The hype surrounding ChatGPT has increased interest in chatbots, virtual agents and large language models, but what are they and what should they be used for? This article aims to describe the differences and confirm which are best suited for customer service environments.
Chatbots
A chatbot is a computer program that is designed to simulate conversation with human users. It is typically used to provide basic automated customer service or to answer simple questions based on series of FAQ-type responses. Chatbots do a great job at answering questions, directing web site visitors and booking appointments. The investment required for a chatbot is generally low if the operator is prepared to compile the predicted questions and respective answers. This is done on a simple web interface where software coding is not required. Chatbots often lack a good AI backend which limits their understanding of your requirement or intent, this forces the chatbot producers to steer the user through a series of buttons and suggestions rather like a visual phone IVR. Chatbots are cheap and quick to produce but will eventually seem limited by both operators and users in a customer service environment where lots of organisations use virtual agents instead. The lack of AI and opportunity to develop further causes the limitation. Chatbots are widely used for web site marketing purposes as an alternative to web forms because the engagement rates are improved.
Virtual agents
Virtual agents are more sophisticated than chatbots and are able to store data throughout a conversation and over several conversations. They can understand more complex conversations and provide more detailed responses. Additionally, virtual agents are often able to learn from their conversations and improve over time with supervised human input. Virtual agents use AI to predict the intent of the user and provide an accurate response, if one has been loaded, to the user. Their ability to quickly ascertain the best possible response or fallback answer is its major differentiator when compared with a simple chatbot, it appears smoother and more human-like even with a small data set. The other differentiator is its ability to connect with any external system via a web API rather than just a limited list of pre-configured applications. Google Dialogflow has "fulfilment" which is based on node.js and Amazon Lex uses LambDa both of which offer limitless integration. Virtual agents offer a wide variety of skills such as retrieving database information, taking credit card payments, upgrading accounts and integrating with legacy back end systems. One of the draw-backs of virtual agent platforms is they tend to have very basic, technical chat interfaces or widgets. Another is their inability to integrate easily with a human chat team without some kind of coding. In summary the virtual agent platform is the most capable for customer service, where the options are unlimited but they do require software development.
Hybrids
Virtual agents (eg Google Dialogflow) are behind many of the top chatbot platforms around today which provide a good web interface to get you started with chatbot tasks but they can also be taken further by bringing in the virtual agent behind the scenes. So in theory this gives you the best of both worlds, a code-free UI with the power of a market leading platform for a relatively low price ticket. Hybrids provide ready-made integrations with many of the top platforms and some differentiate based on the request routing strategy; bot only, bot first then fallback to human or human first then bot. The widget, which is the iPhone sized chat interface which you interact with, can be customised easily and virtual agent features can be passed through to it, such as image display, carousels and gifs. Live chat integration is better with hybrids and in some cases such as with Kommunicate, there is an excellent handoff to human process which invokes when the bot hits a fallback routine. The handoff is barely noticeable by the user and agents are notified instantly and can resume the chat session seamlessly even if they are not staring at the web interface.
Large Language Models (LLMS)
There is a lot of hype in 2023 around LLMs, the most well known currently is ChatGPT. Language modelling is the use of various statistical and probabilistic techniques to determine the probability of a given sequence of words occurring in a sentence. Language models analyse bodies of text data to provide a basis for their word predictions. ChatGPT is a large language model developed by OpenAI. It is designed to generate human-like text based on the input it receives, allowing it to engage in natural-sounding conversations and provide responses that are coherent and relevant to the topic at hand. It is able to process and understand large amounts of text, and can use this knowledge to provide accurate and informative responses to a wide range of questions. Due to the sheer size and complexity of the model behind ChatGPT it cannot be trained for your structured customer service requests and is not 'connected' with the open internet. Furthermore it was trained on data in 2021 so will not understand anything more recent. This makes it unsuitable for use in most customer service environments. However, ChatGPT is a very powerful model which can be used a variety of other environments such as content production, software coding, summarising text and other language-related tasks.
Language chains
LLMs have no capacity for custom dataset development, they have no session memory and cannot work as a 'team' with other LLMs. As previously outlined this makes them unsuitable for customer service currently. Language chains broaden the scope of LLMs somewhat by providing a single chat interface which can handle a variety of LLMs, data sources and computation systems. It can decide which is the most suitable response and respond accordingly. Memory is the concept of persisting state between user requests. Chains can handle memory . ChatGPT, as a comparison has no memory and may provide different answers to the same questions asked repeatedly even over short spaces of time. Whilst chains take us closer to a new super agent, the value lies in its ability to decide on the best response to offer you. It may be difficult for it to decide whether to use your preferred response to a customer over a generic one that is actually irrelevant.
Which is best for customer service?
The best virtual agents are able to complete complex tasks faster than a person would and the more complex the task the more useful the virtual agent will seem to your customers and prospects. Even if the virtual agent is able to focus on a few useful tasks it will be valuable to your customers. Creating even a simple chatbot to work on FAQ questions requires a great deal of time and thought and moving from platform to platform is not easy. So it is important to start with the right platform or risk having to start again from scratch. Hybrids in our opinion meet the requirements for all new, and experienced chatbot operators. There are numerous services available so make sure hybrid has a good backend virtual agent such as Google Dialogflow, Amazon Lex or Microsoft. In summary hybrids can operate effectively in customer service for the following reasons:
The benefits of Hybrids in customer service
- You can easily change the 'front end' whilst retaining the same virtual agent platform. This means if your current virtual agent web application for eg Dialogflow is not very good you can replace it without having to change the back end virtual agent.
- They can be integrated with any unlimited 3rd party system.
- You can use a "crawl, walk, run" approach, starting slowly and developing without having to jump ship.
- Hybrids can handle bots and humans on the same team.
- You can create a very good user experience using virtual agents.
- Security is more of a priority with the bigger tech brands like Amazon, Google & Microsoft.
- The investment in AI by the bigger tech brands will be hard for smaller independents to keep up with.
- This market is very active with lots of consolidation and so it is a good idea to invest your time and money into a platform that will be stable for the next few years.
Hostcomm will be happy to advise and implement a hybrid, virtual agent or chatbot service for you so please get in touch.
