How reliable is the relationship between NPS and revenue?by
The value of NPS is a hotly-debated topic, in part because very little research is available on the relationship between NPS (or indeed any satisfaction metric trends) and revenue trends. However, once you cut through the noise, some evidence is available.
Right from the very beginning, right from when Fred Reichheld published his HBR article The One Number You Need to Grow, the subject of the relationship between NPS and revenue has been a controversial one. Some of the debate is sincere. Some of it is simple marketing work created by people who want to promote a different measurement and improvement system.
It is striking that the controversy has been about the number, and not about the validity of the answers to the open text questions that follow the NPS rating. After all, whatever their validity, numbers like NPS, CSAT, Customer Effort and others are not all that important. What is important is what you do to improve.
Customer suggestions for improvements are contained in their answers to the ‘Why?’ and ‘What could we do better?’ questions. I have yet to see any debate around them. Driver questions are used in other rating systems too, and are not covered here.
Meanwhile, debate about the statistical validity of the numbers is common. It prevents many companies from taking action. Unfortunate, to say the least!
There is NPS and there is NPS
The Net Promoter Score is used in a variety of situations, with differing predictive value. Let me state this clearly: for survey-based NPS, overall brand-level NPS and relationship NPS values and trends are the best predictors of revenue or market share. Forget at your peril. Product-level NPS does have equivalent value in single-product companies. That's not so clear for companies that provide multiple products and services as customers may mainly come into contact with people and processes that are not specific to a single product.
Let me give a clear example of a non-relationship: the NPS numbers and trends for your support calls have no predictive value whatsoever for your overall business. None.
Well, I suppose there is one exception. There is what I call ‘Event NPS’. The event in question is the main interaction a customer will have with a business. In the case of an insurance business, an event might be the end-to-end process for a client to find out how to log a claim, log it, interact with the insurance company, and receive payment.
A more common example is for an e-commerce company. The process that goes from searching for or learning about a product, to comparing it with others, selecting it for ordering, confirming the order and (in most cases) paying by credit card, is an ordering ‘event’. In ecommerce, the event represents 70% to 80% of the interaction that we have with the company. Ecommerce event NPS tends to have good predictive value for market share.
In fact, event NPS is not entirely an exception to the rule about transactions. When a group of sequential transactions (an event) represent a large portion of a customer’s interactions with a company, the NPS results naturally come to resemble brand NPS.
The situation is evolving
Predictive NPS is a new category of NPS that did not exist at the time NPS was invented. AI techniques are applied to operational and other data to predict overall brand-level (and therefore market share-predictive) NPS in real time for 100% of customers.
Early client implementations of OCX Cognition's Spectrum AI solution have been providing surprisingly accurate predictions. Other solutions exist too. It's simply a new type of NPS, though with the same intent as survey-based NPS.
Industries vary and improvements are relative
Some market trends cannot be surmounted. If your entire industry is in a 20% annual decline, then you cannot expect that a positive NPS trend will bring sustainable growth. Remember too that the real importance of an NPS value is not absolute, but relative. If your overall brand NPS improves but your main competitor’s improves more, you will lose market share.
Other things matter too
Bain and Satmetrix’s experience suggests that NPS trends explain 20% to 60% of market share trends. While such predictive power is impressive, many other things matter too.
For example, your company may launch innovative, impactful products which – you would hope – affect your market share. Or you may be in an industry where customer satisfaction does not matter much. Retail gas stations are a good example. Their revenue trends are almost exclusively driven by the price of oil. Not much else matters.
So… onwards to the current state of research on the topic
Very little academic research is available on the relationship between any satisfaction metric trends and revenue trends. This is not surprising. After all, most double-blind brand-level NPS research is funded by companies that consider the results highly confidential. However, some articles are available.
Andrew Stephenson, Jana Fiserova, Geoff Pugh, and Chris Dimos won a ‘best paper’ award for their study, published in the Annual Proceedings journal of the British Academy of Management. They studied the NPS and revenue performance of individual stores in Britain’s largest retail furniture chain, BFS.
Customers provided feedback six months after purchase – excellent timing for overall brand-level feedback for that type of product. And yes, the stores with better NPS trends outperformed the others by a significant margin. The paper is here. (This link should open with a completed search for the name Stephenson, and you can download the paper directly from the search result.)
I have also published slightly anonymised versions of what HP found when studying NPS. Yes, there is a relationship between NPS and revenue trends. One important conclusion was that there is a time lag between changes in NPS (compared to competitors) and changes in revenue. The time lag varies depending on customer purchase cycles. Find one of my articles on the subject on LinkedIn here.
Aggressive contradictory views
You know you’re winning when people hate you. OK, agreed, it’s not a sure-fire measure of success. But you do have their attention.
There are individuals and companies who have been somewhat aggressive in pushing contrary views, trying to imply that NPS trends have little or no value. I have seen two main categories of such positions:
- People who want to promote a compound metric which they have invented. This is the most common objection I have seen. I accept that compound metrics assembled from responses to three or more questions may have better predictive value than NPS on its own. Neither I nor Fred Reichheld (author of the original HBR article) have said anything else. The objection I have to such indices is that they are much harder than NPS to implement and to communicate. Try starting every discussion about results with a lengthy explanation of how you derived the score. Watch your audience sleep. I strongly recommend avoiding all compound metrics.
- People whose research is defective or who misrepresent NPS. It is easy to find the 2006 study by Morgan and Rego about the value of different customer satisfaction and loyalty scores. The authors themselves say that they studied NPS and that a different metric is a better predictor of business performance. Their statement is not entirely truthful. They did not study the “How likely are you to recommend…” question at all! They used a different question as a proxy for it, saying they believe it shows the same thing.
This is just one example of misrepresentation. Another peer-reviewed paper covered various industries in Norway. The authors state that NPS trends did not predict revenue over the four-year period during which three industries were tracked. Retail gas stations were one of the industries. The price of oil dropped 20% during the period, invalidating their conclusion.
Comprehensive list of studies
If you are interested in studying the subject in more detail, there is a way to get off to a good start. Jeff Sauro, PhD, has provided an excellent summary of articles that support or attempt to support varying positions on the subject.
It is on his Measuring U website here.
In addition to the original research by Bain and Satmetrix, there is at least some credible peer-reviewed and other literature supporting the relationship between NPS trends relative to competitors and market share trends. Company confidentiality rules mean it is not surprising that huge volumes of such data and research are not publicly available.
I have not seen any contradictory research that stands up to scrutiny. I do agree that complex, difficult-to-communicate compound metrics with better predictive power may exist.
Please allow me to say once again that the NPS numbers are not the most important part of a CX measurement and improvement system. What matters are the actions that you take, based on customers’ improvement suggestions. Those actions drive financial results.
This article adapted from a piece that originally appeared on LinkedIn.