Do you actually want to figure out what your customers are saying in those open-ended survey responses? Many companies have tons of written customer feedback. The problem is… They have no idea what to do with it. In fact, Gartner reports that approximately 80-90% of business data is unstructured. That includes survey responses, customer support emails, and other text forms. The good news is that survey text analysis techniques exist to make that data actionable. Understanding why survey text analysis matters is the first step to unlocking that valuable customer intelligence that has been hiding in plain sight all along.
In this article, we cover:
- Understanding Survey Text Analysis
- Core Techniques That Actually Work
- How To Choose The Right Approach
- Common Mistakes To Avoid
What Is Survey Text Analysis?
Survey text analysis is the process of organizing and making sense of written survey responses to determine how customers feel and what they think.
Think about it like this…
When someone types that “The checkout process was confusing and took forever” that is a wealth of data. Now, imagine if you got that type of response from 1000s of customers. Trying to read all of them manually is a task too great for anyone.
Survey text analysis techniques aim to make sense of large volumes of text feedback without spending weeks manually analyzing each response.
Why Does This Matter For Businesses?
The thing most companies fail to realize…
Those open-ended survey questions can provide the richest customer insights. Close-ended questions provide numbers and statistics. Open-ended questions give reasons.
Why? Because the reasons provide a full picture of what is really going on.
Companies acknowledge that managing unstructured data is their biggest data challenge. That includes survey responses, support tickets, and call transcripts.
Without proper survey text analysis techniques:
- Customer complaints can get lost
- Product improvement opportunities go unnoticed
- Trends are hidden in the data
- Decisions get made on a hunch instead of evidence
It’s super frustrating.
Core Survey Text Analysis Techniques
On to the good stuff. The survey text analysis techniques that actually work. This is what data teams are using to generate actual results from messy feedback data.
Sentiment Analysis
Sentiment analysis is a process of determining whether survey responses are positive, negative, or neutral.
It works like this…
The technique scans text feedback and assigns an emotional score to each. A response such as “Absolutely love this product!” will be flagged as positive. Something along the lines of “Worst experience ever” will be flagged negative.
This way, you get a quick snapshot of overall customer sentiment without reading every response.
But don’t stop there…
Sentiment analysis techniques really only show a part of the whole picture. Combine them with other techniques for much more context.
Topic Analysis
Topic analysis is a process of grouping responses by theme or subject.
For example, you might see a survey about a software product get the following topics from the text analysis:
- Customer support issues
- Pricing concerns
- Feature requests
- Onboarding experience
- Bug reports
The topic analysis technique will use tags to identify each response by its core meaning. It will do so beyond simple keywords to look at context.
The result is that you will be able to easily see what topics customers are talking about most.
Keyword Extraction
This is the simplest of the survey text analysis techniques out there…
Keyword extraction is a process of identifying the most common words and phrases in survey responses. It is a very quick way of identifying patterns without heavy lifting.
Want to see what customers are talking about most in their responses? Run a keyword extraction and see what rises to the top.
The catch is that these keywords are devoid of context. For example, “great” could be talking about great service or great disappointment, depending on the sentence around it.
Theme-Based Categorization
Theme-based categorization is a process of grouping similar answers together even if the exact wordings used vary.
Imagine that one customer wrote that “shipping took too long”. Another customer wrote, “waited weeks for delivery”. While the phrasing is not the same, these customers are clearly talking about the same thing.
Theme-based categorization techniques spot those type of relationships between answers. They look at meaning, not exact phrase matches.
Manual vs Automated Analysis
There are two main approaches to applying these survey text analysis techniques. Each has its place and time.
Manual Analysis
Manual analysis means that humans read each response and make decisions about them.
The good things are that:
- It is great at catching nuance.
- It understands sarcasm and context.
- It is flexible enough to handle edge cases.
The bad things are that:
- Manual text analysis is terribly time-consuming.
- Categorization decisions can be inconsistent from one person to another.
- Manual analysis does not scale.
For small surveys of under 100 responses? Manual analysis can work just fine. Beyond that? Manual analysis is a choke point waiting to happen.
Automated Analysis
Automated analysis is the application of algorithms and software to analyze responses at scale.
Today’s tools use natural language processing (NLP) technology to understand text in a human-like way. Machine learning based NLP can even identify misspellings, slang, and entirely new words and phrases that it has never seen before.
The good things are that:
- Automation can process thousands of responses in a matter of minutes
- Categories are assigned consistently each time.
- Automated approaches can scale as data volume increases.
The bad things are that:
- Automation always has an initial setup cost.
- Highly-specific context can sometimes get lost.
- Software tools need good-quality training data
For most businesses that handle regular surveys, automated analysis is the way to go.
Common Mistakes To Avoid
Done poorly, these text analysis techniques can cause as many problems as they solve. Here are the mistakes to watch out for…
Ignoring context. The meaning of a word depends on its context. A single word can be a positive or negative sentiment depending on the sentence it is in.
Over-relying on single techniques. Always use multiple analysis techniques together for the richest insights. Sentiment analysis, for example, is even more powerful when combined with topic analysis.
Not acting on what is found. Analysis for the sake of analysis is a waste. Text analysis techniques only have value when insights change how decisions are made.
Skipping data cleaning. Don’t try to clean data after the fact. Survey text analysis techniques will give you garbage insights if the input data is unclean. Remove duplicates, irrelevant answers, and spam responses before analysis begins.
Getting Started With Survey Text Analysis
Want to get started right away?
Pick a survey and one technique to begin with. See what insights you can glean before scaling out to more surveys and techniques.
Survey text analysis is more accessible than ever. The tools available today make it possible for any sized business to do text analysis. No data science degree necessary.
The thing that is most important is that insights gleaned from text analysis actually be used to drive decision-making. That is where the value comes from.
Bringing It All Together with Survey Text Analysis
Survey text analysis techniques take seemingly insurmountable mountains of written feedback and condense them down into clear, actionable, and human-readable information.
The techniques covered here, sentiment analysis, topic analysis, keyword extraction, and theme-based categorization, form the basis for any survey text analysis approach.
Manual versus automated methods can be used, depending on survey size and available resources. Automation is the key to managing growth in survey text analysis.
The bottom line is that those open-ended survey responses are a gold mine. The right survey text analysis techniques just need to be used to extract it.
