the feedback that we received from the customer service agents in the beginning is it was great to not be great in those first set of interactions. They just wanted to try it.

[00:09:55] Robert Zirk: This illustrates the value of a safe environment that allows agents to practice and learn from their mistakes without the concern of providing a wrong answer to a real customer.

[00:10:06] Traditional in-person role plays lack the realism that builds true confidence. Your agents probably aren't professional actors. And Mitch noted that hearing someone speak in person isn't the same as what you might hear over a phone line.

[00:10:21] Mitch Lieberman: In a classroom, I have a leader — let's say you're my leader — I get your voice and I wouldn't believe you if you tried to pretend to stutter or to kind of add static, you know, or to do something that's not natural. If an agent doesn't believe that the simulation is valid and it feels robotic and it feels computer AI, then they won't take it seriously or they won't feel like they're actually getting it.

[00:10:46] Robert Zirk: Real life scenarios are imperfect and simulations need to evoke that.

[00:10:51] Mitch Lieberman: I want static lines, baby crying in the background, dog barking in the background, car noise, car horns. I want a variety of accents, ages, understanding. I want an interesting mix between men and women. And we get that with AI 'cause we can tune it.

[00:11:13] So what we get is we get a wide variety. I can take a training simulation that is a clear, crisp, 40-year-old young woman who is asking a question, who is clear as day and switch that to somebody that's of a different age, older or younger, who has a thicker accent of a variety of means, or there's static in the background, and try to ask the same question. And so I get the variety, I get the distribution across different types without changing the actual call type. It's just the voices because that's reality.

[00:11:45] Robert Zirk: And to help enable that realism, TELUS Digital partnered with ElevenLabs, a voice AI research and development company focused on building and deploying realistic, expressive, low-latency voice technology platforms. Their ElevenAgents platform powers the voice simulations that agents interact with during training.

[00:12:07] Jack Piunti leads go-to market engagements with enterprise partners and customers at ElevenLabs. He explained that when you're training someone for a job where voice is the primary method of communication, your training needs to happen through voice as well.

[00:12:22] Jack Piunti: You move from a cognitive exercise to an experiential one, right? So reading a script on screen and then typing a response, it exercises completely different parts of the brain than actually talking to someone who sounds upset, who interrupts you, who pauses because they're confused, who you maybe can't understand because they're speaking really quietly. In a contact center, we're training people to be on the phones. Everything happens through voice, and so the tone of the caller, the pace of the conversation, the ambient frustration or relief or anxiety — if your training doesn't include those elements, you're preparing agents for a different job than the one they're actually going to do when they go live and start taking calls.

[00:13:02] Robert Zirk: To make this work seamlessly, ElevenLabs has engineered their voice AI technology so that voice synthesis models, conversational AI and latency optimization all work in concert. Jack explains how this integration creates that natural, human-like experience.

[00:13:21] Jack Piunti: We've built all of those models to work together so that you can have a really natural sounding, low latency, basically indistinguishable from human conversation. And so what the TELUS Digital team has done is taken that foundation and built Agent Trainer, built simulated customers on top of ElevenAgents, focusing on the domain knowledge of what those training scenarios look like, what a contact center agent would need, while allowing ElevenLabs to do what we do really well, which is that voice AI piece and the orchestration of the voice component of an AI agent.

[00:13:57] Robert Zirk: So what does that actually sound like in practice? Let's hear a brief example of how an agent-in-training might work through a realistic customer scenario using Fuel iX™ Agent Trainer, which leverages ElevenAgents. In this demonstration, a middle-aged customer is having difficulty accessing their phone.

[00:14:16] Toby: Good morning. My name is Toby. How can I help you?

[00:14:18] Customer: Good morning, Toby. I'm calling because I'm having a bit of a problem with my new phone. I just got it yesterday and now it's locked and I can't remember the PIN. I'm calling you from my friend's phone right now.

[00:14:30] Toby: I'm very sorry to hear that. I'd be happy to help. Do you know what kind of phone you bought? Maybe it's an Apple iPhone or some kind of Android phone.

[00:14:37] Customer: I believe it's an Android phone. I set it up yesterday, but now I can't get past the lock screen.

[00:14:44] Toby: Do you remember if you set up some kind of ID with your face or maybe via your finger?

[00:14:49] Customer: Yes, I did. I used my index finger to set up fingerprint verification.

[00:14:54] Toby: Okay, great. So what I'd like you to do is place your index finger on the button in the bottom center of your phone and just leave it there for a moment.

[00:15:03] Customer: Oh, it worked. My phone is unlocked. Thank you so much. I can see everything now.

[00:15:08] Toby: Great! Have a great rest of your day.

[00:15:10] Robert Zirk: That's the kind of realistic practice environment that's transforming how contact centers prepare their agents. That realism comes from the ElevenAgents platform, which, as Jack noted, is powered by an incredibly diverse voice library.

[00:15:25] Jack Piunti: I think it really stems from ElevenLabs's origins as a global company, right? We're over 400 employees in 50 plus countries. Our co-founders are Polish, grew up in Warsaw, educated in London. We've got huge hubs in London, Warsaw, New York, San Francisco, Tokyo, Bangalore, São Paulo, I mean, truly a global organization. And that's reflected in the variety of voices available in the ElevenLabs voice library.

[00:15:51] There are over 11,000 voices in over 70 languages with every regional accent and sub-accent, right? So you know, Spanish on the Iberian Peninsula is very different than Mexican Spanish, which is very different than Columbian or Argentinian, and then you have sub-accents inside those countries, just like we do in English. Differences between the west coast of Canada versus Toronto versus New York, where I'm from in California, where I live now in Indiana. Like, those are very different.

[00:16:20] And so being able to train folks in the language and the accent that they're going to be serving customers in means that the agents get exposure to all of the different varieties of the way that people speak, again, before they take that first real call.

[00:16:35] Robert Zirk: Another key component in creating realistic training scenarios is the pacing and flow of the conversation. Jack explained why the ultra low latency of ElevenLabs technology is so important in maintaining a realistic simulation.

[00:16:50] Jack Piunti: Conversation has a rhythm and when an AI's response breaks that rhythm, people immediately snap out of the experience and are like, "Okay, yep. You know, I'm doing a training scenario with an AI."

[00:17:01] You need it to be as real as possible because you want the trainee, you want the human agent to develop habits that are gonna transfer to real calls. Like, if they're adjusting their pacing and timing to accommodate AI that responds really slowly, they're actually building bad habits. It needs to feel as close to identical as possible to the real thing. And that really starts with the latency, the turn-taking, the back and forth, the conversational feel that matches a real human-to-human conversation.

[00:17:29] Robert Zirk: Reading a script on paper doesn't capture what actually happens in a real customer call. The emotion, the tone — you can describe these in parentheses, but you can't truly convey them without hearing them. That's where Jack highlights one of the strengths of ElevenAgents: the ability to convey that emotional nuance through voice.

[00:17:50] Jack Piunti: People can laugh, can cry, can whisper, et cetera. And so it's not just speaking words, it's conveying emotional states.

[00:17:57] When my internet is down or I can't figure out how to do something or my cell phone bill is 3x what I thought it would be, like, I'm going to be frustrated. And you just couldn't express that prior to ElevenLabs.

[00:18:10] The voice reflects that with changes in pace, in tone, in intensity. You can simulate a customer who's hesitant, unsure, someone who speaks fast, someone who speaks slowly, someone who needs extra patience.

[00:18:25] And that range matters because the most challenging calls aren't the straightforward ones, right? We need to help train folks in contact center and in any other customer-facing role for the edge cases, the hard cases, not just, like, that happy scenario.

[00:18:39] Robert Zirk: When training simulations include emotional nuance, agents learn to approach customer inquiries with greater empathy and understanding.

[00:18:47] Jack Piunti: A flat voice teaches you the process, right? You learn the steps. Okay, I'm gonna pull up the account, I'm gonna verify identity, we're gonna walk through troubleshooting, et cetera. An emotive voice, a natural voice, teaches you the art or the craft. You learn how to deescalate, you learn how to match the caller's energy, when to slow down, when to speed up, when to be direct, when to be a little more empathetic.

[00:19:09] And the agents who excel in contact center environments are not necessarily the ones that know the script best. They're the ones who can read the room, the ones that have that EQ to be able to read the caller and understand the emotions that are being conveyed by the voice, right?

[00:19:25] Someone saying, "Yeah, I'm fine," versus, "Yeah, I'm fine." Those are two very different responses and if you just have a monotone voice in a training scenario, you're not going to be able to have agents practice those two different scenarios.

[00:19:40] Robert Zirk: The realism Jack described, provided in a safe environment where agents are being coached and not judged, is something that became clear with Fuel iX Agent Trainer usage data.

[00:19:51] Jack Piunti: Agents were using this far more because they really liked it, like, they felt like they could get a ton of at-bats, again, without being judged.

[00:20:00] The second part of that is the confidence factor. People weren't scared anymore when they got on real calls. They'd already taken very challenging calls in their training scenario, and so they were ready. The psychological readiness turns out to be big on both agent ramp time as well as attrition.

[00:20:20] If you feel prepared, you feel ready, you get on and you're like, "Yeah," you know, "I knocked those calls out of the park." It really matters.

[00:20:28]

[00:20:28] Robert Zirk: That confidence translates directly to retention. And agent retention translates to better customer outcomes. More tenured agents have a better understanding of the products or services, know how to navigate complex customer situations, and have handled enough calls to stay calm under pressure.

[00:20:47] That experience has a compounding effect. The longer they stay, the more skilled they become, and customers benefit from that expertise.

[00:20:56] I asked Mitch how the impact of effective training simulations can be conveyed through metrics.

[00:21:02] Mitch Lieberman: Among the areas where customers get frustrated — so where can we focus on the customer outcome is the most important piece — is: Do I have to call back again?

[00:21:12] We've all been in that experience where we thought we had the answer. We get off the phone, we try the solution and it didn't work. So the indicator for that is called first call resolution or first contact resolution. So we would like that to be a high number. We'd like to solve it in one phone call.

[00:21:30] The other is what's also referred to as average handle time. Now, handle time is inside baseball or it's an inside metric. The customer wants to solve their problem, and if they believe that you're gonna help them solve their problem, they'll stay on the phone a little bit longer and we shouldn't rush them to get off the phone.

[00:21:48] The combination of these two elements, if we give agents the tools they require to solve these problems, reduce handle time, resolve question on the first contact, they feel very good about it and they've solved a customer's problem and when you give them the right tools, it's not as frustrating. You know, if you're trying dice onions, getting ready for dinner, dinner and you have a really dull knife, it's just not a fun experience. You're sitting there struggling. So the right tool for the job is super important.

[00:22:17] Robert Zirk: As the training process evolves, Mitch sees an evolution for the role of human trainers as well, enabling them to provide more personalized and effective feedback that helps reduce an agent's nesting period.

[00:22:30] Mitch Lieberman: We allow leaders to be leaders. We allow them to do the really hard things. In the industry at large, if you read articles, AI is gonna take away the boring rote stuff. And what it's gonna allow is gonna allow us to be more leaders and humans as opposed to a coach trying to evaluate if you're ready and coach you and pretend to be a customer all at the same time, which is hard.

[00:22:55] Robert Zirk: After a Fuel iX Agent Trainer scenario, agents receive immediate feedback on what they did well and where they can improve. They also get scored on specific skills on a scale from one to five. That granular feedback helps agents understand what to work on while giving leaders visibility into patterns across their team.

[00:23:16] Mitch Lieberman: They can listen in and see the evaluation criteria, look through it, they can go back and listen to some of the calls, see how you did, and then jump in and say "Here's some things that I think are really gonna happen that I would offer you some suggestions."

[00:23:30] And to really uplevel and to take coaches, leaders, trainers to kind of really live at the top of their license, which is a little bit of a healthcare term, which is to really elevate to the level that we expect and not just to do the simple stuff.

[00:23:45] Robert Zirk: But how do you know which scenarios agents need to practice? That's where quality management comes in. Mitch explained how real call data informs training processes.

[00:23:57] Mitch Lieberman: Quality management is taking actual customer-agent conversations and running them through, you know, a similar scoring mechanism as you would for a training simulation. And we want to use the same engine to rank and report.

[00:24:12] We can put together a team to say where the team is doing well or we can take individuals' last 10 calls and we can check to see if those skills and categories of skills — empathy, knowledge, all the ones I shared earlier — are still true and they're still following up.

[00:24:30] Sometimes people need a refresher and they didn't do what we hoped that they would do, but we can give it back to them. I can monitor their call after the fact. I can score it. I can then create either a coaching session or a scenario for a role play and say, if it was me, "Hey Mitch. Some of the calls that you took this week, you were struggling on your script adherence. You didn't really ask them the right questions," or "Your knowledge, in terms of asking questions to get to the core of the issue, you could use some work there. But no worry. Here's a simulation that's gonna help you focus on that. It'll just remind you to follow that process."

[00:25:05] So it is a tight loop. We call it an agility loop. You can call it an infinity loop. It goes by many names. We do close the loop and circle back with making sure people are doing the right things.

[00:25:15] Robert Zirk: To enable continuous improvement as products, services and industries evolve, Jack sees agent training moving beyond the traditional onboard and go model.

[00:25:26] Jack Piunti: Right now the model is onboard, train, throw people on the phones. We do some monitoring and maybe there's some coaching and feedback, but it's probably less than we'd all like to do because everyone's busy and you're really just hoping for the best.

[00:25:38] With AI voice simulation for training scenarios, it's accessible and scalable to have agents do a 15 minute practice session before a shift or to take some of your lower performing agents and be like, "Okay, you know what? We're gonna actually take an hour out of taking real calls. We're gonna run through a couple of training scenarios that have been set up to mimic where your weakest area has been or where you've been struggling."

[00:26:05] The beauty of Agent Trainer is it's all configurable by natural language. So a contact center manager can configure and set up a simulation for one of their employees in a matter of minutes.

[00:26:18] The most innovative companies already are at a point where they have a hybrid AI/human workforce where AI handles certain things, humans handle certain things, and those humans that are handling those things are empowered by AI.

[00:26:33] Robert Zirk: So agent training is evolving. It's becoming continuous, adaptive and powered by realistic voice simulations like the ones provided through Fuel iX™ Agent Trainer. I asked Mitch: for CX leaders considering implementing AI training simulations, what does it take to get started?

[00:26:51] Mitch Lieberman: It doesn't require a large team. We can consume evaluations that you have historically from your quality management systems, it can consume redacted call transcripts and we can create scenarios right from there. So you don't have to start with "How do I create a scenario?" We make it really easy to onboard and ramp and to try it out, but leaders, try it out. And that's the main thing. It offers a lot of value and it's not a very technologically hard product to get going.

[00:27:22]

[00:27:29] Robert Zirk: Thank you so much to Mitch Lieberman and Jack Piunti for joining me and sharing their insights today. And thank you for listening to Questions for now — a TELUS Digital podcast.

[00:27:40] For more insights on today's big questions in digital customer experience, follow Questions for now on your podcast player of choice.

[00:27:49] I'm Robert Zirk, and until next time, that's all... for now.

[00:27:53]