Customer Effort Score (CES): measure effort, not delight

How to calculate the Customer Effort Score, which question and scale fit and how precise your result is. With a free CES calculator, example and template.

The Customer Effort Score (CES) measures how easy it was for customers to resolve their issue. In the Gartner variant this page uses, respondents rate their agreement with a statement such as "The company made it easy for me to handle my issue." on a scale from 1 to 7. The CES is the share of respondents who chose 5, 6 or 7 (Gartner 2018):

CES = valid answers of 5 to 7 / all valid answers × 100%

Some teams calculate the mean of the seven points instead. I'll show you both variants below with an example, and the calculator handles both.

CES calculator

Further down, I explain how the calculator uses the number of answers per point to work out the CES, its interval and the comparison of two measurements.

The example file shows how to arrange your data: Excel, CSV. It contains the 160 answers from the hotline example below and also calculates the share of 5 to 7 and the mean, each with its 95% confidence interval, directly in Excel.

Under "Variant", choose "Share of 5 to 7" or "Mean 1 to 7". For the original five-point question from the HBR article, the calculator also offers "HBR 1 to 5". The calculator runs in your browser and doesn't send any data.

Where the Customer Effort Score comes from

In their 2010 Harvard Business Review article "Stop Trying to Delight Your Customers", Matthew Dixon, Karen Freeman, and Nicholas Toman revealed a counterintuitive finding: customers punish a company for poor service far more than they reward it for exceptionally good service. Trying to exceed expectations is disproportionately costly and yields comparatively little. What matters more is making the path to resolving the customer's issue as easy as possible.

Out of this came the Customer Effort Score (CES), a measure of the perceived ease of resolving a problem.

How CES is measured

In the original HBR article, CES was captured with a direct effort question: "How much effort did you personally have to put forth to handle your request?" The response scale runs from 1 (very little) to 5 (very much). On this scale, a low value means little effort.

Gartner describes the variant used here as an agreement question: "Please indicate the extent to which you agree with the following statement: The company made it easy for me to handle my issue."

The response uses a seven-point scale:

1. Strongly disagree
2. Disagree
3. Somewhat disagree
4. Neither agree nor disagree
5. Somewhat agree
6. Agree
7. Strongly agree

The CES is the percentage of respondents who answered with at least 5 ("somewhat agree" and above), the top-3-box.

Valid answers here are responses on the agreement scale from 1 to 7. Missing answers, "don't know" and values outside this scale are not included and are reported separately. The same applies to the mean. Without valid answers, no CES can be calculated.

The score ranges between 0% and 100%. A higher score means customers found it easier to resolve their issue.

How to word the question

Here are the seven points again, with the numbers they're coded with in the questionnaire:

The scale is symmetric: three points of disagreement, three points of agreement and "Neither agree nor disagree" in the middle. Label every point, not just the two ends. Then respondents know what a 5 means, and so do you when you analyze the results.

You can adapt the statement to the touchpoint, for example "The company made it easy for me to handle my return." After that, though, don't change the wording again. Otherwise, at the next measurement, you're comparing a different question.

And what if you work with the original question from the HBR article, asking about effort from 1 (very little) to 5 (very much)? Then the direction is reversed, and a low value is good. Results from this question can't be compared with either the share of 5 to 7 or the mean of the seven-point scale. Decide on one variant and name it in every report.

What the numbers say

Dixon et al. (2010) reported the following relationships with behavioral intentions for the original effort measure:

• 94% of customers with low resolution effort said they intended to buy from the company again.
• 88% said they intended to increase their spending on the company's products and services.
• 81% of customers who had difficulty getting their issue handled, by contrast, planned to spread negative word of mouth.

The finding suggests that investments in reducing friction (self-service paths, a clear FAQ architecture, short hotline wait times, consistent information across channels) pay off more for loyalty than investments in delight initiatives.

When to use CES

CES is especially well suited after service interactions of a problem-solving nature: hotline calls, support tickets, complaint handling, self-service sessions. For pure purchase touchpoints (first purchase, checkout), CSAT is usually more appropriate. For assessing the long-term relationship, NPS is a useful complement.

Practical recommendation

Use the CES question right after service interactions, ideally within 24 hours, while the memory of the specific effort is still fresh. For answers below 5, use a filter to show an open follow-up question: "What made it difficult?" This yields the most concrete ideas for improvement.

For an overview of using NPS, CSAT, and CES together, see the page NPS, CSAT, and CES in QUESTIONSTAR.

How to calculate the CES: formula and example

There are two ways to calculate the CES: the share of answers from 5 to 7, as described above, and the mean. I'll show you both, first with a very small example and then with a realistic one.

A very small example

Let's look at ten customers who rated the CES statement after returning a product. Their answers: 2, 3, 4, 5, 5, 5, 6, 6, 6 and 7.

• Share: Seven out of ten chose 5 or higher. CES = 7 / 10 = 70%.
• Mean: The sum of all answers is 49. Mean = 49 / 10 = 4.9.

Both numbers describe the same ten answers, just in different ways. The share only asks how many agreed. The mean also takes into account how strongly. If the three customers who gave a 5 gave a 7 next time, the share would stay at 70%. The mean would rise to 5.5.

Worked example: 160 answers

An internet service provider asks for the CES after every call to its hotline. In one month, it collects 160 answers (made-up numbers):

Variant 1, the share.

Step 1, count ratings of 5, 6 or 7: 40 + 50 + 30 = 120 answers.

Step 2, divide by all answers: 120 / 160 = 0.75. The CES is 75%.

The 20 disagreeing answers (4 + 6 + 10) and the 20 in the middle don't count toward the CES, but they are in the denominator. A common mistake is to leave out the midpoint. Then you'd get 120 / 140 ≈ 85.7%, almost eleven percentage points too high, just because the denominator is wrong.

Variant 2, the mean.

Mean = sum of (point × number of answers) / all answers

Step 1, multiply each point by its number of answers and add everything up: 1 · 4 + 2 · 6 + 3 · 10 + 4 · 20 + 5 · 40 + 6 · 50 + 7 · 30 = 4 + 12 + 30 + 80 + 200 + 300 + 210 = 836.

Step 2, divide by the number of answers: 836 / 160 = 5.225, rounded ≈ 5.23.

Share or mean?

In the Gartner variant used on this page, CES is calculated as a share. It's easy to explain: "Three out of four callers found it easy." The mean, on the other hand, also responds to changes within the agreement range, as the small example shows. However, it assumes that the distances between the points are equal, that is, that the step from "Neither agree nor disagree" to "Somewhat agree" is just as large as the step from "Agree" to "Strongly agree". That's an assumption, not a property of the scale.

If you're unsure, use the share. Whichever you choose, stick with it and state which variant you mean. A CES of 75% and a CES of 5.23 are two different metrics.

In QUESTIONSTAR

By the way, in QUESTIONSTAR you calculate the CES in its share variant with the NPS question and your own thresholds. Set up the question with seven points, coded from 1 to 7, and set points 5 to 7 to count as promoters. The share of promoters in the report is then your CES. Make sure you read the share there and not the Net Score. The Net Score subtracts the share of the bottom group from it and makes no sense for the CES.

This route doesn't give you the mean. For that, enter the number of answers per point from the report in the calculator above under "Enter counts". Or export the raw data as CSV, XLSX or SPSS and calculate in Excel, SPSS or R. For both variants, the calculator gives you the interval and the comparison of two measurements, which the report doesn't show.

How precise is your CES?

The internet service provider measured a CES of 75%. For the following interval calculations, we assume independent random observations with one answer per person. When comparing the two months, we assume two independent samples. If the same people are surveyed again, the comparison needs an analysis for paired answers. With a different sample of 160 callers, the result might have been 71% or 78%. The 95% interval shows how far a result can vary just because of who happened to be selected. The 95% describes the procedure: if you repeatedly drew random samples and calculated an interval for each one, about 95 out of 100 intervals would contain the true value. If mainly the most annoyed customers respond, though, the interval doesn't help. I explain the idea in more detail in the article on the Net Promoter Score.

The interval for the share

First, the simple calculation. The standard error of a share is:

Standard error = √(share · (1 − share) / number of answers)

In the example: √(0.75 · 0.25 / 160) ≈ 0.0342, so ≈ 3.4 percentage points. Multiplied by 1.96, that's ≈ 6.7 percentage points. The interval therefore runs from ≈ 68.3% to ≈ 81.7%.

The calculator uses the Wilson interval, a standard method for shares (NIST/SEMATECH n.d., section 7.2.4.1). Unlike the simple calculation, it always stays between 0 and 100%, even with few answers or with shares close to 0 or 100%. The formula looks harder than it is. With n = 160 answers, the share p = 0.75 and z = 1.96:

Center = (p + z² / (2n)) / (1 + z² / n) = (0.75 + 0.0120) / 1.0240 ≈ 0.7441

Half-width = z / (1 + z² / n) · √(p · (1 − p) / n + z² / (4n²)) = 1.96 / 1.0240 · √(0.001172 + 0.000038) ≈ 0.0666

The interval runs from 0.7441 − 0.0666 ≈ 0.678 to 0.7441 + 0.0666 ≈ 0.811, so from ≈ 67.8% to ≈ 81.1%. The center has moved slightly toward 50%. With 160 answers, the two calculations hardly differ. With 20 answers, they differ noticeably.

For the measured CES of 75%, the 95% Wilson confidence interval is approximately 68% to 81%.

The interval for the mean

For the mean, you first need the standard deviation. It shows how widely the answers are spread around the mean of 5.225. To get it, take the distance of each point from the mean, square it and multiply by the number of answers:

(Values rounded to two decimal places.)

Standard deviation = √(total / (number of answers − 1)) = √(339.90 / 159) ≈ √2.138 ≈ 1.46

Standard error = standard deviation / √(number of answers) = 1.462 / √160 ≈ 1.462 / 12.65 ≈ 0.116

For the mean, you use a value from the t-distribution instead of 1.96. With 160 answers it's ≈ 1.975, and with very many answers it approaches 1.96. The half-width of the interval is therefore 1.975 · 0.116 ≈ 0.23. The interval runs from 5.225 − 0.23 ≈ 5.00 to 5.225 + 0.23 ≈ 5.45.

With fewer than 30 answers, the calculator points out the small sample size. For a rating scale, the t-interval is an approximation, and it can be particularly unreliable then. The width of the interval for the mean also depends on the spread. If all answers are the same, the formula gives a standard error of zero. That doesn't mean certainty about the mean in the population.

Comparing two measurements

After the first month, the internet service provider shortens the announcement in its phone queue. In the second month, it collects 120 answers (again made-up numbers):

Calculated as above:

Has it become easier? Let's do the math.

The share has risen by 7.5 percentage points. The standard errors of the two shares are ≈ 3.42 and ≈ 3.47 percentage points. The standard error of the difference is √(3.42² + 3.47²) ≈ 4.87 percentage points. So the difference is z = 7.5 / 4.87 ≈ 1.54 standard errors. In the two-sided test used here, an absolute z of about 1.96 or more counts as statistically significant at the 5% level. The p-value is ≈ 0.12.

The mean has risen by ≈ 0.32. The standard error of the difference is √(0.116² + 0.125²) ≈ 0.170, and the ratio is t = 0.317 / 0.170 ≈ 1.86. This comparison of two means is called Welch's test (NIST/SEMATECH n.d., section 1.3.5.3). It doesn't assume that both measurements have the same spread. The p-value is ≈ 0.06.

Both variants point upward, and neither reaches the usual threshold. That doesn't mean the shorter announcement achieved nothing. It means that a difference of this size, with 160 and 120 answers, can also arise by chance. Another month of measurement helps here. In this example, the mean comes closer to the threshold. It uses all seven points and also picks up shifts within the agreement range. Still, decide in advance which variant you'll compare. If you calculate both and then report the more favorable one, you're tailoring the result to fit.

What is a good CES score?

There's no universal threshold. How high the CES turns out depends on which issue you ask about, how you word the statement and when you ask. A complaint is rarely as easy to resolve as a change of address. That's why QUESTIONSTAR doesn't provide industry benchmarks.

My advice: compare yourself with yourself. Always ask with the same wording, the same scale and at the same interval after the contact, and look at the interval whenever something changes. Also compare your touchpoints with each other, for example hotline, email and self-service. Where the CES is lowest, read the answers to the follow-up question "What made it difficult?" first.

The CES alone doesn't tell you why an issue was a hassle. Nor does it tell you whether a customer will come back. It measures how easy a specific contact felt, and that's exactly what it's good for.

CES, CSAT, NPS: which metric for what?

The CES is one of three widely used three-letter metrics. They measure different things.

Dixon et al. (2010) relate their statement about better prediction to the original five-point effort question. For the agreement version presented here, check the relationship with your customers' later behavior separately. In practice, the three don't exclude each other. The section When to use CES above explains how to combine them. You can read more about the other two in the articles Net Promoter Score (NPS) and Customer Satisfaction Score (CSAT).

Template for your CES survey

If you don't want to start from scratch, use the template Customer Effort Score: service touchpoint series in QUESTIONSTAR.

To copy a template into your account, open it and use the button inside. If you don't have an account yet, register for free first. Then check whether the wording and the scale match your variant, as described above under How to word the question. You'll find more templates on this topic under Templates.

Frequently asked questions about the Customer Effort Score

How do you calculate the CES?

For the statement "The company made it easy for me to handle my issue", we use the agreement scale from 1 (strongly disagree) to 7 (strongly agree). Divide the number of answers of 5, 6 and 7 by all valid answers and multiply by 100%. With 120 out of 160 answers, that's 75%. For the mean, multiply each point by its number of answers, add everything up and divide by the number of all valid answers. On this agreement scale, a higher share or mean means less effort. More on this above.

CES vs NPS: what's the difference?

The CES asks about the effort involved in a specific issue, the NPS about the willingness to recommend your company. In the share variant used here, the CES lies between 0 and 100%. The mean of the same agreement scale lies between 1 and 7. For both, a higher value means less effort. The NPS lies between −100 and +100. You measure the CES right after a contact, while the NPS is better suited to the relationship as a whole. More in the article Net Promoter Score (NPS).

What scale does the CES use?

In the Gartner variant, an agreement scale with seven points, from "Strongly disagree" (1) to "Strongly agree" (7), with "Neither agree nor disagree" in the middle (Gartner 2018). The original question from the HBR article used a scale from 1 to 5, where a low value is good. The results of the two variants aren't comparable.

What is a good CES score?

There's no universal threshold, and QUESTIONSTAR doesn't provide industry benchmarks. Compare your CES with your own earlier measurements, with the same question and at the same interval after the contact, and pay attention to the interval. More on this above.

When should you measure the CES?

Right after a contact in which someone wanted to resolve an issue, for example after a call to the hotline, a support ticket or a complaint. Ideally, ask within 24 hours, while the memory of the effort is still fresh.

Sources

Dixon, M., Freeman, K. and Toman, N. (2010). Stop Trying to Delight Your Customers. Harvard Business Review, July-August 2010. hbr.org

Gartner (2018). What's Your Customer Effort Score? Article by Jordan Bryan, July 12, 2018. Archived original version

NIST/SEMATECH (n.d.). e-Handbook of Statistical Methods. Sections 7.2.4.1: Confidence intervals, 1.3.5.2: Confidence Limits for the Mean and 1.3.5.3: Two-Sample t-Test for Equal Means.

I hope you can now not only calculate your CES but also judge how much it tells you.

I wish you customers who say: that was easy!

Dr. Paul Marx

Author: Dr. Paul Marx
Dr. Paul Marx is the founder and Managing Director of QUESTIONSTAR and was a professor of marketing at the University of Siegen.