In the highly competitive world of business, understanding and enhancing the customer experience is of paramount importance. By leveraging sentiment analysis on customer feedback, businesses can understand and dig deep into the emotions of their customers, allowing them to identify and address pressing concerns, and fine-tune their products and services. As a result, organizations can boost customer satisfaction and loyalty, paving the way for sustained revenue growth.
This article explores how to augment the customer experience through feedback sentiment analysis, demonstrating the variety of actionable solutions it offers businesses.
You may be wondering how sentiment analysis can truly make a difference in your organization. As you read through this article, you’ll learn what role customer sentiment plays in business. In addition, you’ll learn how to use sentiment analysis to drive emotional impact. Furthermore, you’ll learn how cutting-edge technologies can help you overcome the challenges of delivering high-quality sentiment analysis.
In business, everything begins with the customer. All our revenue is voluntarily given to us by our customers. Most of our activities and expenses are all about creating value for the customer, spreading the gospel of that value, and convincing customers one by one about how products and services will solve their problems and improve their lives. Every day the customer gives, or take away, and his/her gratification or irritation can either make or break our business.
At the same time, we – as humans – tend to be driven by our deep emotions. Underneath our motivations, there’s really nothing else. Joy, satisfaction, pain, frustration, and anger drive our behavior and drive our learning. Give one delight, and one wants more of it and may recommend it to friends. Deliver one frustration, and one will start avoiding you or – worse – resent you and harm your reputation.
Our deep dependency on our customers and this deep emotional drive explain why sentiment analysis on customer feedback is so important in this overly digital world. You simply cannot act on the customers’ pains and frustrations if you are not aware of them and their causes. You cannot intensify the good of your product either even if you cannot recognize the fundamentals and the details in your product that your customers truly love and cherish.
To better understand sentiment analysis’s practical applications, imagine its benefits in the retail industry. Let’s say a retail company uses sentiment analysis to identify customer pain points in their online shopping experience.
By analyzing customer feedback, they can discover recurring issues with website navigation and the checkout process. And with this information, the company may target improvements to their website. This will result in a more seamless and enjoyable online shopping experience for their customers.
This example demonstrates how sentiment analysis can be leveraged to pinpoint specific areas that need attention. This allows businesses to make informed decisions and enhance customers’ experiences.
Let’s do another one but this time, in a SaaS environment. A software company leverages sentiment analysis to prioritize bug fixes and feature updates based on customer feedback. They can find out what’s frustrating users and which features are most requested by analyzing user comments and reviews.
By doing this, the company can allocate resources efficiently and focus on improving things that will have the biggest impact on user retention and satisfaction.
Like the first example, this one illustrates how sentiment analysis can help businesses make decisions based on data and improve their products and services according to customer preferences.
Now you know how it works in the real world, let’s further explore ways how to use sentiment analysis to drive emotional impact in more detail.
One needs to identify something that drives emotions and is actionable. The first step is what we call sentiment analysis. In its fundamentals, it’s about identifying the emotional tone – the sentiment – of a sentence. In practice, it’s about recognizing the text ‘I love the app’ as positive, ‘I saw the app’ as neutral, and ‘the app is driving me crazy’ as negative.
This sounds nice and easy and for the user – easy it is. Lumoa’s user interface (UI) provides a clear and intuitive way to visualize sentiment analysis results. In the Lumoa UI, users can see a graphical representation of the sentiment scores for different categories or topics, as well as individual customer feedback with their corresponding sentiment labels. At the same time, the results can also be filtered by date, sentiment category, or other relevant factors, making it easy to find trends and patterns. Here’s a view of the sentiment analysis in the Lumoa UI.
Yet, it’s quite a challenge to deliver high-quality sentiment analysis in a real-world application. This difficulty is raised by several factors. One core issue is the diversity of languages. For example, while Lumoa can deal with 120 languages through translations, it also supports more than a half-dozen languages as translation targets. It means that sentiment detection needs to operate over multiple languages. Another issue is the diversity of humane expressions.
The richness of emotionally loaded words that range from ‘good’ to ‘niggle’, different ways to express irony or negation ‘could have been better’ and plain typing mistakes ‘it was great’ mean that the machinery needs to master extremely wide vocabulary and nuance of expression.
A third more practical issue relates to the unit of sentiment.
Many sentiment analysis solutions make the mistake of predicting sentiment on too high levels, or in a worst-case scenario, on the document level.
This makes it difficult to assign a sentiment to a specific theme. For example, the text ‘content is great but the app crashes’, has 2 separate sentiments: one related to the application content and another to its quality.
Nowadays, the-state-of-the-art sentiment detection is typically done with pre-trained and large deep-learning networks. Lumoa is no exception, as we are using an XLM-RoBERTa, which is a large multi-lingual language model pre-trained with 100 languages and 2.5TB of text. For reference, that’s about 100 times the size of the 2023 English Wikipedia. The use of this model solves many issues. The cross-language nature of this model means that it can understand about 100 different languages without separate training.
Large language models contain huge amounts of individual details (270M parameters), which capture diverse vocabularies and human expression. Lumoa also has additional logic for separating text for sensible units for sentiment analysis, and for incorporating additional contextual data in predictions. As a result, Lumoa can reach 80%-85% sentiment detection accuracy across various languages, which is considered a highly competitive result in the industry.
While one needs to identify emotional context to drive emotional impact, this is only a part of the story. According to the Lumoa Actionability framework, for an insight to be actionable, it needs to have 3 different properties:
The reason why these 3 different aspects need to be present is somewhat obvious. Let’s consider a statement ‘recent update was ok’. Even if the insight has a clear cause (‘update’) and it’s a novel (‘recent’), if it doesn’t imply an issue or a pain it’s not something you can or need to fix. At the same time the statement ‘everything has become horrible’ may be impactful and novel, but if it lacks a clear root cause, you cannot really act on it.
On the other hand, while the theme ‘price is still high’ implies a clear issue and its cause, if the same theme has been present from day 1 and no one has acted on it: one will likely act on it now. It’s quite typical that there is always someone complaining about the price or some other aspect of a service, which is less of a bug and more a feature of the business model.
So, in practice, the sentiment can become an unactionable ‘vanity metric’ unless it’s bound to the cause of the sentiment and to some kind of trend analysis. For example, in Lumoa this coupling of the emotional impact of the cause and novelty is done in 2 places:
The following screenshot demonstrates how the different aspects work when combined. The top left answers to questions: “What the sentiment is and how it has changed?”. There, you can see the total positive sentiment for November and that it has dropped significantly. The top right answers the question “Why has it dropped at a high level?”. There you can see for example that the sentiment has dropped by 0.23% points, because of the Price and Payment topic.
On the middle-left, you can find this answer more granularly and find that the main subtopic driving the change is Discounts and Promotions. On the middle-right, you can find the exact causes behind the Discounts and promotions negative sentiment. One example is the actionable insight related to the Pull&Bear application Setup and Registration topic. When you read the feedback, the problem becomes clear: customers are not getting the promised 10% discount for downloading the app.
In this way, the sentiment analysis can be combined with cause and trend analytics to form a coherent overview, that leads you from the top-level change (-7.3%) to changes to the concrete actionable details, that drive the emotional impact.
Because emotions drive customer behavior, sentiment analysis on customer feedback forms a tool for driving improvement in customer sentiment, customer behavior, and the business itself. In turn, these can drive directly not just loyalty, but also conversion to paying customers by guaranteeing a better early-stage customer experience. Sentiment analysis can also help organizations become more customer-centric, by making it better tuned to the customers’ pain and delight.
Still, sentiment analysis is not trivial, and it comes with its caveats and challenges, that need to be managed to deliver high quality. Sentiment analysis is often not enough alone to deliver value, but it needs to be combined with some kind of cause/theme analysis and trend analysis to provide insights that are interesting and actionable.