Maya Angelou once said “people will forget what you said, people will forget what you did, but people will never forget how you made them feel.” Results from a recent Mckinsey study demonstrate what this means for businesses: After a positive customer experience, more than 85 percent of customers purchased more. After a negative experience, more than 70 percent purchased less. So getting this wrong can prove a costly exercise.
Rather than rely on assumption, how can a business know exactly what makes the customer feel like they are receiving superior service?
The answer lies in the deep analysis of customer sentiment. Tracking simple metrics like NPS (Net Promoter Score) is pointless if customer feedback on how to increase that metric is not strategically used to make positive change, even if there is a direct link to revenue. In this article, we reveal that customer sentiment analysis is not just about assigning a positive/negative label. Plus we explain how to deepen this analysis and benefit practically from the results.
Isn’t NPS enough?
Raw metrics like NPS provide some level of awareness of customer sentiment. However, research suggests that NPS is too broad a brush, missing a lot of useful detail on how customers feel. Even before looking at the specific reasons a customer may be unhappy, a lot of detractors can still talk positively about elements of the service they receive. Promoters can detail elements they don’t like, despite liking the service as a whole.
By analyzing the sentiment more accurately, and in particular finding the things people are really unhappy about, you can focus more on what will make a difference. Someone who is a detractor but expresses positively is unlikely to tell the whole world about their issues. Someone who is vehement will.
What is sentiment analysis?
Most commonly sentiment analysis refers to assigning a metric to a piece of text that details how positive or negative said text is. This metric is also called polarity, because it returns a value along a single dimension ranging from +1 (extremely positive) to -1 (extremely negative). By using thresholds, comments can then be split into defined buckets: positive, negative or neutral. Regardless of whether the sentiment is bucketed, the scores need to be normalized to get rid of particularities in how people express themselves.
How does an algorithm do this? The simplest type of algorithm uses a dictionary to look up which words or phrases indicate which sentiment. If a text says ‘all you need is love’, it marks it as positive. If a text says ‘I still haven’t found what I’m looking for’, it marks it as negative. This kind of dictionary or knowledge-based approach works to some degree. However, advanced algorithms that use machine learning can capture nuances. For example, the fact that ‘bloody’ may actually be a positive indicator if used in the phrase ‘bloody excellent’. These algorithms learn from large data sets, often already graded by people as to what constitutes positive and negative.
Advanced algorithms that use machine learning can capture nuances, such as the fact that ‘bloody’ may actually be a positive indicator if used in the phrase ‘bloody excellent’
Using the same principle, both dictionary and machine learning based approaches can also determine more specific emotions such as frustration, anger, or joy.
Accuracy depends on how closely the test data set resembles the dataset used to develop the dictionary or the machine learning model. Both can perform poorly if the datasets have little in common.
Why is customer sentiment analysis different?
Applying sentiment analysis to what customers are saying is a little more difficult for two reasons: how much they say, and how they talk.
The first difference is easy to understand: reviews and NPS comments are normally short snippets of text. This means that each piece of text provides only a little information when deciding its classification. Most algorithms are designed to handle large pieces of text like news articles. They might not work on customer reviews.
The second is that how a customer writes a support ticket is different to how they will respond to a survey, or write a Facebook post to their friends. In the case of a support ticket, they will (hopefully) be more factual and a Facebook post will be informal and less specific. If the same analysis is used then the sentiment assigned to each will be less accurate. It is important to tailor the analysis to the medium.
Why aspects of customer sentiment matter?
Customer sentiment isn’t often useful on its own because it does not describe why customers feel that way. What are customers specifically happy or unhappy about? In order to find out, businesses need to identify aspects that customers comment on. For example, in a hotel, the aspects may be ‘customer service’, ‘room size’, or ‘cleanliness’.
An example of how aspects and sentiment analysis categories can be used in a code frame
A deeper analysis can also find specific recurrent themes. For example, ‘not enough bread at breakfast’ or ‘room service is too slow’. These are important because they are easier to understand and act upon. Determining these themes is the holy grail of actionable insight. It will improve customer service effectively in a targeted manner.
The difficulty with both aspects and themes extraction lies in identifying the many different ways people may refer to the same thing. For example, when talking about ‘cleanliness of hotel rooms’ customers may be using words like ‘clean’, ‘tidy’, ‘dirty’, ‘dusty’ and phrases like ‘looked like a dump’, ‘could eat off the floor’. An effective customer sentiment analysis exercise should be capturing not just the sentiment, but also the aspects and the themes in each piece of customer feedback the business receives.
What are some use cases of customer sentiment analysis?
So you are interested in collecting sentiment what can you do with it? Here are 5 use cases that will enhance your customer experience strategy.
An example interface used by Google Shopping to show sentiment of aspects in customer reviews.
1. Target individuals to improve their service
By capturing customers who feel strongly negative towards your product or service, customer service can deal with their issues specifically. Imagine the fury of a customer who leaves a comment that’s -0.95 negative. Whether it is through personal contact or through prioritizing their tickets, action can help defuse the situation. If Dick’s Sporting Goods can identify those with real problems in the returns process, they can make sure customer service prioritizes them.
2. Track customer sentiment over time
Tracking customer sentiment attached to specific aspects of the business is more effective than tracking just the NPS. Analysis can explain why the NPS score has changed; or if the score hasn’t changed, what may have changed in the aspects. In the Dick’s Sporting Goods example above, the specific aspects worth tracking are ‘customer service’ and ‘return policy’.
3. Determine if particular customer segments feel more strongly about your company
When paired with demographic and other quantitative data it is possible to segment the customer base and look at their sentiment in isolation. For example, do customers who spend less feel more negatively (and therefore it is a barrier to them spending more) or are the return policy issues from customers in Miami and not those in New York?
4. Track how a change in product or service affects how customers feel
As the business changes so does the customer sentiment. Publishing a marketing campaign or press release, changing your product’s interface or price structure can have an effect. Tracking customer sentiment can measure this.
A change in score can indicate if a change has resonated with the customers emotionally and was successful. Tracking the sentiment will help you fix any blunders quickly.
If you know aspects and themes in each response, you can also answer questions like: For how long do people react negatively to a product price increase, or do they really love the new feature added? If Dick’s Sporting Goods releases a campaign focusing on competitive prices they would clearly see if people react well.
5. Determine your key promoters and detractors
Customers may be commenting on many aspects of your business, but which areas are affecting your NPS score? A little bit of data science will help you answer this question. By correlating aspects with promoters and detractors it is possible to show which influence the raw NPS score more.
Convinced? Here is how you can get started
Hopefully, by now, you are asking how you can apply sentiment analysis to your own customer feedback. The first and most important thing is to be collecting data. While there are a lot of companies that specialize in collecting customer feedback, even a simple thing like running surveys through Survey Monkey or using the built-in capabilities of customer service software like Zendesk can yield a lot of interesting data.
Next, comes the analysis: assigning sentiment scores and determining key aspects for each comment. You can do this by either by reading and coding the comments manually or by using software or products that focus on customer sentiment analysis.
Finally, use the results of the analysis to implement our five suggestions on how to improve your customer experience. Dashboards and reports can be used to get this new information in front of the people who need to see it.