• FryAI
  • Posts
  • Thinking Machines On The Phones: Replicant AI's Call Center Revolution

Thinking Machines On The Phones: Replicant AI's Call Center Revolution

Welcome to this week’s Deep-fried Dive with Fry Guy! In these long-form articles, Fry Guy conducts an in-depth analysis of a cutting-edge AI development or developer. Today, our dive is about Replicant, a platform for AI-driven call centers. We hope you enjoy!

*Notice: We do not gain any monetary compensation from the people and projects we feature in the Sunday Deep-fried Dives with Fry Guy. We explore these projects and developers solely for the purpose of revealing to you interesting and cutting-edge AI projects, developers, and uses.*


(The mystery link can lead to ANYTHING AI related. Tools, memes, and more…)

Don’t you just love customer service calls? You call the help line to solve your issue, wait on hold for two hours, explain your situation, get transferred, wait on hold again, and then get transferred again … rinse and repeat! This process is one of the more frustrating sides of the customer experience and can also leave companies with bad reputations.

Replicant.ai serves as a contact center automation platform that helps with customer service. It offers human-like AI voice-bots which can answer questions specific to the company and help you solve your problem much faster than traditional methods.


Benjamin Gleitzman is the CTO and co-founder at Replicant.ai. He grew up in Morgantown, West Virginia and became interested in robotics from a young age. Gleitzman’s true passion of technology was born, however, when he received the opportunity to go to Japan on an exchange program. Through this experience, he became mesmerized with how Japan seemed to be “living in the future” with their technology. He ended up getting a job at a company called Advanced Telecommunications Research. At that time, the president of that company had created an android, humanoid version of himself. When this person would leave on trips, he would leave the android in his office to take his place. Gleitzman was in awe of this technology, but he found a flaw: everything about the robot seemed like a real human, except “as soon as it opened its mouth, you could tell that it was fake.” This experience stuck a cord within him to explore human-like voices.

Fast-forward twenty years … Gleitzman and his co-founders began to wonder why, given the fact that everyone has Siri in their pocket and Google Voice in their home, can we not have seamless conversations with a machine? As Gleitzman said, “It seemed there was some missing piece as to why we are not having these great conversations with machines, and so we set out to craft great conversations.”


In the past, conversations with machines have been choppy and misleading, leaving many with bad experiences. However, with the recent surge of voice synthesis combined with the development of AI and large language models, we are entering a space where these robots deserve a second chance.

Let’s paint a picture of how Replicant works from the caller side. Imagine your car breaks down, so you decide to call roadside assistance. If Replicant is implemented on the other line, the call is answered instantly with no hold time. On the phone with the thinking machine, you can describe your problem, such as a flat tire. Replicant can, while simultaneously talking to you, call outbound to all the tow truck shops nearest to you and ask each of them if they are available to pick you up until it finds you a match. In this way, it connects inbound phone calls and outbound phone calls to give the user the most efficient experience, all while staying on the same line.

Replicant began their experimentations with restaurant orders. They would call into restaurants using these “thinking machines,” which would speak to people who had different phone clarity, accents, and who were in a rush. The people taking the orders said they loved this experience because the machine was straightforward, polite, and there was no unnecessary chit-chat—it made the ordering process more efficient. This is the goal of Replicant, not to “create more unwanted calls or ‘robo calls’ in the world, but to look for places where there is an inefficiency and create some resolution there.”

Since these early tests, Replicant has valued an open and honest approach to their communication. Though they work extremely hard to make their voices sound indistinguishable from a human, they always disclose to the caller at the beginning of the conversation that “I am a thinking machine on a recorded line.” This can sometimes turn off human callers who don’t believe an AI agent can solve their problems, in which case Replicant has implemented a number of conversation design principles. For example, if someone wants to speak with an agent, the thinking machine might say, “I’m happy to get you over to an agent, but it is going to be a 25 minute wait time. In the meantime, can I get your policy number?” This is a way for the thinking machine to pull the person back into the conversation and train them that “they might have had some negative experiences with machines in the past, but this is a new generation of this technology, so give it a chance.”

Replicant is focused not only on speaking like a human, but also on listening and understanding. This includes not only processing the words people are saying but understanding the vernacular people are using as well. As Gleitzman points out, “In West Virginia, people don’t speak how they do in California. In Detroit, they are not going to use the kinds of phrase people do in New Orleans. We want to understand everyone equitably, and not just people who sound like me.” The thinking machine is also focused on the intent of the call, rather than targeting or flagging specific words. For example, if someone swears at the AI, the bot will most likely ignore that and focus on the other content which was said by the person, as long as the conversation keeps going forward. Of course, companies can use Replicant in a way to implement their own policies, but this design is built with an understanding of human emotion and frustration, especially when it comes to issues people have with products or services provided by the company.

In addition to helping the company’s side of the call, the thinking machine is also programmed to do much of the heavy lifting for the caller. For example, if booking a hotel, the person should not have to define and verify how many adults and children they have. Instead, powered in large part by large language models, Replicant has a focus on understanding humans as humans. To illustrate, if someone says “It’s gonna be me, my hubby, and my 9-month-old,” the thinking machine should be able to understand that as two adults and one child given contextual reasoning. This allows for more natural, seamless communication that not only speaks like a human, but understands and thinks like one as well.


The average call center has 7 months of retention for their employees, and that comes at no surprise. As Gleitzman explains, “It is a difficult job. You are churning through tickets, people are angry and yelling at you, you have to get up to speed on regulations, and most tasks are mundane.”

The idea behind Replicant is not to eliminate the need for human agents, but to free up humans to deal with more complicated issues and less mundane tasks. This is an elevation of the role of humans in the contact center. Gleitzman tells of his experience working at a call center: “I loved when I got those interesting calls that required research and creativity as a human rather than my 17th password reset of the day.” Replicant does all the mundane identification of the person and narrowing down of the issues, so if and when the call is escalated from the thinking machine to the human agent, the customer can be in and out on that call in 30 seconds rather than a 45 minute hold time followed by authenticating themselves and explaining their situation over and over again. For this reason, Replicant calls tend to be 35-40% shorter than human-to-human calls. Using the thinking machine on the frontlines allows humans to do what they are good at—empathy, creativity, and building relationships—rather than trying to churn through as many tickets as they can in an hour, most of which are mundane and frustrating on both ends of the call. This elevates the level of humans at call centers, giving the human agents more meaning and fulfillment each time they pick up the phone.

Given the help the thinking machine gives human agents, it comes as no surprise that when Replicant was deployed at a large contact center, the disability days their agents were taking dropped by 70%. In other words, 70% less people were sick on the job (or at least called in sick), when the Replicant thinking machine began to serve on the frontlines of the customer service calls, filtering out some of the most frustrating and mundane tasks for human workers.


Customer service calls need an overhaul. On one side of the phone, there is a frustrated customer in need of resolution. On the other side of the line, there is a frustrated agent who has been dealing with the brunt of negative human emotion all day. On both sides of the phone, there is little enjoyment in these conversations.

Customer service calls might not ever be particularly enjoyable experiences, but they can certainly be more efficient ones. This is what Replicant is here for: to give callers a more efficient way to get their issues resolved and to elevate the role of human agents to use their creativity and bring fulfillment to their calls.

Calls with a machine are not like they were twenty years ago, ten years ago, or even five years ago. Thinking machines are here, they are on the other line, and they are ready to be used.

Want to try talking to the thinking machine?: Replicant Interactive Demo