The Gist

  • Generative AI, meet CX. In a CMSWire exclusive, we learned about the release of Level AI's AgentGPT, a generative AI product aimed at improving the customer experience and assisting contact center teams.
  • Not like ChatGPT? Unlike ChatGPT, AgentGPT is trained on a client's proprietary customer conversational data and internal data stores, allowing it to analyze queries with multiple attributes and recommend solutions accordingly.
  • Contact Center thumbs up, thumbs down. AgentGPT is built on Level AI's existing Natural Language Understanding (NLU) product suite for contact centers, and can improve its responses through agent feedback using a "thumbs up, thumbs down" system.

With the goal of elevating customer experience and assisting contact center teams, Level AI announced today its newest generative AI product, AgentGPT — calling it a “highly trained version of ChatGPT.”

Unlike ChatGPT, the wildly-popular OpenAI generative chatbot that debuted last November but does not have access to proprietary data, AgentGPT is trained on a client’s proprietary customer conversational data and can apply a company’s internal data stores to assist customer service teams.

According to Ashish Nagar, CEO and founder of Level AI, it’s built on Level AI’s existing Natural Language Understanding (NLU) product suite for contact centers.

Like ChatGPT, it can respond to customer queries. But AgentGPT, which self-learns from conversations and improves based on agent feedback, also claims to understand the nuances of human language.

Alpha Release Gleans Encouraging Reviews

Level AI released an alpha version on Jan. 17. Nagar told CMSWire customers have tried queries their agents face on a day-to-day basis, particularly the ones that are "complex and could have multiple nuances," he added. "AgentGPT also displays all relevant queries for the customer input entered; this way agents do not have to limit their thinking to understanding the problem in a single dimension."

With the public release of AgentGPT, Nagar said he hopes it will be used to reduce average handle time (AHT), lessen onboarding time of agents and improve overall customer satisfaction in a conversation.

Unpacking the Complexity: How Generative Technology Deciphers Nuance

With access to the unique proprietary information of each organization where it's deployed, AgentGPT analyzes queries with multiple attributes and recommends a solution for each, whether it's a query related to a company payroll system, or a check-capturing system used by a restaurant. 

Say, for example, a query is submitted that said, “my restaurant logo is not clear on gift cards.” There could be multiple ways to interpret “logo is not clear.” It could mean logo’s image is blurred or cut off from the gift card, or that the logo has text over it. The customer could also be referring to the issue of an incorrect logo.

Nagar said that AgentGPT is able to understand the context behind the words spoken and provide a solution without the need to clarify what they actually mean when they speak in sentences with multiple interpretations. In the case with the query, “my restaurant logo is not clear on gift cards,” AgentGPT is able to recommend a solution with steps including, “explain to the customer that the logo needs to be a perfect square with a minimum size of 180x180.”

Screenshot of what one would see using Level AI's AgentGPT for customer service questions.

Learning Opportunities

Self-Learning AI: Thumbs Up, Thumbs Down

According to Level AI officials, AgentGPT improves through agent feedback, with a built-in system where agents can give a “thumbs up” when they find a solution relevant to the query inputted and “thumbs down” when they do not find it relevant. These serve as indications to the model on what has worked well and what hasn't. It then takes this feedback into consideration and automatically takes action based on that feedback.

"What is exciting about our AgentGPT is it does not only change the ranking of that particular query for which feedback is provided, but improves the ranking of related queries or customer concerns too," Nagar said. 

Related Article: CMSWire's Top 10 Customer Service/Call Center Articles of 2022

AI Not Aimed at Taking Human Jobs

While AgentGPT is not meant to eliminate human jobs, Nagar said it can solve the problem of extra staffing by augmenting (or enhancing) an agent during a customer call or chat, helping to reduce the time spent per conversation. 

"The agent does not have to go through multitudes of knowledge articles to find the solution or spend time contacting their managers or senior agents when they are stuck on a customer issue, or not clear on what solution they should provide to customers," Nagar said. "If agents spend less time per conversation, they can take more conversations in a day. This could solve the problem of extra staffing. Moreover, we have seen high attrition rates in contact centers. AgentGPT could reduce onboarding time for new agents and groom them quickly to take customer calls like an expert, further solving the problem of extra staffing."

The Rise of Generative AI: A Growing Market Fueled by CX Vendors

Level AI is not alone in making advancements with generative AI as it seems everyone and anyone is trying to ride the hot chatbot wave.

Some vendors joining the party:

Customer experience professionals themselves are exploring new ways to integrate generative AI. All the while, the major tech giants are battling it out to take the lead, while ensuring (trying to ensure) responsible use of this growing tech.

fa-regular fa-lightbulb Have a tip to share with our editorial team? Drop us a line: