The most common problem with customer satisfaction surveys is not a low score, but that no one knows what to do with it. The survey goes to everyone, the average rises by two tenths, a slide appears in a meeting, and that is the end of it. Yet the most valuable part of such a survey is usually not the rating, but the one sentence the customer adds beside it.
CSAT, NPS, and CES—three different questions
These abbreviations are sometimes used interchangeably even though they measure different things and suit different moments. The choice is not about fashion, but about exactly what you want to learn.
Scroll the table sideways to see all columns.
| Metric | What it asks | Best time |
|---|---|---|
| CSAT | How would you rate this specific service interaction? | Immediately after the conversation or case is closed |
| NPS | Would you recommend the company to others? | After a longer relationship, not after a single interaction |
| CES | How much effort did it take to resolve the issue? | After a multi-step process |
For website customer service, CSAT is enough in most cases. It is closest to the event, easiest to interpret, and hardest to manipulate. NPS is often misused to assess a single conversation, which it was not designed to do—asking whether someone would recommend the company after a two-minute chat usually measures their attitude toward the brand, not the quality of that interaction. If a case took several steps and you want to see where the customer lost time, add the separate effort question described in the guide to customer effort score.
Start with one metric and keep it for at least six months. Changing the method every quarter leaves you with no meaningful basis for comparison.
When to ask so that people actually respond
The response rate falls sharply as time passes after the event. A survey emailed three days after a conversation mainly captures the extremes: customers who are very satisfied and those who are deeply disappointed. A question asked immediately after the conversation closes captures the normal distribution—and that is the distribution you can use.
- Right after the conversationThe highest response rate, while the details are still fresh.
- After the issue is resolvedA good fit for cases that stretched across several messages.
- Not during the conversationAsking for a rating before resolving the problem feels dismissive.
- Not after every messageOne rating per case. Repeating the question lowers the score by itself.
It is also important to prevent one person from submitting a dozen ratings for the same conversation. This is less about abuse than data hygiene—in Meetlead, one rating is recorded per visitor and conversation, so repeated clicks by the same person cannot shift the average.
A quick rating with an optional comment, once per conversation, with no reminders. Explore the product.
The beginning of a customer relationship is another valuable moment—the first few days reveal more than a survey after six months. We discuss this in our article about the first days of working with a customer.
How to phrase the question
A customer service survey has one advantage over marketing research: it can consist of a single question. Add a second, third, and fourth, and you turn a quick reaction into homework. The response rate then falls so low that the result stops meaning anything.
- Ask about the service the customer just received, not the company as a whole.
- Use a short scale—five points are enough and easy to read on a phone.
- Do not lead the response with wording such as “Was our help helpful?”
- Provide a comment field, but do not make it mandatory.
- Do not ask for contact details—you already have them from the conversation.
- Show a thank-you message after the rating and do not ask another question.
Only a minority of respondents usually leave an optional comment, but that minority provides the material for change. A rating tells you that something was wrong. A sentence such as “I waited 20 minutes even though your website says you respond immediately” tells you exactly what to fix.
Requiring a comment improves the quality of each response but markedly reduces the number of responses. If you require one at all, do so only for negative ratings.
How to read the result without fooling yourself
A customer satisfaction score is one of the easiest figures to distort unintentionally. Not because anyone is lying, but because a selected group of people responds in selected situations after selected conversations.
Scroll the table sideways to see all columns.
| Pitfall | What it looks like | What to do |
|---|---|---|
| Small sample | The average jumps every week | Report the number of responses alongside the score |
| Survey shown only after successful cases | A very high, stable score | Check which conversations actually trigger the question |
| Operator asks for a good rating | The score rises with no change to the process | Explicitly rule this out in your service standards |
| Average with no breakdown | One number for every channel | Separate chat, phone, and support tickets |
| Ignoring non-respondents | No response is treated as neutral | Treat a low response rate as a separate signal |
It is worth comparing the rating with service time, because together the two figures tell you more than either does alone. Falling satisfaction alongside improving response times is a classic sign of service that is fast but superficial—we describe this pattern in our article about first response time in live chat.
A rating rarely works in isolation—it is best read in a short, recurring report alongside service times and repeat contacts. We explain how to assemble one in our article about customer service reporting.
What to do with a response
A survey is worthwhile only if its results lead to changes. The routine below takes a few dozen minutes each month and turns a collection of ratings into a list of concrete actions.
- Read every comment.There are usually fewer than you think, and they carry the substance.
- Group recurring themes.Three similar comments form a pattern, even if they are not statistically significant.
- Separate service from product issues.Some dissatisfaction concerns price or availability, not the way the conversation was handled.
- Follow up on extremely low ratings.A short message asking what went wrong can be the least expensive way to save the relationship.
- Choose one change.One measurable improvement per month is worth more than five made at once.
- Revisit the numbers after a month.Check whether the change moved anything before declaring the matter closed.
Recurring comments are also ready-made material for content and a knowledge base—and if you use an assistant to answer questions, they become a direct list of gaps to fill. We discuss this in our article about AI in customer service. If the comments concern lost cases rather than slow responses, the right place to go next is our guide to customer service ticketing systems.
The lowest ratings usually come from complaints, which need their own process rather than merely a better tone. We cover that process in our article about handling customer complaints.
