The customer landscape is no longer represented by a one-dialogue offering. When customers share their feedback with you, they expect you to listen and act. That’s why many businesses and institutions are leveraging machine-learning techniques, such as sentiment analysis (also known as emotion AI or opinion mining) to detect and extract sentiment from Voice of the Customer data.

Sentiment Analysis Defined

Sentiment analysis categorizes utterances sourced from a variety of places, such as surveys, reviews, or feedback by polarity (positive/negative, high-intent/low-intent, urgent/not-urgent). This type of insight is useful when summarizing public opinion, but not for a more complex, actionable, and accurate analysis.

We’re Not Mad, We’re Just Disappointed

It’s a great time to be a product owner—customers are generating huge amounts of intel on your product by expressing their opinions and emotions on forums, blogs, social media, and other platforms. This data holds the key to your customer’s expectations and desires. Not leveraging this data to its full potential would be a waste of free resources. Opinion mining this data to return a broad overview is a good starting point to understanding your customers, but it has its limitations.

Sentiment Analysis is unable to grasp the complexity of language

Human language, in its intricacies and complexities, cannot be categorized into only three buckets (positive, negative, neutral). To understand a broad range of intents and emotions, product owners need to move beyond a one-dimensional scale. This system portrays similarities to the current political landscape, for example, where it has become impossible to categorize everyone as left or right. Human behavior, language, and intent just aren’t that simple.

To be fair, sentiment analysis can be more complex than just placing utterances in three buckets. By running different types of sentiment analysis, you can categorize on a broader spectrum (like a 1-5 rating), detect several emotions (shock, anger, happiness, frustration, etc.), perform an intent-based analysis, or use aspect-based analysis to determine exactly what the sentiment was about.

Nonetheless, phrases are still being organized into limited or misplaced categories, when there are thousands of shades and degrees of human communication.

Unhelpful in the designing and planning phase

Before running a sentiment analysis, it’s important to ask:

How will we make this insight interpretable and actionable?

Capturing and acting on directional customer feedback is critical to planning future business decisions. Let’s say we determine that 82% of customers think negatively about our product. That still leaves us with many unanswered questions: Do we make the product cheaper? Do we make the product easier to use? Or do we scrap it altogether?

Without deeper analysis, you’re left with a ‘throw everything against the wall and see if it sticks’ method. It will take copious amounts of resources and testing to gauge the exact opinion of the product, therefore making it very difficult to improve.

Sentiment Analysis is Often Inaccurate

As I mentioned above, human language is complex and emotion AI is not at a place where it can detect the nuances, subjectivity, complexities, cultural variations, slang, and context of the written word. To derive an accurate analysis, its crucial to keep the context of every utterance intact.

Here’s an example.

If you just went on a first date, and she describes you to her friends as ‘nice’, then that date probably did not go well for her. Similarly, if you go to a restaurant, and describe the food as ‘interesting’, then you probably meant ‘it was bad, but I don’t want to be too blunt.’

In both scenarios, the utterances would most likely be categorized as ‘positive.’ It takes so much contextual understanding for a machine to pick that up correctly.

Are we equipping teams with the best tools to support a deeper analysis?

Teams are starting to leverage new technologies to support a deeper connection with their customers. Our tool, HumanFirst, equips teams with the ability to perform multidimensional and granular text analysis.

With HumanFirst, you can analyze voice call center transcripts, live chat logs, customer reviews, help center tickets, search queries, etc. into perfectly categorized and accurate knowledge, while keeping the context intact. What you’ll end up with is a multi-dimensional, modular hierarchy of intents made up of tens, hundreds, and thousands of utterances, coming directly from your customers. HumanFirsts' cascading hierarchy can reach any type of preferable specificity, so you will get actionable insights that help in the designing and planning phase.

HumanFirsts' cascading hierarchy of intents and utterances

It’s uniquely designed to be more granular. Instead of just capturing sentiment, you can discover every intent, opinion, or feeling from your data, with machine-learning-powered workflows to accelerate this process. You can actually toggle a granularity bar based on how specific you want your topics to be. With this approach, no dialogue will go unnoticed.

It’s time to move on

Until technology can perfectly measure skepticism, anxiety, hope, manipulation, and have all contextual understanding, we’ll undoubtedly miss a large part of the conversation. If businesses really want a two-way dialogue with their customers, sentiment analysis is not enough.

Read more about HumanFirst here.