Artificial Intelligence (AI) is revolutionizing how businesses operate. For companies, the path is full of potential yet accompanied by complexities. In this article, we would like to guide you through this intricate AI landscape, by unraveling two key AI technologies, the established voicebot and the rising LLM agents.
When you receive a phone call reminding you about an upcoming appointment or pending repayment, you could already be talking to a voicebot. Voicebots utilize Automatic Speech Recognition (ASR), Natural Language Processing (NLP) and Text to Speech (TTS) technologies. They can really understand how people talk and stimulate human-like interactions based on customer reactions.
A common use case for voicebots is 24/7 customer service. They can handle common inquiries from inbound calls at any time, so human agents can focus on more complex issues.
Voicebot also excels at high volume outbound calls for reminder, promotion and leads filtering. They are effective to help businesses to boost their customer and revenue growth. Voicebot can make customer outreach calls on a massive scale to help remind all relevant customers, or quickly filter and qualify prospects in the entire CRM contact list. With voicebots handling initial outreach, human agents can follow up on qualified, sales-ready leads.
As we introduced in our previous insight article, an LLM agent is a combination of a large language model with three other key components for planning, memory and tool use. We can also refer to an LLM agent as a LLM powered autonomous agent system, in which the LLM functions as the brain.
LLMs have unveiled their powerful Natural Language Understanding (NLU) capabilities. But LLMs alone are like brains in vats, and they can not realize a wide array of real-world applications. To unlock the full potential of LLMs, we need to build systems that can acquire and apply knowledge to solve practical problems. That’s where LLM powered autonomous agent system or LLM agents comes in. LLM agents can perceive, reason and act like our copilot, beyond just conversation capabilities. In theory, when giving LLM agents any general objective and a robust enough agent system, they can plan for task completion step by step. They are also capable of calling available agents or tools for help when necessary.
Voicebots adhere to rigid pre-determined dialogue flows. They only understand the expressions and words they are trained for. For instance, a seemingly simple use case like transferring a hospital patient’s call from the main line to different department extensions could require a matrix of dialogue flow. This could take weeks for conversation designers to develop.
In contrast, LLM agents possess a new edge in their potent Natural Language Understanding (NLU) capabilities. They can understand diverse human expressions, comprehend in context during real-time conversations. If your query includes something LLM agents don’t understand, they can continue the dialogue until they achieved clarity. This closely resembles how a human would ask follow-up questions to understand better.
Furthermore, while voicebots can perform what they are trained for efficiently, they lack autonomous planning or decision-making capabilities. Nor can they adapt to dynamic external environments.
In comparison, LLM agents are capable of self-learning new knowledge across languages and knowledge domains. They possess both short-term memory from in-context learning, as well as long-term memory from search and retrieval. They also can perform reasoning and planning, breaking down large and complex tasks into smaller, manageable steps.
Let’s consider a customer refund request scenario to understand how a voicebot and an LLM agent would act differently.
A voicebot can understand all expressions and keywords meaning refund if trained for them. It can read out the refund process from a FAQ list when a refund is requested. However, a voicebot won’t be able to handle queries beyond the FAQ list.
On the other hand, an LLM agent can call different tools and leverage industry best practices to resolve the issues. Moreover, it is capable of continuous learning and optimization, acquiring new knowledge and best practices overtime. When a client requests a refund, a LLM agent can ask for the purchase ID, then call relevant agents or tools (i.e. logistics tracking sheet) to check delivery status. They can even soothe the client, suggesting waiting for a couple of days before processing the refund. Meanwhile, LLM agents would analyze order time, and assess whether the no-fault refund period has expired, or check the customer’s credibility status to decide if a refund is eligible.
Looking ahead, we expect LLM agents to go far beyond inbound and outbound customer engagement. They will fully integrate with the enterprise software system, and become the core command center of the entire system. By then, LLMs will be industry experts and decision makers. They would understand domain-specific enterprise knowledge and dynamically call different tools to automate task completions. The agents or tools being called by LLMs can vary, including common office systems such as CRM, ERP, OA and PMS etc. We envision LLM agents to be versatile copilots, managing workflows alongside employees and assisting customers.
Despite the promising future of LLM powered autonomous agent systems or LLM agents, their development is still in early stages. LLM agents can alleviate the burden of training and setting up intricate dialogue flows. They have the capability to learn and optimize with minimal human intervention. However, the training of such LLM agents demands significant computational resources and GPU capacity. Therefore, LLM agents might not always be the most cost-effective solution for your business.
Voicebots, conversely, serve as robust business copilots, and are capable of driving impressive results. They enhance customer satisfaction by providing round-the-clock services, augment payment rates through automated reminder calls at scale, and efficiently qualify leads for human agents to follow up, among other tasks.
Consequently, we recommend businesses leverage voicebots for relatively straightforward applications like sending reminder calls, executing preliminary lead filtering, or addressing routine inquiries. Businesses should consider tapping into the power of LLM agents when they necessitate complex interactions, dynamic reasoning, industry-specific knowledge, or integration with a variety of enterprise systems.
The world of artificial intelligence offers a broad spectrum of solutions for businesses. Whether it’s a voicebot or an LLM agent, the choice depends on your specific needs and resources. The key to success lies in understanding these AI systems, and choosing the right AI assistant to navigate your business towards a prosperous future.
Zhiheng Xi et al. (2023). The Rise and Potential of Large Language Model Based Agents: A Survey. Retrieved from: https://github.com/WooooDyy/LLM-Agent-Paper-List