Large language model-driven artificial intelligence (AI) chatbots have gained interest from business leaders across a range of industries in recent weeks. One such chatbot, called ChatGPT, generated particularly remarkable headlines in the tech industry by attracting more than 1 million users in the first week of its release.
According to Juniper Research, ChatGPT and other turbo-charged models and bots will become increasingly important in consumer interactions over the next few years. By the end of 2023, up to 70% of customer conversations will be handled by AI-powered chatbots, according to a recent research by the analytics firm.
The increasing reliance on AI to improve customer experience (CX) and expedite interactions is highlighted by this. Businesses have a lot of potential to leverage chatbot technology to enhance marketing campaigns, provide individualised services, and overall increase efficiencies as chatbots become more human-like in their dialogues.
Although speech recognition and natural language processing (NLP) have a lengthy history in customer management and call centre automation, experts in the industry believe that the new large language model (LLM)-driven chatbots could fundamentally alter the future of CX.
LLMs, according to Sean Mullaney, CTO of search engine SaaS platform Algolia, “are fundamentally transforming the way search algorithms work.” Unlike LLMs, which are effective at understanding word meanings and can retrieve more pertinent stuff, traditional search engines do not match individual words from a query with the terms in a huge index of content.
LLM-based chatbots and virtual assistants have made it possible for customers to communicate with organisations in a more conversational and natural way. This has improved the CX for customers throughout their whole customer experience. As a result, LLMs are now a go-to option for businesses trying to improve their efforts in customer service, sales, and marketing.
However, putting the new bots into use won’t be easy. First-generation chatbots have already demonstrated that success is not a guarantee.
Despite their adaptability, many first-generation chatbots find it difficult to comprehend complicated requests or queries and have trouble keeping context throughout an engagement. Because chatbots are frequently confined to a small number of conversations, this has occasionally led to a stiff or rigid customer experience. Interactions frequently end up going through a person.
According to a recent survey by AI company Conversica, users’ experiences with first-generation chatbots fall short of what they had hoped for. According to the company, if the answers don’t specifically address their needs, four out of five customers give up on the chat experience.
According to Jim Kaskade, CEO of Conversica, “first-generation chatbots rely on predetermined scripts that are laborious to programme and extremely harder to maintain.” They also restrict users to prewritten message responses and can’t grasp simple questions. Applications with LLMs that are enterprise-ready and AI-equipped, like GPT, can have an impact, he said.
ChatGPT transforms the field of conversational AI.
LLMs influenced by ChatGPT can help organisations display their material to clients in a more engaging way by adding various conversational styles and content tones. The effectiveness of LLMs’ responses and overall CX can be continuously improved by learning from and adapting to customer encounters.
According to Dan O’Connell, chief strategy officer at AI-powered customer intelligence platform Dialpad, LLM-based chatbots like ChatGPT can be used by agents as editing and suggestion tools to improve their interactions with consumers. O’Connell told VentureBeat that they “can be used in a variety of ways to save time and append information, but also to efficiently identify subjects, action items, and map sentiment.”
“The issue with models like ChatGPT is that it’memorized’ all it could find online into only 175 billion numbers (5,000 times fewer than the human brain). Consequently, ChatGPT can never be completely certain of the solutions it provides, according to Pieter Buteneers, director of cloud communications system Sinch Labs. “Remembering every tiny detail is impossible, especially if we’re talking about keeping all the information online. As a result, it will always say whatever is the first thing that comes to mind.
Despite its shortcomings, the newcomer ChatGPT outperforms existing chatbots in one key area: it is excellent at determining user intent, preserving context, and staying highly engaging throughout the discussion. Additionally, ChatGPT’s potential for NLP and its capability to quickly react to inquiries have caused businesses to reconsider their present chatbot architectures designed to improve customer experience. I’m ChatGPT, hello. Any questions?
While the GPT-3’s NLP architecture generates an output that seems like it “understands” the query, content, and context, traditional chatbots enable interaction in an apparently intelligent conversational style. The current iteration of ChatGPT does, however, have several disadvantages, such as the potential for producing inaccurate information and even politically wrong comments. Even the OpenAI team has cautioned against using ChatGPT to ask for factual information.
The secret to creating LLMs with remarkable skills, according to Jonathan Rosenberg, CTO and head of AI at contact centre platform supplier Five9, is to use AI methods like zero-shot learning, as ChatGPT did. When an input that wasn’t addressed during machine training is presented to a machine learning model, the process is known as “zero-shot learning.”
According to Rosenberg, what distinguishes GPT-3 from its predecessors is that it has grown large enough to produce meaningful answers to any query without having been specifically trained on it. The design of GPT-3 isn’t drastically different from those of its forerunners, though. Instead, zero-shot learning wasn’t effective until the model size above a certain threshold, after which it simply began to function considerably better.
Kurt Muehmel, everyday AI strategic advisor at AI-powered analytics platform Dataiku, stated that models like ChatGPT “won’t be able to replace what organisations perform within the contact centre with typical conversational AI.” “Companies that deploy them need to develop processes to ensure that there is a constant human expert assessment of the responses and to correctly test and maintain the systems to ensure that their performance does not degrade over time,” the report states.
However, organisations must see chatbots and LLMs like GPT as useful instruments for carrying out particular tasks rather than just as trendy novelties. To optimise their influence, organisations must create and implement use cases that provide the business with real benefits. Thus, these AI technologies can significantly contribute to operational efficiency and business success.
“ChatGPT offers opportunities since it can recognise more emotional nuances in text because of its advanced emotional understanding capabilities. Because the human element still needs to play a crucial part, what businesses are currently doing in the contact centre won’t be completely replaced by this, according to Glassbox CTO Yaron Gueta. Companies will be able to have much less call diversion between the chat channel and call centre, as ChatGPT can improve end-user experience during chat conversations, which is where it will have the biggest benefit.
Similarly, Yori Lavi, a cloud expert at data analytics platform Sqream, advises that it is crucial to keep in mind that constant monitoring, testing, and training are essential. He stressed that it’s crucial to frequently inform models like GPT of the value and danger associated with their conclusions.
“Chatbots should never make high-risk decisions without first being validated or evaluated. As a result, organisations should work on developing chatbots that can address complicated needs and draw on prior inquiries and contextual information to better customer experience, according to Lavi.
Using cutting-edge LLMs for improved CX
According to Deanna Ballew, SVP of product for Acquia’s digital experience platform, advanced LLMs like ChatGPT will be used as a dataset and a conversational AI capacity, while other technologies will improve ChatGPT for training.
“In 2023, there will be a lot of experimentation and the introduction of new products to enhance ChatGPT’s commercial worth. This will also affect how customer service representatives respond to customers, whether they do so with chatbots or by swiftly applying ChatGPT on their own data, according to Ballew.
The growing usage of these models in customer care and support, according to Danielle Dafni, CEO of generative AI firm Peech, implies businesses will need to continue investing in creating more complex chatbots, resulting in greater CX. But there is a benefit.
Companies who use these models to enhance their chatbots’ capacity for detecting and responding to interpersonal emotions as well as other capabilities will be well-positioned to offer better customer service and an enhanced user experience, according to Dafni, who spoke with VentureBeat.
“Bots like ChatGPT and conventional LLM chatbots will keep developing and getting more smart at comprehending and reacting to customer interactions. With greater public awareness, more users would anticipate GPT-level conversational functionality from chat features, displacing first-generation scripted bots, according to Conversica’s Kaskade.
The adoption of web chat solutions with generative AI capabilities, according to him, is just at the cusp of adoption given the present breakthroughs. In the next three years, he envisions these becoming commonplace in both B2B and B2C transactions.
