Sean Cho
January 31, 2023
•
8
min read
There's a lot of buzz surrounding ChatGPT right now. Despite it being a research preview for the public, it's a huge leap forward compared with current chatbot solutions. Its encyclopaedic knowledge and uncanny ability to answer questions like a human are likely to transform the customer experience when it is finally available as an API.
According to Gartner, chatbots will be the main customer care channel by 2027. If ChatGPT is any indicator of how AI, machine learning and language models are evolving, this target seems probable as consumers shift away from email, IVR and phone to save time and get instant answers via natural language interactions.
However, as Forrester discovered in a CINNOX-commissioned survey, some consumers still prefer dealing with a human since communicating with a person offers the emotional connection that self-service touchpoints lack. A chatbot, no matter how intelligent, is still a machine that lacks the subtleties and complexity of the human touch.
Read on to learn more about ChatGPT's impact on the customer experience and how you can leverage the smarts of AI-powered chatbots and provide a personal touch at the same time.
What is ChatGPT?
ChatGPT is a chatbot that uses deep learning techniques to generate natural-sounding human-like text. It does this by training on a massive dataset of text from the Internet, which allows it to understand and generate text on a wide range of topics. It then uses algorithms to predict the next word in a text based on the preceding words.
When ChatGPT is finally available as an API, businesses will be able to leverage the technology behind it in their apps or on CX platforms like CINNOX. A hotel and resort, for example, could use chatbots to take bookings and upsell rooms in a human-like manner. Here's one possible scenario between a guest and the chatbot:
Customer: I want to book a sea-facing room on the highest floor this coming Saturday. For 2 guests and 3 nights, please.
Chatbot: Sure, we have a standard room available at $685 per night.
Chatbot: By the way, we also have one junior suite left for just $99 extra per night. It's 48 sq ft in size compared with 32 sq ft for standard rooms. Would you like to see some photos of it?
Customer: OK. Why not?
What is the difference between ChatGPT and other chatbots?
Chatbots have been around since 2016 and they have had varying levels of success in customer support and service. So, what's the difference between ChatGPT and other chatbots? While both are similar in that they process and generate natural language text, there are four main differences between ChatGPT and today's chatbots:
- Training data. ChatGPT is trained on a large dataset of text from the Internet, which allows it to understand a wide range of topics. Chatbots, on the other hand, are limited to specific data and rules, and hence a narrow range of topics.
- Flexibility. ChatGPT can perform more tasks—such as language translation, summarisation and question answering—which allows for more flexibility in its use. Chatbots, however, are typically designed to perform specific tasks only.
- Natural language. ChatGPT uses deep learning techniques to understand and generate human-like text. Other chatbots usually use rule-based or pattern-matching techniques, which may not be as accurate or natural-sounding.
- Interaction design. ChatGPT is a language model, and its main purpose is to generate text from a command prompt. Other chatbots are designed to interact with users, using decision rules to guide users throughout the conversation.
As a result, ChatGPT is a more powerful and flexible tool than other chatbots. Normal chatbots, on the other hand, are more focused on specific tasks and user interactions like capturing and qualifying leads. Despite ChatGPT's vast knowledge, other chatbots may be better at walking a prospect through an insurance policy quotation, for example.
ChatGPT and the Customer Experience
I was curious to know what ChatGPT thought of customer experience in general and how it can improve it further. So, I typed: "What can ChatGPT do for the customer experience?" and it replied with the following:
ChatGPT can be used to enhance the customer experience in several ways, including:
Chatbots and Virtual Assistants: ChatGPT can be used to create chatbots and virtual assistants that can understand and respond to customer inquiries naturally and conversationally. This can help improve the efficiency and effectiveness of customer service.
If you had a chance to try ChatGPT, you could be forgiven that you're interacting with a live agent. After typing your question, it pauses like any human would and then carefully parses its answer word by word as if it was in deep thought and typing it out. Contrast this with a typical chatbot that displays pre-defined answers and options in an instant.
Text generation for customer support: ChatGPT can generate personalized responses, such as email or chat replies, to customer inquiries, which can help speed up the resolution of customer issues.
When used as an API in a CX platform like CINNOX, ChatGPT could train on specific datasets from a company's customer support, sales and service database. It could then make sense of all the emails, names, addresses, orders, audio transcripts, SMS and much more, to tie it all together and respond to an enquiry faster than any live agent.
Text summarization: ChatGPT can be used to summarize customer feedback or support tickets, which can help businesses quickly identify and address common issues.
Good customer experience means quickly getting to the core of a problem. What are customers enquiring about, and what solution do they seek? ChatGPT could label chat enquiries in real-time, indicating whether it's a complaint, compliment or service request, for example. Enquiries could then be routed to the best agent for a positive experience.
Language Translation: ChatGPT can be fine-tuned on a specific task of language translation, which can help customer service representatives to communicate with customers who speak different languages.
ChatGPT is multilingual. It can understand questions and generate replies in almost 100 languages! Type a question in French, Spanish or Chinese, for example, and it will reply in the same language. What this means is that just about every customer on Earth can be understood without needing a live agent who can read and write their language.
Sentiment analysis: ChatGPT can be fine-tuned on a specific task of sentiment analysis, which can help businesses understand customer opinions and emotions, which can help improve the customer experience.
Like today's chatbots, ChatGPT can be trained to recognise human sentiment. The use of caps lock, punctuation, happy or angry emojis as well as complimentary or offensive words can indicate what the customer is truly feeling in the chat. This feature, however, should be used together with a live agent which we will explore in the next section.
Limitations of ChatGPT for the Customer Experience
By now, you might think that implementing a powerful AI solution like ChatGPT would automatically mean a better customer experience. Not necessarily! While AI-powered chatbots have the technological advantage, it is the human element in customer service that ultimately makes a relationship successful—and creates happier customers.
Let's look at its responses when I asked, "What are the limitations of ChatGPT for the customer experience?". Plus, we'll also evaluate measures that could be taken to soften its sometimes emotionless approach, and make the customer experience more human.
ChatGPT, like any other language model, has some disadvantages, including:
Lack of common sense: ChatGPT is based solely on patterns it has learned from its training data and does not possess common sense or the ability to reason like a human. This can lead to nonsensical or irrelevant responses to certain inputs.
ChatGPT can give very convincing, grammatically correct and well-structured answers. However, as acknowledged in its FAQs, it may also generate incorrect or harmful replies because its dataset is not connected to the Internet and is limited to content produced up to 2021. Plus, it may also make up facts or “hallucinate” answers at times!
Solution. In situations like these, it is best that the customer has an option to escalate to a live agent should answers go astray. In addition, agents should also monitor AI-powered chats for inappropriate, incorrect or harmful responses—and quickly take over the chat should the customer experience be in danger of turning negative or unpleasant.
Bias: ChatGPT has been trained on a large dataset of text from the internet, which can introduce bias into its responses. This bias can be based on factors such as gender, race, and socioeconomic status.
As an AI solution that uses machine learning algorithms, the quality of ChatGPT's output is only as good as its data input. Its current dataset is 570GB in size or 300 billion words. Future versions of ChatGPT are likely to improve when it trains on terabytes of data. For now, it may make assumptions when interacting with customers.
Solution. To reduce bias when chatting with customers—assuming that they speak English for example—agents can see the country location, browsing language and browsing history in CINNOX to personalise the chat further. Using these insights, agents can anticipate enquiries knowing what language they speak and what they have been viewing.
Lack of context awareness: ChatGPT does not have the ability to keep track of the context of the conversation, leading to inconsistencies or irrelevant responses.
In the current version of ChatGPT, it can remember what the customer said earlier in the conversation, but there is a limit to how much information it can retain. For now, its memory is limited to 3,000 words and it cannot reference past conversations. This may produce disconnected chat experiences where previous mentions are not referred to.
Solution. If an AI-assisted chat escalates to a live agent, he or she can check the chat history in CINNOX to understand the context of the interaction and refer to past chats. If customers want to speak with the same agent for continuity and familiarity, they can use CINNOX's web link feature to call or chat with them again and build rapport.
Inability to understand idiomatic expressions and sarcasm: GPT-based models have a hard time understanding idiomatic expressions and sarcasm, and can produce inappropriate or confusing responses in those cases.
Human languages have their double meanings, puns and humour. Currently, AI cannot discern this play on words from the ordinary use of language. For example, if a customer types "I'm on the fence about that offer" to mean that they are undecided, AI may reply with, "I'm sorry to hear that. I hope you get off the fence safely."
Solution. To avoid confusing customers or making them feel ridiculed or frustrated, live agents should take over the chat if they come across conversations that are being misinterpreted by AI bots. For clearer communication, customers can also use the CINNOX widget to switch between chat, video and voice calls to communicate their point across.
AI and human agents still need to work together
To finish on a lighter note, I asked: "ChatGPT, do you often get compliments or brickbats?"
As a language model, I am not capable of receiving or experiencing emotions, so I do not receive compliments or brickbats. My purpose is to assist users by providing accurate and relevant information or completing language-based tasks to the best of my abilities based on the input provided to me.
At least ChatGPT is honest about what it can and cannot do. For people that desire a human touch in the customer experience, it may be difficult to build rapport or connect with an soulless chatbot. For now, live agents are still needed to deliver that special touch missing in AI-assisted chats and build successful long-term relationships.
