Customer effort score calculation: formula, example and reporting worksheet

Customer Effort Score (CES) usually summarizes valid ratings with an arithmetic mean: add the valid ratings and divide by the number of valid ratings. The calculation is easy; interpretation is not. Before reporting CES, preserve the exact question, scale endpoints, direction, touchpoint, denominator and field period. Some instruments assign higher numbers to easier experiences, while others assign higher numbers to greater effort, so a larger score can mean opposite things.

What is Customer Effort Score?

Customer Effort Score is a post-interaction survey metric intended to summarize how easy or difficult a specific task felt, such as resolving a support case, completing onboarding or making a purchase. It is not the same as CSAT, which asks about satisfaction, or NPS, which classifies recommendation ratings.

What is the CES formula?

For the common mean-score version, CES = sum of valid effort ratings ÷ number of valid effort ratings. Exclude blank and invalid records under a written rule, then report their counts separately. Do not silently convert missing values to zero.

Worked calculation

A 1–7 ease question receives 128 valid ratings whose values sum to 710. The mean CES is 710 ÷ 128 = 5.55 after rounding to two decimals. On this instrument, 1 means very difficult and 7 means very easy, so a higher score represents lower perceived effort. If 12 additional invited customers skipped the item, report 128 valid ratings and 12 missing responses; do not divide by 140.

Why can CES direction be reversed?

Current first-party documentation demonstrates both conventions. HubSpot documents a 1–7 customer-support rating where 1–3 is Difficult and 6–7 is Easy. SurveyMonkey gives a 1–5 example where lower values indicate less effort and higher values indicate more difficulty. Neither coding direction is inherently wrong. The failure is publishing “CES 4.8” without the question, range and polarity.

How do you harmonize opposite directions?

Prefer reporting each instrument in its original form. If a documented analysis requires a common direction, reverse-code before combining. For a scale from a to b, the reversed value is a + b − original value. A 1–5 effort rating of 4 becomes a 1–5 ease-oriented value of 2 because 1 + 5 − 4 = 2. Keep the original variable, create a new derived field and record the transformation in the data dictionary.

Reusable reporting worksheet

Store metric_name, question_version, question_text, touchpoint, scale_min, scale_max, low_label, high_label, direction, valid_n, missing_n, invalid_n, rating_sum, mean_ces, distribution, population, field_period, weight and verification_date.

Comparability gate

1. Confirm the exact wording and referenced task are unchanged.
2. Confirm the scale range, labels and direction are unchanged.
3. Confirm the invitation point and elapsed time after the interaction are comparable.
4. Confirm the population, channel and inclusion rules are comparable.
5. Compare missingness and the full response distribution, not only the mean.
6. Use design-aware uncertainty for sampled or weighted results.
7. Keep segmented results separate when the mix of touchpoints changed.
8. Record any reverse coding or other transformation.

Transparent method

We verified the mean calculation against SurveyMonkey's current CES documentation and the opposite-direction 1–7 implementation against HubSpot's current survey-response property documentation. The polarity table, harmonization formula, worksheet and comparability gate are editorial synthesis designed to make the calculation reproducible. They are not a universal CES standard or benchmark.

Limitations

An arithmetic mean compresses an ordered response distribution and may hide polarization. CES does not identify the cause of friction, prove that an operational change caused a score change or predict retention for every product. Wording, response timing, channel, case complexity, sampling and nonresponse can change the result. There is no defensible universal “good CES” across opposite scale directions and different touchpoints; use stable internal comparisons and publish the complete specification.