ChatGPT - Are the customer service bots finally useful?

LinkedIn and all the business and customer service journals are full of stories about the generative artificial intelligence (AI) chatbot ChatGPT. This bot was only launched in November 2022, but within a week it had a million users and soon notched up 100 million regular users.

The initial message was focused on the ability of ChatGPT to create comprehensive answers to questions using natural language. There is no need to speak as if you are addressing a search engine - just talk to the bot and it responds.

The Guardian newspaper in the UK specifically said that there would no longer be a need for any customer service agents if customers can engage with ChatGPT. That sounded ominous, but of course it was really just an attention-grabbing headline.

Since all these early comments about the end of different types of employment there has been something of a reset. People have started realising that ChatGPT is very smart, but it is just a tool that can be one a part of a business solution.

As an experiment, I asked ChatGPT 4 how generative AI can change and update customer service processes. The first time I asked, it created an entire essay so this is the answer when I asked it to answer in a single paragraph sounding like a McKinsey consultant:

“Generative AI has the potential to revolutionize customer service by streamlining processes, enhancing efficiency, and reducing operational costs. AI-powered chatbots offer instant, personalized, and round-the-clock support, effectively managing routine tasks and enabling human agents to focus on complex issues. With continuous learning capabilities, multilingual support, and sentiment analysis, these AI solutions foster improved customer experiences and business outcomes. As generative AI continues to evolve, it is vital to strike an optimal balance between automation and human empathy, maximizing the synergy between technology and human agents to achieve exceptional customer service.”

This is a good answer. It is filled with business buzzwords as well so it took the prompt of sounding like a consultant seriously. This is also a good example of something that ChatGPT does very well. It can take a question, understand it, and search for existing information before synthesising it together into an answer.

This is the generative part when we talk of generative AI. Think how this would work if you asked Google the same question. I tried it and Google listed over 18 million pages of information, including a LinkedIn blog and a CNBC news story. ChatGPT doesn’t point you at links to the information, it generates a bespoke answer to your question based on what is in the library.

So there are some very interesting and immediate applications in the customer service environment:

• Instant response: reducing customer wait time by responding to questions immediately with what should be a credible answer.
• 24/7 support: being there to take questions at any time and not just answering with a link to a FAQ page, but with more helpful advice.
• Basic troubleshooting: being that initial chatbot you get to talk to before talking to a human, but actually being more useful.

The problem with generative systems like ChatGPT is that they usually don’t know when they don’t know the answer. Unlike a more traditional bot, that would just fail to answer, ChatGPT can draw on the library of information to create an answer, but it may not be entirely correct.

As an example, I asked ChatGPT to explain to me how robots were performing customer service functions in 1962 and what has changed. The system responded by saying: “Since 1962, robots have come a long way in terms of their application in customer service processes. Here is a brief overview of the key milestones in their development and deployment…” followed by a history of robotics since the 1960s.

So it is very smart, and very useful, but it requires the answers to be in the training data - it’s not good at new problems that are unknown.

For this reason, it can never replace the empathy of a human customer service experience - nobody wants to call their life insurance company about a family death and talk to a bot. It also may struggle to replace problem-solving activities where a customer has a complex problem that needs investigation.

However, it does look like we can finally get a much better FAQ experience through the use of these generative tools. The podcasts and journalists that said human customer service is finished spoke too soon, but this is an important development - the bots are finally becoming useful and that is exciting.