NPS vs CSAT vs CES: Which Metric Should You Actually Track?

NPS, CSAT and CES all claim to measure how customers feel, and vendors selling each one will tell you theirs is the one that predicts loyalty. They're measuring genuinely different things, and the difference isn't academic — pick the wrong one and you'll spend a year optimising a number that has no relationship to the problem you actually have.

This is the head-to-head: what each metric asks, what the evidence says it predicts, where each one fails, and which to choose if you're only going to run one.

The three metrics at a glance

CSAT: did this specific thing go well?

CSAT is the most direct of the three. You ask how satisfied someone was with a defined thing — a repair, a delivery, a phone call — and report the percentage who chose the top two ratings.

Its strength is precision and speed. Because it's tied to one event, a low CSAT points at something specific: this technician, this route, this Tuesday. You can act on it the same week. It's also the easiest for customers to answer, which shows up in completion rates.

Its weakness is that satisfaction is a low bar. Customers routinely report being satisfied right up until they leave, because "satisfied" describes the absence of a problem rather than the presence of a reason to stay. A high CSAT tells you that you're not annoying people. It doesn't tell you they'd choose you again.

CES: how hard did they have to work?

CES asks customers to rate the effort a task required, usually by agreeing or disagreeing with a statement like "the company made it easy for me to handle my issue".

It has the strongest research behind it in one specific setting. A team at the Corporate Executive Board studied more than 75,000 customer interactions and published the results in Harvard Business Review in 2010 under the title "Stop Trying to Delight Your Customers". They compared CSAT, NPS and their new effort measure, and found effort was the better predictor of loyalty in service interactions — delighting customers didn't build loyalty nearly as reliably as removing friction reduced disloyalty.

CES also has a practical weakness: there is no single standard version. Scale lengths and question wording vary between vendors, so a CES of 5.2 is not comparable with anyone else's. Treat it as an internal diagnostic, not a benchmark.

NPS: how strong is the relationship?

NPS asks how likely a customer is to recommend you, on a 0-10 scale, and subtracts the percentage of detractors (0-6) from the percentage of promoters (9-10). It came out of Frederick Reichheld's 2003 Harvard Business Review article arguing that a single recommendation question predicted growth better than the long satisfaction surveys of the time.

Its real strength is that everyone knows it. It's the only one of the three with widely published industry benchmarks, and it's the metric a board or an investor will ask for by name. As a periodic measure of whether the relationship is getting better or worse, it does its job.

Its weaknesses are structural. Passives — the 7s and 8s — are excluded from the calculation entirely, so a genuine improvement from a 3 to a 6 shows up as no change at all. The score is also noisy at small sample sizes: with 40 responses, two people changing their mind moves your headline number by five points. And because it asks about the relationship in general, a bad score tells you something is wrong without giving you the faintest idea what.

Which should you actually track?

Match the metric to the decision, not to what sounds most sophisticated.

• You want to catch bad jobs before customers churn quietly. CSAT, fired after every job. It's specific, fast, and easy to answer.
• Your support queue or returns process is the problem. CES, on that process only. This is the setting the research actually covers.
• You need a number to track quarterly and report upwards. NPS, with the follow-up question, run on a fixed cadence.
• You have fewer than about 30 responses a quarter. None of them. Read what customers write, talk to them directly, and introduce a metric when you have the volume to make it stable.

What all three miss

Every one of these metrics shares the same blind spot: they only hear from people who answer surveys. That group skews towards the delighted and the furious, and away from the quietly disappointed majority who simply don't come back.

Two sources fill the gap, and you already have both. Behaviour — retention, repeat purchase, reopen rates — is harder evidence than any stated intention, because it costs the customer something. And public reviews are unprompted, specific and written for other customers rather than for you, which makes them considerably more candid than a survey response.

If you want the full picture of how to combine survey scores with the data you already hold, we've covered building the programme in how to measure customer satisfaction.

Keeping the review side of that running is what Dinopix Reviews does — Google and Facebook reviews in one inbox, sentiment analysed so themes surface without manual tagging, and requests sent automatically after each job so you're hearing from more than just the extremes.

Sources

This guide draws on the primary sources below. Research and definitions change — always check the current version.

• Dixon, Freeman & Toman, "Stop Trying to Delight Your Customers" — Harvard Business Review, July-August 2010
• Frederick F. Reichheld, "The One Number You Need to Grow" — Harvard Business Review, December 2003
• Measuring Your Net Promoter Score — Bain & Company, Net Promoter System