Lesson #4: Text Analytics Is More Than A VoC Feature; It's An Absolute Must-Have

Voice of the Customer | Listen or Die | Text Analytics

Reading time: 11 minutes

The following is an excerpt from Listen Or Die by Sean McDade, PhD.

The true “voice” of your customer is the unique words they share with you each time you ask them “Why” or “Please tell me more about that” in your survey. There is gold in each comment that customers share with you. The question is this: how do you best mine for that gold? Enter text analytics.

But first, let’s take a step back. No matter how you are listening to customers (solicited, unsolicited, verified, observed), the data you will ultimately collect will be one of two types: structured or unstructured. Let's dig right in.

Structured Feedback

Structured customer feedback is the most common, the easiest to deal with, and super important in spite of this lesson’s title.

Let’s consider an example. A typical question in your transactional (solicited) survey might be, “How satisfied were you with your most recent experience at Hotel ABC on a scale of 1 to 5, with 5 being very satisfied and 1 being very dissatisfied?”

When customers provide their answer to this question, it comes in the form of a number—in this case, a 5 would indicate very satisfied with the most recent visit to the hotel, a 1 would indicate very dissatisfied, and so on.

Almost any VoC software platform can easily analyze these data and create graphs to aggregate and compare the responses: Maybe 30% of respondents were very satisfied, 35% very dissatisfied, and so forth.

For example, consider this graph, which is easily rendered based on structured responses:

NPS Segments

What’s not easy is unstructured customer feedback that often follows a structured question.

Unstructured Feedback

Unstructured feedback is key to understanding why those satisfaction levels are what they are.

Continuing with our earlier example, the next question in the survey might be “Please tell us why you feel that way.” This is prompting the customer to type in open-ended (unstructured) comments to explain their level of satisfaction with the experience.

A customer might type in something like “The front desk took too long to check in,” “The room was dirty,” or “The people staying in the room next to me were loud and I didn’t get any sleep.” 

And these unstructured comments are not just for solicited, transactional surveys. Unstructured customer feedback is key to unsolicited feedback as well, especially social reviews. Visitors to review sites like TripAdvisor can write entire paragraphs of open-ended feedback on their recent experience.

Unstructured feedback is incredibly valuable because it indicates why people feel the way they do. Without it, CX professionals would be lost. How would they know what to do to fix problems without context?

A Unifying Framework

I have introduced a lot of terms regarding types of feedback. The ones you need to know as a CX professional are solicited, unsolicited, structured, and unstructured.

The following is a unifying framework that gives examples that I hope will make this clear:

 

Feedback Framework

 

Making Sense of Unstructured Feedback

The question then becomes this: how do we make sense out of these unstructured comments from customers?

Unlike structured data, which are easy to display graphically or in tables, each piece of unstructured data is different—unique to each customer who took the time to provide it.

There are two options to handle unstructured feedback: humans or machines.

Humans (Open-Ended Coding)

Humans are the most expensive method to make sense of unstructured comments—often prohibitively so, because not only are they expensive knowledge workers, but also the effort doesn’t scale. 

An approach to unstructured data analysis made popular with market researchers is called open-ended coding. Here’s how it works.

A person is tasked with reviewing a small sample (one hundred is common) of open-ended customer comments and the goal is to identify major “themes” from these comments. They create what is called a “codebook” that contains a handful of major themes (usually five to ten). Then usually another person reviews the remainder of the open ends, one by one, and assigns one of the major themes to each comment. We are talking about thousands of comments in some circumstances and days or weeks of work. The result is that unstructured comments become structured!

Then market researchers create a report that quantities these comments—such as 43% of people mentioned Wi-Fi as a problem, 16% mentioned the spa as an issue, and so on.

The human approach is virtually impossible to scale. For PeopleMetrics’ larger clients, we send out more than 20,000 surveys a week, which results in thousands of open-ended responses. In a given day, even the best human coders might be able to handle a few hundred comments.

Machines (Text Analytics)

Machines, on the other hand, can scale infinitely. All sophisticated VoC software platforms will have a text analytics module available. Let’s dig into the value that text analytics provides.

Using a computer algorithm that identifies common themes or topics by scanning and grouping customer comments, text analytics provides an unprecedented ability to make sense of and take action on large volumes of unstructured customer feedback. Here’s how it works.

Wi-Fi could be a topic, but some customers might call it wireless internet. Others might say only internet or speed of internet or web access. After a little tuning by a human, the machine quickly learns to identify all those terms as Wi-Fi.

The next step is that the machine is trained to understand customer sentiment for each comment. One customer might say, “I love this hotel because the wifi is so fast.” Another customer’s comment could be, “wifi, there was nothing fast about it, so frustrating.” The machine then assigns a sentiment score for each comment related to given topics, usually ranging from –1 to +1.

Then text analytics organizes all topics in a visual display, often in the form of a word cloud. If you’ve seen one, you already know that the size of each word reflects how often the topic is mentioned. Its color indicates the sentiment: red usually indicates negative, green for positive, and yellow for neutral. The image here shows an example from a telecommunications company:

 

Text Analytics

 

Text analytics can quickly identify trends in real time and at scale. This allows you to identify and get in front of major problems immediately!

By contrast, human coders might take weeks to notice that, say, the telecommunication company's pricing is an issue, as in the earlier example.

Still need convincing?

If you’re still concerned about the difference in accuracy between open-ended coding and text analytics, let it go!

Consider the trade-off between greater accuracy on every comment and the ability to spot key trends in real time that will help you improve CX!

In reality, if you don’t have a quick, accurate, and affordable way to process your unstructured customer feedback, collecting it in the first place is a pointless exercise. Text analytics represents the present and future. Without it, you are not getting all you can out of your VoC program.
 
This is an excerpt from Listen Or Die by Sean McDade, PhD. 

Click here to download the book.

Download this chapter + a bonus checklist

 

Sean McDade, PhD is the author of Listen or Die: 40 Lessons That Turn Customer Feedback Into Gold. He founded PeopleMetrics in 2001 and is the architect of the company’s customer experience management (CEM) software platform. As CEO, he guides the company’s vision and strategy. Sean has over 20 years of experience helping companies measure and improve the customer experience. Earlier in his career, he spent five years at the Gallup Organization, where he was the practice leader of their consulting division. His company offers CEM software with advanced machine learning solutions and hands-on analytical support to help companies make sense of their CX data. Sean holds a Ph.D. in Business Administration with a specialization in marketing science from Temple University in Philadelphia. He has published eight articles in peer-reviewed scholarly journals and has taught over 25 marketing classes. Sean was named a 40 under 40 award recipient of the Philadelphia region. He is an active Angel Investor, including investments in Tender Greens, CloudMine and Sidecar.

 

P.S. What did you think of this blog post?

I loved it!  
It was okay.
 
It wasn't valuable.

Posted on 07-31-2018