The Future of Customer Service: LLMs and the Emergence of Specialized Bots


5 min read

The Future of Customer Service: LLMs and the Emergence of Specialized Bots

The landscape of customer service has evolved dramatically over the years. We have witnessed a shift in the way businesses approach customer service, moving from traditional call centres to chatbot technologies. Personally, it feels like this has been a clumsy rise and shift, but we are about to see a move that will truly transform customer service.

I am not sure yet if this is a good or bad thing.

In the early days of online businesses, call centres were the go-to method for providing customer service. This birthed the IVR, and we all had the joy and pain of hold music as well as getting randomly disconnected. This poor customer service was largely because the majority of companies saw call centres as cost centres. Companies like Zappos used this to their advantage and turned their call centres into brand loyalty hubs. There is a story of a customer call that went on for sixteen hours, but this approach to exceptional customer experiences worked, and Zappos was sold to Amazon for $1.2 billion in 2009.

However, most companies treated their call centres as cost centres, often outsourcing queries to other countries where people were following a script of common questions. This approach had its limitations, as it didn't address the real problem – improving the user experience. It was a reaction to the explosion of a new channel known as online. Some businesses started to realise that instead of outsourcing customer service, they should focus on fixing their user interfaces and creating a more seamless customer experience.

Enter the Chatbots

As chatbot technology emerged, businesses began to experiment with incorporating these AI-powered virtual assistants into their customer service strategy. The promise and what was delivered however didn't match, chatbots were FAQs that had been converted into conversational, WhatsApp-style interactions, providing customers with a more engaging and interactive way to get their questions answered but still following a scripted approach.

However, this approach wasn't perfect either, because when customers saw a button that said "chat," they expected to chat with a person, as they typically had an issue and didn't want to be brushed off.

Chatbots weren't great at actually solving problems.

This led to chatbots becoming a triage for FAQs and complaints. Companies used humans for complaints and escalation, while chatbots handled simpler queries (if only businesses had just better presented their faqs). In hindsight, businesses should have had humans front and centre, directing customers to specialised bots that could handle their specific needs more efficiently. The problem is that this approach wasn't cost-effective, so it never had a chance when customer service was viewed as a cost centre.

The LLM Revolution

The introduction of large language models (LLMs) like GPT-4 changed the game again. Now, businesses are starting to put LLMs at the forefront of customer service. LLMs can answer complex questions and direct users to appropriate resources more effectively.

I think we are about to see customer service become human-less. I don't think this is a good idea, but it's going to happen. Let me explain why and how.

LLMs provide businesses with an easy way to offer human-like customer service, arguably 5-10x better than what we have had for the last decade. LLMs already have human-like language of sympathy and kindness and with other advancements in the broader arena of AI, text can now be turned into an actual audio voice or even video, allowing you to call and speak to "someone."

You then have humans supervise/fine-tune the model to ensure that customer interactions stay within the company's brand guidelines.

The problem with LLMs is that their biggest strength is also their biggest weakness: they can be unpredictable with their output. To solve a specific task, you can have a vertically focused bot experience that handles refunds or item changes, for example. This bot has done this process thousands of times before and is continuously perfecting its approach towards the desired outcome of solving the problem at hand.

Now, what happens when there is no bot to handle a problem?

This is where humans come in. Instead of the customer interaction being a live agent or a call to a supervisor, a human will supervise an LLM while it tackles the unknown problem, tweaking responses as needed. These tweaks and supervision will then form the basis of a new, vertically focused bot "exported" from the LLM and continuously fine-tuned over time.

So now we have an LLM model acting as a triage for an unlimited number of bots, with humans fine-tuning the interactions and responses. A call centre of 300 people is now replaced by three domain experts based anywhere in the world. Complaints, issues, or unknown experiences requiring human involvement are significantly reduced, and for the first time, machines may be the better solution (that companies are willing to use) to humans.

Technically, it is a fascinating development, but you have to ask if all customer service becomes non-human, then what do humans do in customer service?

The answer is simple: they make the product better.

They reach out to customers and ask how the product is, without seeking a 5-star review. Human customer service will become outbound and, most importantly, will need to be operated by those who know the product. The question then is which companies will care enough to do this instead of just making customer service humanless and less of a cost.