No Jitter is part of the Informa Tech Division of Informa PLC

This site is operated by a business or businesses owned by Informa PLC and all copyright resides with them. Informa PLC's registered office is 5 Howick Place, London SW1P 1WG. Registered in England and Wales. Number 8860726.

Why Topic Modeling Is My Favorite Gen AI CX Feature

There is a lot of interest in using AI within enterprise communications, and there are certainly a lot of new use cases emerging. My favorite use case works with voice and digital channels, agents, and self-service, and it’s completely invisible to the customer and the agent. My favorite AI use case is topic modeling, or the practice of finding topics or themes within a body of data. 

Before I explain why, let me share one of my pet peeves regarding customer service. I believe contact centers often don’t know why their customers call. It’s a ludicrous accusation. I know contact center conversations are microanalyzed. But, I also know that the bots and IVR menus I encounter never quite seem to fit.

 

No, Not Netflix 

Here’s a simple example. The Gaylord Palms is a major hotel and conference center in Orlando, FL. It hosts large gatherings and events of all sizes back-to-back, year-round. It’s primarily a business conference venue but also doubles as a vacation resort. 

I needed to find the Gainesville meeting room. It’s reasonable to assume that finding meeting rooms in such a such a large venue is common. There were two clues before I reached my room that validated that assumption: they handed me a pocket-sized facility map when I checked-in, and then they promoted a custom wayfinding app in the elevator. 

I did consult the map, and it did list meeting rooms, but not the Gainesville room. So, I opted to call for assistance. This is where I encountered the Gaylord bot for hotel guests. They call the button on the phone “Consider it done.” Unfortunately, the bot didn’t know where the Gainesville room was. Evidently, it didn’t even know the word “Gainesville,” but it did offer me helpful assistance on how to view Netflix. 

Although this particular hotel understood the importance of wayfinding, the Gainesville room wasn’t included on the venue map. I am guessing the bot I encountered was only trained to answer questions applicable to any Gaylord property. The “Consider it Done” bot did not appear to be powered by generative AI, but I have doubts that would have helped as more people likely seek directions to the city of Gainesville than the meeting room. 

So, this experience was further reinforcement that contact centers don’t know why their customer’s call. “Consider it Done” did not get “it” done, and honestly, I forgot about the custom mobile app, but had taken the time to consult the fold-up map.

 

Topic Modeling for CX 

Now, let’s get back to topic modeling. The new generative-AI powered versions help contact centers identify topics in customer conversations. It’s a fantastic use case of AI, and it provides valuable and actionable insights. Topic Modeling can be used to monitor topic trends, support agent training, and define intents by simply “hearing” conversations.

These new topic modeling tools derive insights from the customer’s requests and the responses they receive. They work across channels, and can even discover hidden patterns that often exist. They can operate completely in the background. 

What they don’t do is force conversations into predefined categories. That’s the trap that limits organizations from understanding how conversations change over time – and bars them from understanding what customers really need. New topic modeling tools can even eliminate the need for the FAQ. I’m a glutton for punishment, so I consult many FAQs before attempting to reach an agent, and they rarely have the information I need. 

If you think about it, the notion of a frequently asked questions (FAQs), are really an admission of failure. They further validate that most organizations don’t know what their customers need. If they knew, they might take proactive action to eliminate the need. Since they don’t know, they create a collection of useful answers to assumed questions. Today, FAQs are (or should be) obsolete. If you can’t reduce the frequency of repeated inquiries, there’s something terribly wrong.

Armed with insights from modern topic modeling, where the data has emerged organically from customer interactions, the CX experience can be significantly improved in a variety of ways. It starts with adjustments to IVRs and routing (faster and cheaper than printing new maps). The topics can also be combined with sentiment analysis to better understand how the responses are impacting customer satisfaction. This is one of the simplest ways to understand the voice of the customer (VoC). 

Insights derived from modern topic modeling could create the killer FAQ, but instead, a better CX could involve making those details more knowable and easier to discover, or it could be used to change processes, communications and documentation and thereby eliminate the customer’s need to find this information.

Topic modeling is easy to do, and should be an ongoing process. Customer queries evolve, and responses should evolve too. The Gaylord initially implemented “Consider it Done” in 2005. Perhaps back then, the number one question was how to watch Netflix in the room. The hotel has expanded numerous times, perhaps the Gainesville room didn’t exist in 2005. Topic modeling creates a feedback loop to understand how customer service inquiries evolve.

It's pretty clear that the Gaylord Palms hotel know that guests need wayfinding assistance, but they missed wayfinding in their bot training. I also tried asking “Consider It Done” when they feed the alligators in the atrium. It didn’t know that either. Most organizations need to do better at collecting and understanding all the questions their customers have before implementing self-service bots. Generative AI-powered topic modeling is a good way to collect that information. 

Topic modeling is one of the best ways to start with generative AI. It can be everywhere (calls, emails, chats, texts, and videos) and completely invisible. The information can be used to improve the customer experience by reducing the need for outreach. 

Regardless, effective automation and CX really starts with understanding what your customers want. 


Dave Michels is a contributing editor and Analyst at TalkingPointz.