CONTRIBUTOR
CTO,
Alchemer

Despite all of the hype around AI making business processes easier, before you begin implementing AI for your CX system, you need to know what you really want to automate and why. Then you must find reliable, accurate information to train your AI system and rigorously test it. Rushing to implement AI within your customer experience offerings could lead to another incident like Air Canada’s AI gaffe. In February 2024, Air Canada was required to honor a refund invented by their chatbot (contrary to their policy page). Shortly after, the company deactivated the chatbot.

This is one of the things that Forrester analyst, Rowan Curran, warned about in an April 2023 Forrester podcast. “When we’re talking to clients and anybody suggests creating a customer facing chatbot that has the ability to kind of generate content arbitrarily, that immediately raises a huge red flag for me,” Mr. Curran said. “Currently the risk of generating content that may have a negative effect on your brand, if you’re allowing users to create or you’re allowing the systems to create arbitrary content with no human in the loop, is very, very high.”

Know Your Needs First

Much of AI usage in the CX space has to do with language or prompts that are based on questions and answers. The other use is data that needs language-based feedback. All of these responses come from Large Language Models (LLMs). So, the first thing your CX team needs to do is determine whether or not they have enough data of their own to create an LLM and AI model that accurately gives customers the responses they want or need.

If your team decides on using an outside LLM, then the next thing the team needs to figure out is how they’re going to train the model and provide the context, so the system responds to customers with the right information. It takes a lot of time to train a model from scratch to respond in the right way. But there are existing models out there that do.

Be Conscious of Security and Intellectual Property

Once you’ve trained a model for context and desired output, you’ve really got to consider how you create something that delivers consistent unbiased answers. You need to be conscious that the model you choose doesn’t violate how your customers’ data is secured. And you also need to make sure that you aren’t violating any copyright concerns, because if you’re using the same language as your competitor, and they’ve copyrighted their model, you could have challenges.

In short, your CX team needs to be conscious and involve the right leaders from your Infosec and Technology teams to make sure all of the risks are mitigated.

If you’re running the call center, you need to have a model that knows how to respond to sentiment. By that, I mean the AI system must make sure that any analysis, or questions and answers have the context of knowing whether something is good, bad, happy, sad or mad. Remember, these systems are not having actual conversations, they are using logic to anticipate the next best word or answer. As Mr. Curran says, “Essentially these things are people pleasers overall.”

Get People on Board First

The secret to getting people on board with AI is to help your people understand that the goal of AI is not to put them out of a job but to make their jobs better, and make their work more rewarding for them, the customer and the rest of the business.

Sometimes we get lost in all of the ChatGPT and AI hype, seeing AI as a huge thing, when it really can be something as simple as automating how your business gets its data. Don’t spend your time thinking of every possible use case and all of the magic AI can do.

Instead pick one thing, try it, and then pick the next thing. Because once you get one use case, or one good use for AI or automation, you will become very effective at deploying that across other innovative ideas.