By Pedro Costa Ferreira
- Nowadays, Customer Experience is a key element of differentiation strategies in all industries, especially in mature markets with low opportunities for product differentiation.
- Net Promoter Score (NPS) metrics have been used by organizations of diverse kinds in some way to measure customer satisfaction and loyalty.
- Advanced Analytics disciplines can be used to deepen the understanding of the customer journey within a CX context.
Since its introduction as a concept, Consumer experience (CX) has evolved as a key element of differentiation strategy in all industries, especially in mature markets, which face a scenario of low opportunities for product differentiation. Traditional sectors, such as banking, retail, hospitality, telecommunications, as well as corporations in the modern digital world recognize the role of CX as a crucial part of their business success.
What is customer experience?
We can define CX as the subjective response that customers have by contacting direct or indirectly with a company and its products. It is closely related to customer loyalty management and is measured through different methodologies, like the application of metrics-based surveys with the aim of measuring overall customer satisfaction, purchase intent or recommendation tendency.
And, since the publication of Frederick Reichheld’s seminal article, The One Number You Need to Grow (2003), the Net Promoter Score (NPS) has also been used by organizations of distinct kinds in some way to measure customer satisfaction and loyalty.
The NPS is measured based on the customer’s response to their likelihood of recommending the company/product to other people. The scale runs from zero to ten. When a response is nine or ten, that customer is labelled a “promoter”; when they respond with a seven or eight, they are called “passive”; at the bottom of the scale, a “detractor”. The NPS is calculated from the percentage of promoters minus the percentage of detractors.
Despite being a simple and widely used metric, NPS is only a thermometer. When the score is low, it shows a “fever”, but a deeper understanding of the root causes of the disease is a need that this indicator does not fulfil. The problem is that measuring customer satisfaction doesn’t tell anyone how to achieve it.
In this article, we will explore the limitations of NPS and how Advanced Analytics skills can enable us to delve deeper into the specifics of this metric, use more complex approaches, produce richer assertive insights and, thus, target business actions more precisely.
How can Advanced Analytics be applied to CX?
Let’s consider a telecoms company as an example. The customer journey encompasses the main vectors that influence their satisfaction, such as signal quality (e.g. connection stability, coverage), product (e.g. cost-effectiveness, variety of applications that don’t deduct from the internet), billing (e.g. deadline for receiving the bill, bill with the correct amounts), top-up (e.g. top-up process), communication (e.g. communication from the operator via SMS or calls, clarity in billing).
However, from the customer’s point of view, their perceptions are not so well defined within each vector. The way the customer perceives the service is relatively abstract and changes considerably depending on individual user characteristics and external factors. Not all touchpoints have equivalent value.
Therefore, the consumer experience is multiple and fragmented, it does not depend only on the service, but mainly on how they perceive that service. Individual characteristics (age, gender, income and region) and exogenous factors (the state of the economy, climate and seasonality) have an important bearing on the NPS response.
Not taking all these peculiarities into account and trying to understand NPS only through bivariate analyses can lead to an incorrect diagnosis of the root causes. You should always remember Simpson’s Paradox, which shows that the trend or result that is present when the data is put into groups is reversed or disappears when the data is combined.
From the company’s perspective, the questions are different and always seek to understand the consumer experience in greater depth. The company is asking, for example, whether “road coverage is an important factor for the customer”. In this case, the challenge for the business is to ask the right questions and look for measurable proxy variables that make it possible to correctly infer consumer perceptions.
We realize that NPS is the result of a complex relationship of variables that influence each customer’s response, Understanding the main levers of this metric is an extensive analytical journey.
What skills are needed to deepen your understanding on customer journey?
Meeting the challenge requires multiple competences, analytical maturity and a first-rate multidisciplinary team.
You need to: (I) know the NPS journey and understand what can impact it qualitatively (e.g. design thinking exercises can help raise possible hypotheses about consumer behavior); (II) know the statistical limitations and apply machine learning algorithms to answer more accurately what the real “customer pain” is; (III) have a great deal of knowledge about the operation and structure a datalake (centralized repository of quality data) with important consumer information.
As statistician Edwards Deming would say: “In God we trust; All others must bring data”.
You also need to: (IV) know and have plenty of experience working with the proxy variables that help explain the feeling of the consumer experience (e.g. what are the measurable variables I can use in my model that help explain a particular consumer experience?); (V) know that there is a lack of clear and precise association between NPS and CX attribute scores.
A survey of the NPS modelling literature shows how difficult it is to model this variable. Statistical metrics such as accuracy, recall and precision are around 50 per cent in most published articles.
It is important to realize that (VI) the presence of biases in the NPS responses of the customer journey creates difficulties in fully understanding the growth levers. Problems in NPS research are widely discussed in the literature.
For example, the representativeness of the group of users who respond to NPS surveys compared to the entire base is one of several problems discussed in the associated literature.
Perhaps only a specific type of person is willing to fill in these forms, and this can make the results biased. Other problems are a lack of understanding of what is being asked (surveys that check the customer journey are lengthy and the questions are not always clear) and the wear and tear on users to answer lengthy questionnaires (respondent burden).
These skills enable us to use more complex approaches, generate increasingly rich and assertive insights about NPS and, above all, accurately target actions for the business. Advanced CX Analytics should be worked on three main fronts, each with its advantages and limitations and addressing different business questions. These are: (a) Measuring NPS and Business Indicators; (b) Studying NPS at Journey Level and (c) Modelling NPS at Customer Level.
How do the main fronts of Advanced CX Analytics work?
On the NPS Measurement and Business Indicators front, the Data Engineering team works to provide the data and other technical indicators to correlate with the NPS and the Analytics team generates insights for the business. With the large volume of information, billions of lines in the databases, big data technologies are applied, and Python and Spark are combined to explore the NPS results.
Again, although NPS is relatively easy to measure, it is still just a thermometer. Correlating it with business variables is not trivial, but it does generate significant insights for our partners. For example, it is possible to model, rank and priorities deployments with technology suppliers at the “best” sites, considering the variables with the greatest impact on NPS.
When we move on to the NPS Study at Journey Level, we do a more macro analysis and by correcting views we can process the results of relational NPS surveys and capture the “most aware” customers. The idea is simple, the execution not so much; but as we already know some of the problems with surveys, we work to clean up the “wrong” answers and get conscious respondents.
For example, if the individual answered 10 for the NPS and 0 for the other questions, that respondent didn’t understand the survey or did not answer in a committed way. Whatever the reason, the tendency is for these answers to only generate noise in the analysis and need to be discarded.
This effort is worthwhile because, as well as reducing bias, it reduces the high collinearity (two or more variables vary in a comparable way to each other) that exists between the variables in the relational survey, allowing predictive analysis between detractors and promoters using the attributes of the relational survey.
The insights are interesting and can be complemented by applying natural language processing (NLP) to customer comments on the partner company’s different channels, making the analysis more robust and the insights even more assertive.
Another challenge is modelling NPS at Customer Level. Understanding this general equilibrium problem and modelling it is far from a trivial task, but the results tend to be rewarding.
At this stage, you need to understand that the NPS score the customer gives does not depend solely on an action by the company, but rather on a number of factors, such as personal characteristics (e.g. gender, race, age), perception of all the company’s journeys (e.g. communication with the operator, clarity in billing) and socio-demographic characteristics (e.g. population of the municipality, HDI).
Also, external factors (economic situation, seasonality), latent variables (e.g. mood on the day you’re answering the survey), proxy variables that can make the connection between the customer’s feelings and the business (e.g. signal strength, number of websites trafficked) and the actions of competing companies (e.g. the customer can compare the signal level of their service provider with the signal of the competing company).
What are the results of applying Advanced CX Analytics?
The results are promising and allow for interesting analyses, such as simulating the NPS of the company’s entire base and in cities where there is no survey. It is also possible to infer which are the main offending factors in the NPS and calculate the net impact of actions, making investments more assertive.
It is important to make it clear that for ethical and data protection regulation reasons, the company will not use the results of the models to create actions for specific customer niches. The aim of modelling at customer level is to be able to extract the elasticities that give us the net effect of each action (e.g. improving the signal by 2 p.p. increases the NPS by 0.5 points) and, thus, be able to have more assertive insights and target business actions more precisely.
Implementing and giving cadence to the three approaches we have discussed in this article is a complex task and requires many stakeholders to be engaged with the potential results of the initiatives. Because it is so difficult to model and because it presents so many biases, results only appear with persistence and successive iterations.
It is a continuum of business hypothesis, improved measurement, denial or proof of the hypothesis by the data and then real-life experimentation to prove the theory that the data has pointed to. This is the only way to generate insights that really move the NPS needle.
Finally, it is worth remembering that, while customer experience remains an important differentiator for all companies, data and AI are increasingly moving to the center to ensure that the grand expectations of today’s connected customers are met.
Putting together the puzzle we have discussed throughout this article and understanding the root causes of the countless variations and specificities in the customer experience journey is a sine qua non for long-term survival. What about your company? How is Advanced CX Analytics?
PEDRO COSTA FERREIRA worked as Director of Data & Analytics at EloGroup.
