Key Takeaways
- The NPS vs CSAT vs CES debate has been tested, and the result surprises most guides. In a study of 93 firms, top-2-box satisfaction was the best single predictor of retention, NPS was close behind, and CES performed worst.
- That research used consumers. The Dutch study measured self-reported retention two years later, and its authors note they “only investigate business-to-consumer firms.”
- NPS's creator carved out B2B tech. For database software and computer systems, where executives choose vendors, other questions “were more predictive.”
- Each metric fits a different job. CEB said NPS is “a powerful gauge at the company level” while CES works in service interactions.
- Combining them works better than any one. The same study found “combining metrics, especially the CES with the customer satisfaction-related CFMs, results in improved out-of-sample retention predictions.”
- Ask each survey of the right person, never tie it to bonuses, and check every score against net revenue retention.
The quarterly board pack had three survey scores on one slide. NPS was 42. CSAT was 91%. CES was 2.3. A board member asked which one mattered. Carmen's head of support said CES, because the research said effort predicts loyalty. The CS lead said NPS, because it's the industry standard. The CFO asked why none of them had moved when two large accounts had just downgraded. Nobody had a good answer.
Most NPS vs CSAT vs CES guides give definitions, formulas and a table of pros and cons. They rarely cite the research that compared the three. This post covers what that research found, what it didn't test, and how a B2B customer operations team should use each score.
The three metrics in one line each
NPS asks “How likely is it that you would recommend our company to a friend or colleague?” on a 0 to 10 scale. Promoters score 9 or 10, detractors 0 to 6, and NPS is promoters minus detractors.
CSAT asks how satisfied the customer was, usually on a 1 to 5 scale. Research uses “top-2-box”: “the proportion of customers filling in the two highest-scoring points of the overall customer satisfaction scale.”
CES asks “How much effort did you personally have to put forth to handle your request?” The original scored it from 1, very low effort, to 5, very high effort.
What the research found when it compared them
Three independent studies are worth knowing.
Marketing Science, 2006. Using US satisfaction index data and company financials, the researchers found “average satisfaction scores have the greatest value in predicting future business performance.” Measures based on recommendation intentions had little or no predictive value in their data.
A 2007 replication. Testing NPS in the industries its creator had cited, the researchers found “no support for the claim that Net Promoter is the ‘single most reliable indicator of a company's ability to grow.’” The lead author worked for a rival research firm, which is worth knowing.
A Dutch study, 2015. This is the only study to test all three against retention. It surveyed “6649 respondents, who in total filled out 8924 firm evaluations for 93 firms across 18 industries,” and asked about retention two years later. The results:
| Metric | How it ranked | What the authors said |
|---|
| CSAT, top-2-box | Best single predictor | “the single-best predictor of retention across industries” |
| NPS | Close second | “an effective predictor of customer retention” |
| CES | Worst | “little to no predictive power” |
| Combined | Better than any single metric | “A dashboard of CFMs… is preferable to monitoring a single CFM” |
Source: International Journal of Research in Marketing, 2015.
There are important limits. The study used “customers' self-stated retention as a dependent variable,” not billing records. It covered only Dutch consumers. And the best metric varied by industry. In three of 18 industries, “none of the CFMs are significant.”
A later study added a twist for NPS. Testing seven US sportswear brands over five years, the researchers found “only the more recently developed brand health measure of NPS (using an all potential customer sample) is effective at predicting future sales growth.” In other words, NPS worked when it surveyed the whole market, including people who weren't customers. That's a different use from asking existing accounts.
Satisfaction has its own weakness. CEB found “twenty percent of the ‘satisfied’ customers in our study said they intended to leave the company in question; 28% of the ‘dissatisfied’ customers intended to stay.” Even the best predictor in the Dutch study leaves plenty unexplained.
What the creators said about their own metrics
Both NPS and CES came with limits their authors stated up front. They rarely make it into vendor guides.
One more detail is worth knowing. The 2003 HBR article that introduced NPS disclosed that its data came from Satmetrix, a firm whose board included the article's author. That doesn't make the findings wrong. It's a reason to weigh the independent replications carefully.
The article built NPS from “thousands of customers recruited from public lists in six industries”, all consumer-facing. The recommend question ranked first or second in 11 of the 14 case studies, not all of them. It also named the exception that matters most for B2B software: “In database software or computer systems, for instance, senior executives select vendors … In these cases, we found that the ‘sets the standard of excellence’ or ‘deserves your loyalty’ questions were more predictive.”
The CEB authors who introduced CES were equally specific. Compared with NPS and CSAT, they wrote, “NPS proved better (and has been shown to be a powerful gauge at the company level). CES outperformed both in customer service interactions.” CES was built for one moment: right after a service contact.
That explains the coverage problem in the Dutch data. Only 20.5% of the sample had contacted the company with a request, so the rest couldn't give a CES. A CES average describes, at most, the customers who needed help.
The B2B problems none of the studies solve
The person answering isn't the person renewing
In B2B software, an end user fills in the CSAT survey after a ticket. An admin answers the NPS email. A VP or procurement lead signs the renewal. The creator's own carve-out points at this. A score from users may say little about the buyer's decision.
Scores and behaviour can split
A Cambridge working paper studied one UK B2B service firm. It found “customer loyalty based on the NPS score classified customers into 70 per cent promoters”, while purchase records showed 46% of customers had churned. It's one firm and an unrefereed paper. It's still a warning about treating a survey score as a retention number.
Scores get gamed
NPS's creator now says so directly. A 2021 HBR article on NPS states “unaudited, self-reported Net Promoter Scores undermined the usefulness of NPS,” and names the main abuse: “linking Net Promoter Scores to bonuses for frontline employees.” The proposed fix ties the score back to revenue, using “a battle-tested statistic called net revenue retention (NRR).”
How to use NPS, CSAT and CES in a B2B team
| Metric | Ask it of | When | Use it for | Don't use it for |
|---|
| NPS | Decision-makers and admins | Twice a year, per account | Relationship trend and follow-up conversations | Agent or CSM bonuses |
| CSAT, top-2-box | The person who raised the request | After each resolved interaction | Service quality by team and topic | Account renewal forecasts |
| CES | The person who raised the request | After key service moments | Finding friction to fix | A headline KPI |
This table is our recommendation, based on the sources above.
In practice, most B2B teams will get the most from CSAT first. It covers the most interactions and ranked best in the comparative research. NPS earns its place once someone owns the follow-up with each detractor account. CES belongs on the few service moments where you suspect friction, such as onboarding tickets or billing questions.
Then check all three against the account's net revenue retention every quarter. If a score moves and revenue doesn't, or revenue moves and the score doesn't, the score isn't telling you what you think. Our post on customer success metrics that matter covers the revenue layer, and our guide to customer effort score covers CES in more depth.
Make every score checkable
Our post on AI customer service metrics sets a simple test for any number on a board slide: can someone else reproduce it? Survey scores often fail that test. Here are four checks that fix most of it.
- Publish the response rate and who answered. A score from 40 admins isn't the same as a score from 40 buyers. The 2021 article criticises reports where no details are given about which customers were surveyed or how many.
- Report the distribution, not only the average. Top-2-box for CSAT, and the share of promoters and detractors for NPS.
- Tie every response to an account. A score you can't connect to a customer can't be checked against what that customer did.
- Compare with behaviour each quarter. Plot scores against renewals, downgrades and expansion by account. Where they disagree, read the requests and conversations behind them.
None of this needs a new survey. It needs the survey answers, the account and the revenue in one place.
Scores on the same record as revenue
The board-slide question had no answer because the three scores lived in a survey tool, the downgrades lived in billing, and nobody could connect a detractor to an account. Checking a survey score against NRR needs both on the same record.
MatrixFlows keeps accounts, contacts, requests and survey responses in linked tables in Matrix. Forms can send CSAT or CES after a request closes and store the answer on the request and the account. Views can show an account's scores next to its renewal date and revenue. When a score and the revenue disagree, the team can see the requests behind it.
Back to the board pack. The two downgraded accounts had high CSAT on every ticket. Neither had answered an NPS survey in a year. The people who decided the renewal had never been asked.
Create a Free Workspace → and put survey scores next to account revenue this week.
.svg)
.svg)
.svg)
