Tenyx introduces new conversational AI voice solution to elevate enterprise customer support

Tenyx Inc., the maker of voice artificial intelligence solutions, today unveiled Tenyx Voice, its comprehensive solution for automating enterprise customer voice interactions using large language models without having to walk through tedious menus on the phone.

The company is led by a veteran team of AI and machine learning experts who approached current concerns with AI voice systems that are difficult to work with because they are unnatural, forget user progress and suffer scaling issues. The founders are AI and machine learning experts from backgrounds such as Google LLC, Apple Inc., Amazon.com Inc., IBM Corp. and Salesforce Inc. Tenyx co-founder and Chief Technology Officer Adam Earle told SiliconANGLE in an interview that automated voice experiences can be done much better.

“The key thing is to feel a naturalistic high intelligence experience of a call,” said Earle. “I think if you call customer service today it’s probably 45 minutes on hold and you get a very brittle IVR experience: ‘Please say claims.’ ‘Claims.’ It leads us all to the same thing where we end up swearing at the poor agent.”

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

The Tenyx Voice system operates on an underlying generative AI large language model, meaning that it can understand the conversational context of the speech of the person on the phone as their speaking. In this way, it’s almost like another person on the end of the line listening attentively. Generative AI is the same technology behind AI models such as OpenAI’s ChatGPT that makes it capable of holding human-like text conversations with users.

To make conversations with the AI natural, Earle said, it also has to act and react more like another person. Machine systems are bad at being interrupted and tend to interrupt. Anyone who has used an automated phone system has experienced this problem where it begins listing out a long list of menu options and then when they try to say aloud an option it takes a few seconds after saying the option, such as “Talk to someone,” it keeps talking for a few seconds, pauses and then finally, “Do you want to talk to someone?”

Natural conversations don’t work this way, so the Tenyx Voice system is designed to allow interruptions to happen extremely quickly and it reacts within moments.

Importantly, it also remembers what it has told the user and what it hasn’t because of interruptions. “The second big thing is, once you’ve interrupted, that kind of sets the context for what you haven’t spoken about, and that becomes really important downstream,” Earle added. “So, if the system thinks that it read five things, and you interrupted on number two, it can start to lead to bad things downstream.”

And then there’s the other side of interruptions. In any conversation, there’s give and take. There’s the user interrupting the system, and then there’s the system interrupting the user. For example, Tenyx Voice tunes itself to tell when the user is taking a break to think, pausing to catch their breath, or paging through a calendar to check information before interjecting so that it doesn’t break the user’s train of thought.

At the same time, it’s important for the system to detect contextually when a person is done speaking and ready to move on with the conversation. Tenyx Voice uses a natural conversational tone and turn-taking to understand when it’s time to move on.

Examples could include ordinary pauses understood to be part of the flow as part of its awareness — such as expectations of the information the user is delivering, including the length of phone numbers, brand-specific serial numbers or other conversational cues.

“Awareness of the specific domain that you’re in is necessary to know when you’re done and when you’re not done,” said Earle. “It’s a big part of what we do when building that naturalistic conversation turn-taking.”

Tenyx Voice is already in use by multiple customers of the company including RealtyTrac, a real estate solutions company that hooks investors up with foreclosed properties.

“We are committed to delivering exceptional customer service across our entire portfolio (real estate solutions, credit monitoring, and more),” said David Teng, chief executive of RealtyTrac. “Tenyx empowers us to deliver a unique and differentiated customer experience that represents our brands and helps us deliver on our revenue and efficiency goals.”

Supporting its voice AI system, Tenyx built its own LLM, TenyxChat, that can stand up to some of the most heavyweight models in the market including OpenAI’s ChatGPT. The company also pioneered its own fine-tuning method that helps mitigate “catastrophic forgetting,” a problem for machine learning models where adding new functionality can degrade performance for previously learned abilities.

‍