Text Analytics Enables You to Listen to Your Customers’ Stories, at Scale
Net Promoter Score (NPS) has become a favorite metric in all kinds of customer (as well as employee and other stakeholder) feedback analysis initiatives. Since 2003, when Fred Reichheld of Bain & Company demonstrated its correlation with business growth and profitability, organizations of all industries, geographies and sizes have adopted it as a key indicator of corporate health.
(Net Promoter Score is a service mark, of Bain & Company, Inc., Satmetrix Systems, Inc., and Fred Reichheld. )
As we all know, NPS is based on a single question (“What is your likelihood to recommend company X to a friend or colleague”) to which respondents can answer on a standardized scale from 0 to 10. Responses are ranked according to well-known categories:
To complement this numerical response, NPS surveys often include an open-ended question such as “What is the primary reason for your score?” which allows surveyors to obtain additional information about the causes and circumstances of the scores.
The Two Faces Of NPS
Although the success of NPS is undeniable, it also has its detractors. One reason is its insensitivity to different cultures, which may be more or less inclined to give a high score. But we could also say that NPS has two sides:
On the one hand, NPS is a good predictor of growth and profitability. Companies that improve their NPS tend to experience improved growth and profitability as a consequence.
But on the other hand, NPS is a lagging indicator of satisfaction and other respondent perceptions and opinions. It is as a result of negative (or positive) perceptions that NPS worsens (or improves). When we receive a low NPS we are looking in the rearview mirror at a past satisfaction problem. As some analysts argue, NPS is an outcome metric that indicates what customers do as a result of their experience with us. But NPS itself does not provide information about underlying causes and problems. So if we don’t complement the quantitative analysis of the NPS score with a more qualitative one that identifies the causes, motivations and perceptions of customers, we are undermining our analysis and missing the opportunity to make it more actionable and to make decisions about it.
How to Make NPS Actionable
So you have your NPS scores and you want to act on them. How to improve those drivers that reduce your NPS and foster those that increase it? Well, the first thing to do is to identify those drivers and understand them at a deep level because, otherwise, we are flying blind.
And for this, the numerical ratings are complemented by those open questions that try to elicit the “why”. Qualitative analysis complements the ratings to get insight into that “why” behind the “what”. Ideally, we should make individualized analyses of these responses for detractors and promoters and discover the factors that correlate most with scores of one or the other type, as well as detect the emotions aroused or the intentions they reveal.
But the analysis of these responses poses serious difficulties. Numerical scores can be summed, aggregated, averaged… but how to sum responses in free text? The obvious answer is to analyze them by hand, by human means; but manual analysis of these texts creates enormous problems even when dealing with relatively small volumes (and by the nature of NPS studies these volumes are never going to be absolutely small).
- Human analysis is expensive, slow and suffers from low availability (shifts, sickness…). As a consequence, when the volume of comments to be analyzed is high, sampling techniques are often used (e.g. focused on those responses with very negative ratings). This can lead us to miss important insights.
Despite our preconceived ideas, human analysis is not of high quality: due to the ambiguity of language, different analysts processing the same text may reach different (even contradictory) conclusions about its meaning. Even the same person, depending on the moment, his physical and psychological state and other factors, may view the same text in a totally different light.
Text Analytics to the Rescue
Text analytics can automate the process of understanding verbatims, converting each verb into a set of structured data representing its meaning. These technologies bring high speed and availability to the comment analysis process and the possibility of performing it in high volumes and at low cost. As a result, it is feasible to thoroughly analyze all verbatims and not miss any insights. And although these technologies introduce errors (human analysis, too), they do so with a constant, consistent bias (not random based on personal circumstances) and are easier to compensate for.
But to get the most value from NPS comments, it is not enough to do a superficial analysis of texts based on predefined entities and categories, which structure the comments from our exclusive point of view as a provider. It is essential to apply an approach that applies the customer’s point of view, which we will call Interaction – Perception – Intention:
- What is the cause of customer ratings? We must be able to uncover the elements of the customer’s experience that are the root cause of their rating. What happens during those experiences? Do they have problems with certain attributes of our products? Are there aspects of our service that they find deficient? Or, on the contrary, do they highlight positively some others? We can call these interaction metrics.
- How do these experiences make them feel, and do they elicit relevant emotional responses? Negative emotions carry a lot of weight in experience formation and tend to be more enduring and shared. In contrast, positive emotions strongly influence loyalty and repurchase. It is crucial to analyze sentiment and emotions because they have a strong correlation with customer behavior and company profitability. These are known as perception metrics.
What are they going to do about it? Are they going to take any meaningful action: buy more, recommend, criticize, change supplier? Are they going to follow a different customer journey? We can call these intention metrics.
Dare to discover
Many text analytics solutions apply a conservative approach, based on looking at comments according to predefined categories and topics that have been used to train the system. These may be general or specific models of an industry or company (its products, its attributes…) and the analysis tries to “confirm” the presence or absence of these predefined topics. The problem is that these models usually represent our “view of the world” as suppliers, not as customers. But what happens if an unexpected or completely new topic suddenly emerges? What if our restaurant customers, instead of talking about our dishes or service, talk about the hygiene of our restrooms? Or suddenly start criticizing our supermarket for the lack of anti-COVID measures?Our text analytics tools need to be able to discover the topic structure inherent in a collection of comments and the new and unexpected topics that emerge from the conversation. And all this, without the need for prior configuration and learning. Only in this way will we be able to interpret comments from the customers’ point of view, and not from our preconceived ideas.
Detect emotional bonds
Customers are not rational beings and the vast majority of their decisions are emotional. Studies show that customers who have an emotionally favorable bond with our brand are much more profitable, even more so than those who are simply satisfied. And unfavorable emotions have a great power to contaminate experiences and are correlated with detrimental behaviors such as criticizing or abandoning the brand.
That’s why it’s not enough to detect comments with positive, negative or neutral polarity, but a wide variety of emotions (love, hate, fear, skepticism…) each of which brings a different tonality. And this must be done not only at the level of the overall commentary, but also at the level of the topics discovered in it. Thus we can detect, for example, that customers specifically love the design of our products or hate the waiting time in our queues.
The intention stated in the comments is the best predictor of future customer behavior. It is vital to discern whether a customer will buy, recommend or stop using our products. And to do so in the context of the topics that have been discovered.
For example, if we detect that our new product model is strongly associated with recommendation intentions or that our support service is associated with churn intentions, we can implement specific programs and campaigns to encourage the first situation and avoid the second.
A learning platform
Numerical NPS scores are valuable as predictors of growth, but they do not allow us to understand what is behind the numbers and take specific actions.
As an example, let’s consider the case of an ecommerce business with a more than acceptable NPS score. The company’s management might be tempted to celebrate the results by resting on their laurels. However, detailed analysis of the comments may reveal widespread customer dissatisfaction with an important aspect such as delivery times, a circumstance that generates frustration and may lead to future churn. In fact, the survey is telling us that if corrective action is not taken, the indicators could be very negative in the coming period. Systematic analysis of the comments allows us to discover the correlation of the dissatisfaction drivers with the numerical scores of the customers and to analyze their evolution over time. In short, text analytics of verbatims covering Interactions, Perceptions and Intentions allows us to listen to customers’ stories at scale, understand the business from their perspective and add a qualitative component to the quantitative NPS study.
Where to go from here
So what should be your path towards NPS initiatives that constitute a learning platform? In our view you should try to apply feedback analytics with the following capabilities:
- Apply a discovery approach, which does not start from predefined models based on what you think is important and allows you to detect unexpected insights and emerging topics, to understand your business from your customers’ point of view.
- Detect emotions and intentions around those topics, to identify feelings and predict behaviors that allow you to take action.
- Systematize the analysis: develop models that you can recurrently apply to compare results from different studies, detect trends and evaluate improvements.
- Build a learning platform: share, communicate and apply insights.
Does your text analytics provider provide them? If you want to know how Gavagai can help you don’t hesitate to call us.