Smart virtual assistants: how AI can support customer service

With the launch of large language model applications like ChatGPT in November 2022 and Google Bard in August 2023, the use of generative AI for business purposes is on the rise. So far, content creation for marketing has been a main area of AI activities within companies. But is this all that generative AI is useful for? In the following article, we will explore another promising field for large language models: customer service. We’ll take a closer look at what AI-powered chatbots can do – and discover the limits of AI in direct interaction.

When looking at business predictions for AI applications, it seems like the sky is the limit. Research by Statista expects the market for artificial intelligence to grow to almost 2 trillion dollars by 2023. This incredible number is backed by impressive statistics. Fishbowl, a professional networking app, conducted a survey among working professional in which 43% percent of respondents claimed to have employed AI tools in their daily work. A survey by Sortlist found that 52 % of the participating employers expect to use ChatGPT to answer customer questions in the future. The overall image is clear: the immense curiosity towards generative AI goes beyond pure marketing purposes, even though the actual degree of implementation varies significantly. One business area that shows a particularly high potential for AI implementation is customer service – especially through new and even smarter service chatbots.

Meet your digital customer service colleague

Service chatbots have been a part of the virtual customer shopping experience for quite some time. They are based on machine learning algorithms and provide information on products or answer simple questions. However, before the advent of ChatGPT, training them was time-consuming. Chatbots needed to be fed large numbers of samples for each task and their answers to the most common questions had to be validated individually. Consequently, the areas in which chatbots could interact with customers remained limited. Inquiries that went beyond the standards would be forwarded to human customer representatives. Additionally, users felt that the answers given by conventional chatbots lacked a human touch.

The customer service of the future?

Let’s envision what the future of customer service could look like instead:

Over the past months, you have been learning how to surf. Now you’re ready to buy your own board but are unsure which one is the best fit for your level of experience. You come across a surfboard shop in a nearby town that offers a virtual assistant via a Google search. Here’s how the conversation between the two of you could go.

Virtual Assistant: Hi! Are you looking for anything specific?

You: I am looking for my first surfboard. I am still pretty new to surfing and want one that is easy to handle, but that I can also use as I become more skilled.

Virtual Assistant: Congrats on picking up surfing as a hobby! Where do you plan to go surfing?

You: I am located here at the Atlantic Coast, but I also travel a lot across Europe with my van and want to take my surfboard with me.

Virtual Assistant: Got it. Are there any boards that you have tried out before? Which features did you like? Is there anything else I should know before I make suggestions?

You: [sends two images] Here are two boards that I got to try in surf school. I liked how steady they were but am looking for one that’s easier to maneuver.

Virtual Assistant: It looks like you’ve tried out longboards before. Based on your descriptions, a hybrid surfboard might be a better choice. You can learn more about different types of surfboards here. And here are some of the models that we have in stock.

You: They look good. Is there a way that I can test the first two?

Virtual assistant: Absolutely. The team at our Calais location would love to get to know you. They are available from Tuesday to Friday, 9 am to 3 pm. Would you like me to book a meeting for you and reserve test boards?

Did you notice any differences between the conversation above and the chatbot behavior we’ve become used to? First, the assistant is able to respond to a wider range of questions. Secondly, it can identify elements on images and translate them into text. Thirdly, it can access calendars within the retail company, book meetings and even equipment. And finally, the language used by the assistant is more informal and matches the conversation that would be appropriate in physical stores. A generative AI tool that has been trained for this kind of conversation and is integrated into the IT infrastructure of the company makes this all possible.

The limits of generative AI in customer support

While a virtual assistant like the one above can already be set up today by using generative AI such as ChatGPT or Google Bard, there are limitations inherent to the technology. The chatbot must still be trained on how to reply to questions – even though that may only require three or four data sets instead of thousands, as was the case with previous machine learning algorithms. And, more importantly, the virtual assistant is only able to respond correctly to questions it has been trained for. Other questions may lead to the assistant admitting that it doesn’t know the answer – or even to so-called hallucinations, where the generative AI comes up with answers that are not based on available data. The assistant should therefore always make transparent where it finds its info.

The virtual assistant portrayed above seems to use language very naturally. However, current versions of generative AI still struggle with understanding parts of language that are more subtle, for example humor. And while the chatbot can respond to your questions above, it won’t steer the conversation towards a purchase like a real salesperson might. This can of course be solved by transferring the conversation to a human agent at a certain point, or by asking customers for their contact data and programming automated follow-ups, for example via email.