Anyone who runs customer service sooner or later gets the question from management: what is our NPS? At the same time the customer service tool shows a CSAT number, someone has read that CES is the metric that predicts loyalty, and an industry benchmark from the Swedish Quality Index talks about a customer satisfaction index. This article sorts out what they measure, what the research says about them and how a customer service team with a few hundred tickets a week should choose.
What do CSAT, NPS, CES and satisfaction indexes actually measure?
The four metrics measure different things at different moments, and that difference decides which one you should use. This is how they differ:
| Metric | The question the customer gets | Scale | When it is asked | What the answer says |
|---|---|---|---|---|
| CSAT, customer satisfaction score | How satisfied were you with the help you received? | Often 1–5 or satisfied/dissatisfied | Right after a closed ticket | How that particular ticket went |
| CES, customer effort score | Did the company make it easy for you to resolve your issue? | 1–7 in Gartner's version | Right after a closed ticket | How much effort the customer had to put in |
| NPS, net promoter score | How likely are you to recommend us to a friend or colleague? | 0–10 | At regular intervals, independent of tickets | The customer's attitude to the company as a whole |
| Customer satisfaction index (NKI in Sweden) | Several questions on expectations, quality and satisfaction | Index 0–100 | Annually, often via an external survey | The relationship with the company compared with others in the industry |
NPS is calculated as the share answering 9 or 10 minus the share answering 0 to 6, and can therefore range from minus 100 to plus 100. NKI is the Swedish variant of a customer satisfaction index; the Swedish Quality Index measures it with a statistical model that combines several questions into one index, and reports NPS alongside it.
Why is NPS not enough for customer service?
NPS measures the customer's relationship with the company and was designed to predict growth, not to run a customer service department. The metric was introduced by Fred Reichheld in Harvard Business Review in 2003, where he showed that the share of customers willing to recommend the company correlated with growth in most of the industries studied. The question therefore concerns the company as a whole, and the answer is affected by price, product, delivery and brand as much as by customer service.
For customer service that creates three problems:
- The answer cannot be tied to an action. A customer who gives a 6 may be unhappy with the shipping cost. Customer service can do nothing about that, and does not even know it was the reason.
- The frequency is too low. NPS is typically asked a few times a year to a sample, while customer service needs signals per week and per ticket type.
- The scale is blunt at low volumes. A team with 300 tickets a week and a 20 percent response rate gets 60 answers. With the NPS formula a single customer moves the metric several points.
That does not make NPS wrong. It is the right metric at management level, ideally compared with the industry through a satisfaction index. Customer service contributes to the number, but does not steer by it.
What does the research say about customer effort?
How easy the customer had it weighs more for loyalty than how delighted the customer became. That conclusion comes from a study of more than 75,000 customers who contacted customer service or used self-service, published by Matthew Dixon, Karen Freeman and Nicholas Toman in Harvard Business Review in 2010. The researchers found that going above and beyond made little difference: customers wanted a simple and quick solution to their problem. They introduced the customer effort score and showed that it predicted loyalty better than both customer satisfaction measures and NPS.
Gartner, which manages the research today, describes customer effort as a good indicator of whether the customer buys again and buys more, and phrases the question as a statement to rate on a seven-point scale: the company made it easy for me to resolve my issue. What drives effort is what customer service actually controls: having to contact the company again, switching channels, repeating your story, being passed around.
The customer who emailed about a return, got a reply with a question back, answered, waited and finally called may well answer "satisfied" on CSAT because the return went through. On CES the same customer scores low, and that is the answer that predicts whether the customer shops again.
How do you measure CSAT and CES per ticket without wearing customers out?
With one question, asked once per ticket, shortly after the ticket is resolved, and with a free-text field. Zendesk by default sends its CSAT survey a day after the ticket is marked as solved, asks for a positive or negative rating and an optional comment, and lets the customer respond within 28 days.
A few rules that keep the response rate up and the answers useful:
- One survey per ticket, not per message. A customer with three tickets in a month gets three surveys; a customer with ten messages in one ticket gets one.
- Ask about the ticket, not the agent. "How satisfied are you with the help?" rather than "How satisfied are you with Anna?". The customer often rates the decision anyway, and that should show in the comment, not in the agent's pay review.
- Put CES in the same survey if you want both. Two questions and a free-text field is the limit. More questions lower the response rate more than they add.
- Report the response rate next to the number. A CSAT of 90 percent from 8 percent of customers is a different thing from 35 percent.
Which metric fits which decision?
The metric is chosen by the decision it should drive, not by what the tool happens to show on its dashboard. This is how they fit together:
| Decision you need to make | Metric to look at | Why |
|---|---|---|
| Is the chatbot or the new AI draft working? | CSAT and CES per channel, before and after | The customer judges exactly those tickets; a CSAT that drops after automation is a clear signal, as we describe in the article on grounded chatbots |
| Which ticket types should we write better answers for? | CES per ticket type, plus the comments | High effort points to answers that require follow-up questions or channel switches |
| Does the help centre need new articles? | The comments in CSAT and CES | The free text reveals questions you could not answer, the same method as in the article on knowledge gaps |
| How do we compare with the industry? | Satisfaction index and NPS, annually | Comparable metrics at company level |
| Should we staff differently? | No satisfaction metric; response time and queue | Satisfaction is an outcome, not a planning input |
The satisfaction metrics say how it went, not why, and should be read together with the operational metrics such as response time and resolution rate, described in the article on customer service metrics.
How do you read the result without fooling yourselves?
By always looking at three things at once: the number, the response rate and the comments. The number alone misleads in predictable ways.
- Non-response is not random. Those who respond are more often very satisfied or very dissatisfied. If CSAT rises while the response rate falls, you have probably not improved, just lost the middle.
- Channels cannot be compared directly. The chat gets answers from customers still in the window, email from those who bothered to open one more email. Compare each channel with its own history.
- The rating is often about the decision. A rejected complaint gives a low CSAT even when the handling was correct. Sort the negative answers into wrong answer, slow answer and bad decision. The action is different for each pile.
The comments are the part of the measurement that gives the most. The free text shows which answers are missing, which articles are unclear and where customers switch channels. Read systematically, they give the same picture as an analysis of the inbox does, but from the customer's side.
What to do
- Turn on one question per ticket. CSAT with satisfied or dissatisfied and a free-text field, sent a day after the ticket is resolved. Most tools have it built in.
- Add CES if you have the volume. The statement "it was easy to get help" on a seven-point scale, in the same survey. Start with CSAT alone if you have fewer than a hundred tickets a week.
- Report three numbers every week: share positive, response rate and number of answers. Per channel and per ticket type when volume allows.
- Read every negative comment and sort them into wrong answer, slow answer and bad decision. Decide one action per pile.
- Measure before and after every change in how you answer: new article, new template, chatbot or AI draft. In Supportifier you follow the customer's rating per channel and per article, so a drop can be traced to the answer that caused it.
- Leave NPS and the satisfaction index to management, once a year, and contribute what customer service's numbers say about why.
Common questions
Should customer service measure NPS?
Not as a steering metric. NPS concerns the customer's relationship with the whole company and is affected by price, product and delivery as much as by customer service, so the answer can rarely be tied to something customer service can change. Measure it at company level a few times a year, and let customer service measure CSAT and CES per ticket. Customer service's numbers often explain why NPS moves.
How many answers do you need to trust CSAT?
Enough that a single customer does not move the metric. With 30 answers a week one customer is worth just over 3 percentage points, so look at four weeks at a time or per month. More important than the count is the response rate: report it next to the number, and only compare periods with roughly the same response rate. The comments are useful from the first answer.
Can we measure CSAT on the chatbot?
Yes, and you should do it separately from the tickets a person handled. Ask the same question in the chat when the bot closes a ticket, and follow the share of positive answers per question or article the bot answered from. If CSAT drops or CES rises for a certain ticket type, the source material is missing or wrong, and the handover to a person should then come earlier for that type.
What is a good CES?
There is no industry figure to compare with, because scales and wording differ between tools. Use the metric relatively: per ticket type, per channel and over time. The ticket types that score worst are where customers have to come back, switch channels or repeat themselves, and they should be fixed first. An improvement that holds over several weeks is what counts, not the level itself.
Sources
- The One Number You Need to Grow — Harvard Business Review, 2003
- Stop Trying to Delight Your Customers — Harvard Business Review, 2010
- Improve Customer Experience to Gain Loyalty: Effortless Experience — Gartner, read 2026
- About the CSAT (Customer Satisfaction) user experience for email and messaging — Zendesk, read 2026
- Svenskt Kvalitetsindex modell — Swedish Quality Index, read 2026
Rickard Collander