Follow Datanami:
February 24, 2017

How Topic Modeling Can Change How Brands Interact with Customers

Matt Matsui

Nobody likes a Monday morning quarterback. He’s the guy who always knows exactly how to run a play to score a touchdown…the day after the game. As much as the postgame know-it-all gets under everyone’s skin, many brands are Monday morning quarterbacking with one of their biggest assets: their customers.

Often, companies take a reactive approach to important decisions about their customers because “hindsight is always 20/20.” The problem with hindsight is that it doesn’t always tell brands how to fix issues until it’s too late.

Many companies are using speech analytics to look for exact answers within customer conversation data before taking action but, in the midst of massive amounts of information, that can mean many verbal cues go unnoticed, and those cues point to a bigger picture. But, what if brands can determine even the seemingly unpredictable trends before customers have specifically stated the problem? With topic modeling, it’s possible.

Many brands are using speech analytics to gain insights into customer sentiment and needs. Conversations with customers are recorded, speech is translated to text, and brands receive alerts when a pre-determined keyword or phrase is spoken.

Analysts can then drill down into conversations to further understand what customers are saying so brands can make appropriate changes to a product, service, or marketing campaign. While this is important, it’s impossible to classify every keyword or phrase that may be useful, making it difficult to identify some issues until it’s too late.

For example, several customers of a restaurant chain may call to let the restaurant know that they aren’t feeling well. However, until they start saying specific words and phrases like “nausea” or “food poisoning,” the company may be missing a larger issue. By the time the issue is finally discovered, they’re already far behind.

With topic modeling, brands can see beyond the actual words being used to discover trends and get ahead of potential problems. Topic modeling uses machine learning and natural language processing to find abstract topics in large pools of data so brands can identify the larger trends in the midst of conversations, including topics that aren’t even on the radar.

In topic modeling, algorithms analyze the speech-to-text conversation data, find related words, and extract themes from the word clusters. These statistical models determine the connections between those themes and how they change over time. As the data set expands, the algorithms grow, change, and learn to better process information, so the themes become more accurate.

These models allow brands to categorize conversational data by concept rather than using keywords or phrases, so they can take action sooner, rather than looking in the rear view mirror.

In the restaurant chain example, the food poisoning outbreak is already in process by the time customers start feeling sick. However, if the company uses topic modeling to identify food-related subjects, the algorithms will notice a change in patterns when things like “feeling nauseous”, “not feeling well,” or “not sure what I ate” begin to appear in relation to food.

Once this happens, they can use that information to find other themes, like isolating the particular item that was consumed that ultimately caused the illness. The restaurant can then proactively stop serving that item and use things like loyalty program information to reach out to diners who may have eaten that dish on that specific evening and offer a voucher or incentive for the customer to dine there again. These actions can be taken even before many patrons feel the full effect of the illness.

To build successful engagement and retention strategies that foster loyal relationships with customers, brands must continually look for ways to improve how they listen to customers. Topic modeling allows brands to see the bigger picture using some of the subtler verbal cues that happen during customer conversations, and it gives brands the opportunity to gain deeper and faster insights by identifying topics and trends much more quickly. With this critical information, brands can enact containment strategies to ensure issues don’t spread, be more proactive with customers when a problem has been recognized or a request has been made, and improve overall customer service levels.

Companies get so much value from having an efficient way to make sense of all the customer conversation data that pours in every day, and topic modeling provides insight into those conversations, even when a brand isn’t sure what to look for. With that insight comes knowledge, and that knowledge is accessible more quickly than ever before.

Now, brands can be nimble and more effective to get ahead of problems, make appropriate changes, or put in place better service strategies, unlike the Monday morning quarterback. With topic modeling, companies have the opportunity to gain the insights they need to keep customers happy and ditch the Monday morning quarterback once and for all.

About the author: Matt Matsui is the senior vice president of products, markets, and organizational strategy at Calabrio, where he oversees company-wide go-to-market efforts. Matt joined Calabrio with more than 25 years of experience leading product and marketing organizations for a broad range of companies, including ACNielsen, Cognos, Fair Isaac and numerous early stage analytics firms. Prior to joining the Calabrio team, Matt was a managing partner for Veralytics, a predictive sales and marketing analytics and product insights company.

Related Items:

Secrets of the Data Whisperer

Don’t Be a Big Data Snooper

Datanami