How to Reduce the Risks of Using Chatbots

Can you give your AI chatbots your company's real core values and brand voice? With the White House and the Consumer Financial Protection Bureau cracking down on companies’ use of chatbots, businesses are under new scrutiny to ensure that their AI-powered chatbots aren’t leaving customers in a dreaded doom-loop of non-answers and evasions. Building a chatbot that meets regulators’ (and customers’) expectations requires CTOs and CMOs to ensure that chatbots are trained with the same diligence and values as human employees.

While chatbots have been around for years, only since the rapid popularization and growth of large language model AIs, like ChatGPT and Anthropic, has the market for customer-oriented chatbots exploded. Rather than being limited to certain customer-response flows, modern AI chatbots can answer a huge variety of more complex questions, and thanks to lower computing costs, are now more accessible than ever to the average user.

And while companies have been quick to adopt chatbots, it’s not without some risk. While the White House is mostly concerned with the frustration customers frequently report in interacting with chatbots, companies like Air Canada have found themselves on the wrong end of a lawsuit after their online chatbot hallucinated its own refund policy.

AI is not a set-it-and-forget-it technology, and companies cannot simply deploy an AI chatbot without doing their due diligence to ensure that customers are getting the right information with the right messenger. While the large language models upon which these modern chatbots are built can offer more natural-sounding answers, at the end of the day, they are merely pulling together multiple logic streams to predictively estimate a single answer.

In other words, companies that fail to properly assess their AI streams risk handing their customers off to a more advanced Clippy, Microsoft’s early "Office Assistant" interface. Companies would never deploy an untrained customer service representative, so they shouldn't do it with their chatbots, either.

We’re still in the nascent days of generative AI development, and building truly effective chatbots will take time. But with these bots being looked at legally as binding extensions of the companies that use them, it’s key that businesses understand the potential and pitfalls that come with these new chatbot technologies. Here are some things to consider.

How to reduce the risk—and improve the results—of using chatbots

Data is your best friend

AI struggles in ambiguity. Left to its own devices, AI chatbots will turn to “fill-in-the-blank” sort of guesswork based on its interpretation of the data. A chatbot that’s been told, for example, to ensure that a customer is happy, will do what it understands to be best to make the customer happy—even if that answer is way off base.

The key to combating these errors is data. Not only should you be training AI from the outset on customer data—emails and phone calls, for example, that provide real customer experiences with real human responses—but training should be ongoing as well. Things like a customer satisfaction (CSAT) score and Net Promoter Score (NPS) can help to refine any issues that customers are having with the chatbot in real time.

Garbage in, garbage out

We know that courts and customers are increasingly looking at chatbots as extensions and representatives of the companies themselves. It’s only logical, therefore, that these chatbots should be trained like a human representative of the company would be.

In the same way that a copywriter at your company would have an understanding of the values and tone that your brand takes in written formats, so too should your chatbot. What are the values of your company? How should a customer experience feel? What should it sound like?

The more detailed and specific instructions you can give to a chatbot, the better the responses from the chatbot will be.

Stay customer obsessed

One rule of AI bears repeating: humans have to stay involved with any AI-driven service, but especially those that are customer-facing. Humans need to be around to monitor for biases or mistakes, and customers need a way to get in touch with a real person. There are some situations that AI simply should not be managing. Chatbots are not infallible, nor are they suited to every situation.

Companies need to keep the customer experience in mind at all times. Chatbots are a part of that, but shouldn’t be the entire plan. We’re not yet at a place where chatbots can wholly replace a customer service workforce.

Businesses need to map their customer journey from beginning to end, and understand where the pitfalls are. There are hard and fast rules (a chatbot should never be in a position of collecting sensitive data or password information, for example) and then there are scenarios that occur on a more case-by-case basis: What do you want to do when a high-value customer raises a problem? Are there certain words that should immediately raise a flag to a human representative?

The key to being successful with a chatbot

This line between ease and empathy is key for businesses to be successful with a chatbot, and can be the difference between being viewed as a callous corporation versus inspiring lifelong brand loyalty.

Business leaders need to understand that we’re only at the tip of the iceberg with what AI can do, and the technology isn’t going to be perfect right away, which means using chatbots can involve some risk. The mission of tech implementation teams needs to be to find the “Goldilocks” system, one that accomplishes what it needs to, while staying within the bounds of the company’s fault tolerance.

And that’s not going to happen unless your company’s values get equal billing while you're training your AI.

AI chatbots and small business FAQs

About the Author

Post by: Pete Sena
Pete Sena is the founder/CEO and chief creative officer at Digital Surgeons.

Company: Digital Surgeons
Website: www.digitalsurgeons.com