It’s hardly surprising that people wonder what jobs ChatGPT and technologies like it might replace. Customer service agents jump to mind. After all, if it can convincingly tackle an MBA exam question, why shouldn’t ChatGPT be able to deal with your average customer query?
It’s not that simple though. As exciting a leap forward in AI technology as ChatGPT is, I don’t think it is the silver bullet to automating customer service.
Most chatbots act a lot like Google search - they predict possible answers from a corpus of information available to them. Sometimes the answer is clear and they can answer immediately. Other times there are many possible answers, and so they offer a number of options e.g. links to different documents or web pages. Based on your selection, they learn which was ‘right’ and this changes how they rank these answers in future. This process can be frustrating to users because you are teaching the chatbot, when it should really be a service expert already.
ChatGPT, by contrast, looks across the broadest possible selection of content on a topic and offers a single consolidated view, written in a way that feels as if an expert has authored the ‘single’ response or answer. It is also being taught, but its ability to give contextual, expert-level answers will increasingly become what customers expect from chatbots. As a result, customer expectations will rise and so will frustration levels with most current self-service chatbots.
ChatGPT is not simply a smart librarian. It is able to perform like an expert, offering you a very specific, coherent and mostly complete response to almost any request or question you can think of. And while it does make mistakes, it gets an awful lot right.
ChatGPT already knows a lot about your company’s specific products and services, based on information on your website and in social media. Anything that is in the public domain can form part of ChatGPTs knowledge range.
So yes, for answering questions where the information lies in the public domain, ChatGPT is a very powerful alternative to Google and company chatbots. However, most companies do not host the information needed to resolve service queries or disputes in the public domain. Much of that content sits within different internal systems, or even inside different experts’ brains. So, while ChatGPT can be deployed behind a company’s firewall, getting it trained to the levels required will be no small feat.
Also, many customer service queries and disputes don’t have a single answer that can be documented and made available to ChatGPT. Individual answers usually depend on the customer’s situation. To get to the right outcome, a number of questions need to be asked to gather missing data, and then different rules need to be applied before a decision or action can be reached. Plus some of this data is sensitive.
To get this right, human agents currently follow very specific processes, and their conversations are heavily constrained by company policies and rules. Each answer the customer or system gives shapes where the conversation goes next, and everything has to be done in a consistent, compliant way. Even letters and emails are prescribed in templates, so human agents have little creative licence to write what they like.
To automate this logic, you need a digital brain that will have the right conversations with your customers, based on your very specific rules, guidelines and data requirements. This digital brain needs to ensure accuracy and consistency, but also adjust to each customer’s context. And it should also advise each customer based on their varying situations, needs and problems. ChatGPT can’t do that.
ChatGPT is very efficient at answering most information-based questions. But when data is sensitive, accuracy is critical, and answers and actions are heavily shaped by rules, other technologies may be better suited.
With the recent investment by Microsoft into OpenAI, the owner of ChatGPT, it will be blended into most of what you do on a daily basis. It will help you write documents and letters, and solve your complex spreadsheet formula problems. And it will be a powerful first source for customers to turn to when they have an informational query about your company
Resolving specific queries and disputes where context, rules and accuracy matters, will continue to be managed by other, better suited, technologies - including human agents and virtual agent (VA) technologies designed specifically to handle these engagements. Virtual agents’ logic is shaped by rules, yet they can have personalised conversations that feel authentic. The difference between ChatGPT and virtual agents is that the process VAs follow and the outcomes they reach are assured, and in line with compliance rules and back-office system requirements. In short: Virtual agents are under your control; ChatGPT isn’t.
Technologies like ChatGPT will certainly impact knowledge workers as a category, but other smart automation technologies will have a greater real-world impact on customer service, despite the current hype.