Keiningham and colleagues published three articles between 2007 and 2008. This is the first of the two 2007 articles, and he was cited in the 2019 WSJ article. In Keiningham et al. (2007a), the authors examined 8,000 customer responses for common metrics in retail banking, mass-merchant retail, and Internet service providers (ISPs) across two years. Customer retention and share-of-category spending were then analyzed in the second year.
The authors point out that many consultants and business managers frequently ignore or misunderstand the connection between beliefs/attitudes (i.e., satisfaction) and intentions. As a result, it is not uncommon to hear consultants and managers say something to the effect that they have gone “beyond” customer satisfaction to measuring customer loyalty, as in the books Customer Satisfaction Is Worthless. Customer Loyalty Is Priceless (Gitomer, 1998) and Beyond Customer Satisfaction to Customer Loyalty (Bhote, 1996). Reichheld’s Ultimate Question is another example of this.
They used four behaviors: change in share-of-wallet (i.e., SOWt2 SOWt21), SOW, customer retention, and customer recommendations. Their Net Promoter Score was based on a five-point fully labeled scale:
5 Definitely would recommend them
4 Probably would recommend them
3 Might or might not recommend them
2 Probably would not recommend them
1 Definitely would not recommend them
The Net Promoter Score was calculated as the 5s minus the 1–3s, with the 4s as passives. The average correlation for their NPS item and repurchase intention were higher than the overall satisfaction item. The average fisher-transformed correlation for the NPS items were r = ~.31 across three measures (compared to satisfaction r = ~.25), showing similar results to our comparison of NPS and satisfaction in 12 industries, where NPS was a nominally better predictor of future growth.
The authors reported that repurchase intention best predicted retention and recommend intention best predicted future recommendations. In other words, their Net Promoter Score was the best predictor of intent to recommend and retention.
The authors then examined how well a single-item measure compared to a multi-item measure, testing Reichheld’s claim that more items added “insignificant predictive advantage.”
Note, the authors in a later paper advocate for a single-item measure of satisfaction (converted to relative ranking; see Keiningham et al., 2015). In that article the authors stated “With regard to the use of single-item measures in general, although marketing academics typically prefer multi-item measures, single-item measures of overall satisfaction have been used in many prior studies and shown to perform adequately.”
The authors used an ordinal dependent (0: not retained; 1: retained but did not recommend; 2: retained and recommended). They show that there is a small increase in predictive ability. The difference between the single best predictor and the best model with multiple predictors varied by adding between .1 to 1.6% in creases in R2: 5.9 vs. 7.3 in banking; 8.7 vs. 10.3 in retail and 7.4 vs. 7.5 in ISP. The increase was statistically significant, but whether it is practically significant, or as Reichheld alludes, insignificantly predictive is a matter of opinion.
The authors concluded that recommend intention does provide insight into future recommend behavior but not as a sole indicator. They base this on NPS and all measures having modest correlations, that recommend intention was not ALWAYS the best predictor of their loyalty metrics (retention, share of wallet, and recommendations), and that multiple measures were better at predicting the retain and recommend variable. The authors are concerned about the potential misallocation of resources by focusing only on customers’ recommend intentions.
Takeaway: The authors found that a five-point version of the NPS was a better but comparable predictor than satisfaction on future recommendations, intent to purchase, and retention. NPS was not ALWAYS the best predictor of their loyalty metrics (retention, share of wallet, and recommendations), and multiple measures were slightly better at predicting the retain and recommend variable.