Complicated-sounding terms like “machine learning,” and “sentiment tracking” are thrown around a lot in today’s marketing and customer experience circles, with new AI-powered tools being announced every week.
Though most business owners know that these types of software are important, they may be turned off by the tech-heavy language. However, you don’t need to be a professional programmer to understand the basics of text analytics and how it can drastically improve your CX efforts. We break it down for you below.
Text Analytics Definition
First things first, let’s define what text analytics actually is. Text analytics is the process of extracting quantitative data from written (qualitative) information. The goal of text analytics is to analyse qualitative feedback quickly and at scale to uncover trends and patterns. There are a large variety of ways in which this process can be useful for businesses, and we’ll outline some examples below.
What Do Businesses Use Text Analysis For?
Customer Feedback: Perhaps unsurprisingly, the main reason businesses use text analytics is to analyse customer feedback. Sending out surveys asking customers to rank responses on a numerical scale can only get you so far. The more valuable analysis usually comes from freeform responses that can be gathered through detailed customer feedback surveys, social media mentions, online reviews, and more. By using text analytics tools, businesses can cut down on the employee bandwidth needed to manually sort and categorise responses from these important feedback sources.
Risk Management: Enterprise-level companies, such as financial lenders, are starting to use textual analysis to identify risky investments or lending practices. These types of tools can quickly pull information about a specific company, including news articles and reports, and categorise this information based on a set of pre-determined rules to see if industry experts view them as a good investment.
Targeted Marketing: Text analysis can also be used to better refine audience segments for more accurate marketing. These tools can gather demographic and psychographic information about a user’s interests and buying habits online in order to build out more detailed personas that can be served with the right types of ads.
How Does Textual Analysis Actually Work?
Now that you have a general idea of what businesses use text analytics for, we’ll dive into what the AI is actually doing. This is just a basic overview to help you better understand the process, as machine learning is, without a doubt, a very complicated subject. After all, natural language processing mimics the workings of the human brain!
Step 1: Tagging and Chunking
The first step machines take to understanding language is to go through each piece of textual feedback and tag each word with a part of speech, such as noun, verb, adjective, etc. Next, the sentence is chunked into phrases based on where these parts of speech occur. These are usually categorised as noun phrases, verb phrases, and prepositional phrases. If you’re a little rusty on elementary school grammar, prepositions describe spatial relationships, for example “on,” “after,” “into,” etc.
Step 2: Parsing
Once the sentence is tagged and chunked, the bot will separate it into different elements or sections based on the defined phrases. This step is important because a single piece of customer feedback can have multiple meanings or sentiments. For example, the phrase “Love your product, but it’s a bit expensive” consists of multiple elements including “product,” “price,” “positive emotion,” and “slightly negative emotion.”
Step 3: Rule Setting and/or Topic Modelling
Now that you understand how the AI reads and categorises pieces of qualitative information, we can get into the analysis part of the process. There are two main ways to go about this: rule-setting or topic modelling. With rule-setting, the AI is going off of a pre-determined set of rules laid out beforehand. For example, the word “expensive” could have a rule attached to it that directs the AI to add this review to both the “price” and “negative emotion” categories. The pro of using rule-setting is that you can be reasonably sure of accurate results from the beginning, but the con is that the set-up time takes much longer.
Topic modelling is what’s known as “unsupervised” machine learning (whereas rule-setting is supervised) because the AI essentially learns how to categorise and analyse on its own based on recurring themes and sentiments. Just like humans, the bot will learn as it goes and gets more accurate over time. The pros and cons of this method are just the opposite of rule-setting: There is virtually no set-up involved in the beginning so you can start using the tool right away, but it can take some time before results are accurate enough to be truly actionable for your business.
Take a look at the infographic from Chattermill below for a visual illustration of these points, along with 5 more real-world examples of how companies can use text analytics to improve CX.