Bought It AI, a conversational AI agency serving to enterprises, has unveiled Agent Copilot, an AI assistant for customer support and gross sales operations.

Agent Copilot combines generative AI with guard rails to empower brokers to rapidly and precisely entry info from complicated information sources, resulting in enhanced buyer interactions and response occasions, mentioned Peter Relan, chairman of Bought It AI, in an interview with VentureBeat.

This resolution makes use of Bought It AI’s distinctive TruthChecker AI, designed to determine and forestall inaccuracies in responses generated by massive language fashions (LLMs).

Agent Copilot affords companies the flexibleness to customise their AI expertise by deciding on from a variety of LLMs, together with ChatGPT, GPT-4, Llama-2, MosaicML, and Flan-UL2. The selection of LLM will be tailor-made to fulfill the particular information base necessities and measured efficiency towards the related paperwork.

“You’re the pilot and the AI is the copilot,” mentioned Relan. “The massive classes are gross sales and advertising and marketing, banking and customer support.”

Additionally, Agent Copilot will be configured to combine with varied information sources, similar to PDFs, internet pages, paperwork, and shows, enabling complete assist for multi-turn dialogs based mostly on the compiled information base. For organizations with information privateness issues, the answer additionally helps on-premise set up of enterprise-specific fine-tuned LLMs.

“The dangers related to incorrect info are just too excessive,” mentioned Relan. “With Agent Copilot, we provide companies the chance to decide on essentially the most optimum LLM for his or her wants, making certain the very best degree of accuracy and effectivity of their buyer interactions.”

Relan additionally highlighted the affect of Bought It AI’s TruthChecker, which considerably decreases hallucination charges in responses generated by the bottom mannequin, leading to near-human degree accuracy for multi-turn dialogues in information retrieval eventualities.

“The copilot may also help them reply questions, like whether or not you will have a deductible on your insurance coverage or you should use out-of-network docs or these issues,” Relan mentioned. “The agent sometimes has entry to a information database, or set of paperwork that might be internet pages. As a substitute of looking out by way of all that, suppose you’ll be able to floor it in a conversational approach. The copilot’s job is to go get the reply and are available again with it.”

That makes the agent extra productive, he mentioned.

“We’re seeing beta prospects deploy this as a result of the choice is longer calls,” Relan mentioned.

I requested if the pricing was getting decrease or was nonetheless underneath management. Relan mentioned the prices are beginning to transfer decrease for such queries.

And since Relan’s agency was aimed toward stopping generative AI from “hallucinating,” additionally it is conscious of the “guardrails,” Relan mentioned. These must be in place to make the tech clear in order that if it does provide you with a flawed reply, it received’t be like a “black field” the place you’ll be able to’t inform the place it got here from or if it’s flawed.

“We’ve got one thing we name a reality checker, which can truly test the solutions and supply an audit path,” he mentioned.

Agent Copilot is constructed on Bought It AI’s advisable future-proof structure, Enterprise Giant Language Mannequin Structure (ELMAR), and is presently present process an open beta program with a number of enterprises for Agent Help use instances.

The answer is anticipated to be extensively out there in September 2023. Its options embrace information base connectors to customized information sources and paperwork, a alternative of LLMs, TruthChecker AI for hallucination detection, fluid conversational dialogues with mitigation and disambiguation, energetic studying and fine-tuning for optimum accuracy, on-premises deployment with open-source LLMs, and seamless integration with different agent, CX, and CRM options.

Relan mentioned there are massive enterprises, monetary companies companies and insurance coverage companies in beta testing.

“They’re principally saying with out the guardrails, we’d not be contemplating this,” he mentioned.