CSAT, NPS and CES: which customer satisfaction metric to use

Ask three teams how happy their customers are and you'll get three answers with three acronyms. Support quotes CSAT, the founder quotes NPS, and someone in product has read that CES predicts loyalty better than both. All three can be right, because they measure different things.

The trouble starts when a team picks a metric because it sounds serious rather than because it answers a question they need answered. NPS sent after every chat tells you nothing about the chat. CSAT sent once a year tells you nothing you can act on.

This guide explains what each metric measures, when to send it, how to word it, and how to turn the answers into changes. Mid-July is a good moment to set it up: volumes are usually calmer than in spring, and you'll have a few months of baseline before the busy end of the year.

The problem: scores nobody can act on

Most satisfaction programmes fail in one of these ways:

• The wrong metric for the job. An NPS survey after a password reset measures the product, the price and the weather, not the reset.
• Survey fatigue. Customers get a survey after every email, rate everything five stars or nothing at all, and the numbers stop meaning anything.
• A number without reasons. "CSAT dropped four points" starts a meeting but not a fix.
• Nobody closes the loop. A customer writes "this was useless" and never hears back.
• AI and humans mixed together. A blended score hides whether the AI agent or the team is the problem.

Choosing the right metric, then building a simple routine around it, fixes most of this.

CSAT, NPS and CES at a glance

	CSAT	NPS	CES
Measures	Satisfaction with one interaction	Overall loyalty and willingness to recommend	How easy it was to get something done
Typical question	"How would you rate the help you received?"	"How likely are you to recommend us to a friend or colleague?"	"How easy was it to handle your request today?"
Scale	1–5 (or emoji, or thumbs)	0–10	1–7 (very difficult to very easy)
Score	% of 4s and 5s	% promoters (9–10) minus % detractors (0–6)	Average, or % of 5–7
When to send	Right after a conversation closes	Every few months, by email	Right after a task or process
Owned by	Support	Leadership, the whole company	Support and product
Best for	Coaching, content fixes, AI quality	Long-term trend, benchmarking yourself	Removing friction from processes

Step 1: Start with CSAT after every conversation

For a support team, CSAT is the right first metric. It's specific, immediate and easy to act on.

How to set it up:

• Trigger it automatically when a chat closes or a ticket is solved. For chat, ask straight away; for email, a few hours later is fine.
• Keep it to one question, plus an optional comment box. Every extra question cuts responses.
• Use simple wording: "How would you rate the help you received?" Avoid "Did our agent exceed your expectations?"
• Tag every conversation with a reason (order status, returns, billing, bug, complaint), so you can see CSAT by topic, not just overall.
• Report AI and human conversations separately. If the AI agent's CSAT is well below your team's, look at the help content it answers from.

A low CSAT on one topic is a process problem. A low CSAT for one agent is a coaching conversation. A low CSAT everywhere is a staffing or policy problem.

Step 2: Add CES where friction costs you

Customer Effort Score asks how easy something was. It's the best metric for processes customers have to get through: returns, refunds, cancellations, account changes, self-service flows.

Good places for CES:

• after a return or exchange is completed;
• after a self-service flow or help article ("Did this solve your problem easily?");
• after a subscription change or cancellation;
• after onboarding or setup.

Word it as an agreement statement or a direct question: "The company made it easy for me to handle my issue" (strongly disagree to strongly agree), or "How easy was it to return your order?" Low-effort answers tell you what to keep; high-effort answers, with a comment, tell you which step to remove.

Step 3: Use NPS for the big picture, sparingly

NPS asks how likely someone is to recommend you. It reflects everything: product, price, delivery, support, brand. That makes it useful for leadership and useless for coaching a support agent.

How to run it well:

• Send it on a schedule, for example twice a year, by email, to a sample of active customers.
• Don't send it right after a support conversation. You'll measure the conversation, not the relationship.
• Always ask "why?" The follow-up comment is where the value is.
• Compare against yourself. Your trend over time matters more than an industry average.

Step 4: Ask at the right moment

Moment	Metric	Channel	Timing
Live chat closed	CSAT	In the chat widget	Immediately
Email ticket solved	CSAT	Email	A few hours later
Return or refund completed	CES	Email	Same day
Help article viewed	CES or thumbs up/down	On the page	Immediately
Active customer for 3+ months	NPS	Email	Twice a year
Subscription cancelled	CES plus a reason question	Email or in-app	Immediately

Limit how often one person can be surveyed, for example no more than one survey per customer every few weeks, except post-conversation CSAT.

Step 5: Turn answers into changes

A weekly 30-minute routine is enough for most small teams:

1. Read every negative rating and comment from the past week.
2. Tag the real reason: slow reply, wrong answer, unhappy with a policy, product problem, AI couldn't help.
3. Fix one thing at the source: rewrite a help article, change a saved reply, adjust a policy, or add a Q&A for the AI agent.
4. Close the loop: reply personally to customers who rated badly within a day or two. "You rated our help as poor, and you were right. Here's what we've done."
5. Track the trend monthly, by topic and by AI versus human.

Common mistakes

• Rewarding agents for scores alone. It encourages asking for good ratings rather than solving problems.
• Surveying only happy customers. Selective surveys make the numbers meaningless.
• Changing the question every quarter. You lose the trend.
• Treating no response as satisfied. Silence isn't a five.
• Never reading the comments. The score tells you where; the comment tells you why.

How to get started this week

1. Day 1: turn on post-conversation CSAT in your support tool. In Tidio, that's one toggle in the settings, with editable texts.
2. Day 2: set up reason tags for conversations if you don't have them.
3. Day 3: pick one process where effort matters (returns, cancellations or onboarding) and add a CES question at the end.
4. Day 4: decide when you'll run NPS, and with which tool, if you need it at all this year.
5. Every Monday: read the negative ratings, fix one thing, reply to unhappy customers.

For the full set of support numbers worth tracking, see the customer service metrics that matter, and for turning the comments into product changes, read how to turn support conversations into product insights.