How Many People Actually Recommend?

Jeff Sauro, PhD • Jim Lewis, PhD

blog feature image with people talking around table with smartphonePurchases and referrals are two closely tracked company metrics. If you can predict the extent to which people will purchase and recommend, then you will have some idea about future revenue and growth. Much of that forecasting is done through surveys using behavioral intention questions.

The behavior for purchasing is, of course, actually buying something. The behavior for recommendations is recommending (for example, verbally in conversation or via text), and the intention is what someone states they will do. So, how do you measure the intention of a future behavior? You ask people about their intentions.

But if people say they’re going to recommend a product or company to someone, will they actually recommend it?

In this article, we review data on how well measures of intention to recommend match future recommendation behavior.

Measuring Recommendation Intention

While you can certainly ask whether people will recommend a product, experience, or service using a yes or no question, such binary measures limit your ability to differentiate the fair from the fanatical. By using a multi-point scale, you can better differentiate the intensity of the intention.

One of the more (in)famous scales used to measure intent to recommend is, of course, the Net Promoter Score (NPS). It’s derived from the single eleven-point likelihood to recommend (LTR) question (e.g., “How likely are you to recommend this <company/website/product> to a friend or colleague?”), which has response options from “0: Not at all likely” to “10: Extremely likely.”

In the original NPS research, respondents who provide a top-two-box rating of 9 or 10 are categorized as promoters (people with a strong intention to recommend).

It’s understandable that those with a tepid maybe or probably attitude won’t necessarily follow through. But what about those who very confidently state that they will? The definite yes? The extremely likely crowd?

Likelihood to recommend is slightly different from intention to recommend, but given our earlier work on subtle differences in the wording of this item (e.g., “will” versus “would”), we suspect any differences would be small (although potentially worthy of additional research).

Published Data on Recommendation Follow-through

How likely people are to follow through on their recommendation intention is a question that’s been asked for some time, but there’s scant data in the published literature on how accurate such measures are. The reason for the dearth of data is that it’s hard to record the recommendations. Unlike with verifying purchases, which have more of a paper trail (receipts and sales records), for estimating recommendation behavior, researchers usually must rely on self-reporting to estimate recommendation behavior (although there are some ways to track referrals when there’s an incentive associated).

The next methodological challenge is something called common method bias. If you ask someone whether they are likely to recommend AND if they already recommended, these two measures collected at the same time may be affected by the current attitude (usually positive). That is, it could be that people who have a favorable attitude are more likely to report both that they will recommend and that they have already recommended. To reduce the impact of common method bias, a longitudinal approach can be used where intention is asked at one point and then recommendation behavior is measured at a future time point.

We found four sources that analyzed longitudinal recommendation data—two from the marketing literature and two that we published. The first was an article published in the Harvard Business Review (Kumar et al., 2007; not peer reviewed). Kumar and colleagues worked with two companies (financial services, telecom) that had programs allowing them to track which customers recommended their service to others—actual referral behavior rather than a self-report.

In their first study of 6,700 of the financial service’s customers, 68% expressed their intention to refer the company to other people. Within the next year, 33% of the sample (just under half of those who said they would recommend) followed through. They also reported a similar analysis of a telecom company (n = 9,900) where 81% of clients surveyed said they’d recommend the company but only 30% (just over a third of those who said they would) actually did. In a second study using the same participants, they described successful incentive campaigns to increase customer referrals (like the successful referral campaigns conducted by PayPal and Dropbox)—but for this article, we focused on the percentages of intention and recommendation without financial incentives to recommend.

The Kumar et al. (2007) study is unclear how participants were asked whether they would recommend (we emailed this inquiry to the authors but received no response). This is important because if participants in the study were simply asked a yes/no question, then the intensity of the behavioral intention to recommend could run the gamut from weak to intense—essentially equivalent to responses from 7 to 10 on the LTR item. On the other hand, if the customers designated as likely to recommend were those who selected 9 or 10 on the eleven-point LTR item, then their response intensity would be known to be high. Given the scant description in the article (“we polled a set of their customers … on their referral intentions and then tracked their behavior”), it seems likely that the intention to recommend in this research was established with a yes/no question.

In both cases, fewer than half but more than a third of the participants who reported an intention to recommend followed through. You could look at this as a glass half empty because most customers who said they would recommend didn’t … but from a glass half full perspective, at least a third of those who said they would recommend ultimately did. And remember, these were tracked behaviors—not self-reports.

The next source we examined was a peer-reviewed paper by Romaniuk et al. (2011), where the authors studied how many people followed through on TV show recommendations (something highly affected by word of mouth) and provided detailed metrics. Participants in the study (n = 235) were asked how likely they would be to recommend using a verbally administered probability scale similar to the eleven-point Juster scale instead of the likelihood anchors typically used for LTR. Because LTR and Juster scale formats differ in ways that tend to be more cosmetic than substantial, (e.g., full versus endpoint labeling of response options), the format differences likely have little effect on respondent behavior.

Participants who provided ratings from 0 to 4 were classified as non-intenders; those with ratings from 7 to 10 were classified as intenders (equivalent to a combination of the NPS classifications of neutrals and promoters). One week later, 138 participants who had agreed to be recontacted were interviewed and asked whether they had recommended the show. Note that the sample sizes in this study, while adequate to detect large effects, were relatively small when broken down by response options. For example, there were only 16 people who selected 10 and only 47 were intenders.

Romaniuk et al. reported about half (53%) of those indicating the strongest intention to recommend (10 out of 10 on their scale) reported following through and recommending the TV show. More broadly, 30% of intenders reported recommending while 95% of non-intenders did not recommend. This suggests that estimates of future non-recommendations by non-intenders are more accurate than estimates of future recommendations by intenders. This is reasonable because the path from non-intention to non-recommendation requires the participant to do nothing in contrast to the path from intention to recommendation, which requires participants to encounter an opportunity to recommend and then exert the effort to recommend.

Next, we turn to two longitudinal studies we conducted. The first was a longitudinal analysis of intent to recommend. In 2018, we asked over 1,100 participants from an online U.S. panel to rate their likelihood to recommend:

  1. Several common brands (e.g., Apple, Best Buy, Target)
  2. Their most recent purchase (specified by the participant)
  3. Their most recently recommended company/product (specified by the participant)

For example, participants reported having recently recommended or purchased from companies like Kroger (online and in-person), Steam, Shoes.com, and Sportsman’s Warehouse.

We then invited all respondents to a follow-up survey conducted 30, 60, or 90 days later (the full sample was about evenly divided across the time periods). Aggregating across the 90 days, around a third of all respondents reported recommending a brand or product. The higher the LTR rating, the larger the percentage of people who reported recommending. For example, across the ways of assessing recommendation intention, the percentages of promoters who reported making a recommendation in the lookback period were 54%, 66%, and 42% for, respectively, common brands, most recent purchase, and most recently recommended (see Table 1 below). The respective recommendation percentages for the most extreme responders (LTR rating of 10) were 53%, 71%, and 73%. One notable deviation from expectation in this analysis was a substantial percentage of those who rated themselves as least likely to recommend reported having recommended their recent purchase (non-intenders: 40%, detractors: 39%, passives: 56%).

The final source is our longitudinal study of online grocery shopping behavior. In early 2022, we surveyed 390 existing users of eight U.S.-based online grocery services. Respondents used the LTR item to rate how likely they’d be to recommend the service they used most frequently. In follow-up surveys distributed between 30 and 60 days later, we asked participants whether they had recommended the previously rated grocery service to anyone in the last month, receiving responses from 320 (82%) of the original participants. Over half (58%) of promoters (fewer than 20% of detractors/non-intenders) reported recommending.

Table 1 summarizes the results from these four studies (Sauro, 2019; Sauro & Lewis, 2022; Romaniuk et al., 2011; Kumar et al., 2007). While there is variation among the studies, there are some clear patterns when averaged. Most (63%) of the people who were the most likely to recommend (the 10s) reported recommending (see the column “10s” in Table 1). As the intensity to recommend decreases, so too does the reported recommendation rate, with 18% of detractors (0–6) and 16% of non-intenders (0–4) reporting recommending.

SourceDetractorPassivePromoter10s7–10s (Intender)0–4s (Non-Intender)
Sauro (2019): Common Brands12%28%54%53%41%10%
Sauro (2019): Most Recent Purchase39%56%66%71%63%40%
Sauro (2019): Most Recently Recommended 6%24%42%73%36%10%
Sauro & Lewis (2022): Online Grocery14%43%58%65%52%16%
Romaniuk et al. (2011): TV Programs53%30% 5%
Kumar et al. (2007): Financial49%
Kumar et al. (2007): Telecom36%
Average18%38%55%63%44%16%

Table 1: Percentage of respondents in various likelihood-to-recommend categories who later reported recommending or providing positive word of mouth (self-reports except Kumar et al.; data across recommendation measures not fully independent in Sauro, 2019; small differences between percentages are not statistically or practically significant).

We found three other sources that associated recommendation intention with self-reported recommendation but, for various reasons, didn’t include them in Table 1. The first two were De Run and Ling’s (2006) estimation of the probability of service recovery strategies in fast food restaurants and Sauro’s (2019) “did” versus “will” recommend analysis. Both sources have the potential for common method bias (past recommendation asked at the same time as likelihood to recommend—not longitudinal studies). De Run and Ling estimated the share of all recommendations from promoters in the same survey, but they did not report what percentage of promoters recommended. In our 2019 analysis, we also asked about past recommendation behavior and intention to recommend in the same study and found between 64% and 69% of promoters reported recommending—which is on the higher end of the other sources reported in Table 1 (although not substantially different).

The third source not included was a paper by Burnham and Leary (2018) that described a longitudinal study using an online panel asking about recommendation intention. Three weeks after the initial survey, they followed up with their participants to assess (1) whether participants provided recommendations (referred to as positive word of mouth) and (2) the extent to which participants had an opportunity to recommend during the lookback period. The analytical focus of the paper was on structural equation models estimating the influence of various predictors on reported recommendations, so the authors did not provide recommendation rates or percentages of responses to the LTR item. Their model showed that recommendation behavior depends not only on the likelihood to recommend but also on opportunities to recommend, providing some explanation for why the percentage of actual recommendations over a specified time will be lower than initial ratings of likelihood to recommend.

Discussion and Summary

Aggregating across four longitudinal data sources (two external publications and two of our datasets) we found:

Those most likely to recommend mostly recommend. Across the studies that reported this data, about 60% of people who selected the most extreme response option (10 on a 0–10-point scale) reported recommending. A similar percentage (55%) of promoters (those selecting an LTR of 9 or 10) also reported recommending. For extreme responders (who selected a 10) and promoters (who selected 9 or 10), the estimates of recommendation behavior ranged from 42% to 73% (only one estimate was less than 50%). It looks like between 50% and 60% of promoters ultimately make recommendations. We can also see that promoters were about three times as likely to recommend than detractors (55% vs. 18%).

Those expressing any intention to recommend have a surprisingly high rate of recommendation behavior. Across the four studies and seven estimates of recommendation behavior, 44% of participants classified as intenders recommended. Intenders were defined as the combination of NPS passives and promoters indicated by selecting from 7 to 10 on a 0–10-point recommendation scale or yes on a yes/no likely-to-recommend question. This recommendation rate is less than the estimates for extreme responders but still substantial. The recommendation rates across all seven estimates ranged from 30% to 63%. Even when the strength of the intention to recommend is diffuse, it’s reasonable to expect the percentage of recommendations from intenders to be less than half but more than a third.

Recommendation rates were comparable for tracked versus self-reported. An interesting secondary finding from aggregating this data was that the average tracked recommendation rate of 44% for intenders (Kumar et al., 2007) was the same as the self-reported recommendation rate averaged across the other three studies.

Those least likely to recommend generally don’t. Across these studies, there were two ways to identify those who were unlikely to recommend. One was the NPS designation of detractor (those selecting 0–6 on the LTR item) and the other was non-intenders (those selecting 0–4 on the LTR or Juster items). The average over four estimates of recommendation rates from detractors was 18% and from five non-intender estimates was 16%.

Recommendation rates for recent purchases were unexpectedly high. One context of recommendation explored in Sauro (2019) was for recent purchases. Even for detractors and non-intenders, about 40% reported recommending their most recent purchase. For now, we can only speculate about potential causes for this spike in recommendations. It’s possible that, having made a purchase, the attitude toward that purchase is affected by the resolution of cognitive dissonance over time (e.g., “Even though I don’t usually make recommendations, in this case, I must have made a good purchase decision so when the opportunity presented itself, I recommended the product”).

It isn’t realistic to expect recommendation behavior to equal recommendation intention. Some researchers have negatively framed the key finding as “only” about a third to a half of those who express an intention to recommend follow through with a recommendation behavior. Even the lowest estimate of 30% from Romaniuk et al. (2011) is far from zero. Once someone expresses an intention to recommend, there are obstacles to actual recommendation including never having an opportunity to recommend or not having a strong enough intention to exert the effort to recommend (Burnham & Leary, 2018). A more positive framing of the key finding from these studies is that there is a strong relationship between the intensity of the intention to recommend and future recommendation behavior, so changes to the quality of UX and CX that affect the likelihood to recommend are important.

You might also be interested in
0
    0
    Your Cart
    Your cart is emptyReturn to Shop
    Scroll to Top