CSAT vs NPS vs CES: Formulas, Scales and When to Use Each

When should you use CSAT, NPS or CES?

Use CSAT to judge individual conversations, CES to find friction in a process, and NPS to track overall loyalty. If you can run only one survey, run CSAT after support.

1. After a support conversation closes: CSAT, optionally with a CES question for complex cases.
2. After a process completes (return, KYC, booking change): CES.
3. Every quarter or half-year, across the base: NPS, separated by segment and country.
4. After onboarding: CES plus a short NPS.
5. For agent coaching: CSAT with comments, reviewed in context. Do not rank agents on NPS.

Pair survey scores with operational metrics such as first response time and reopen rate, as described in the omnichannel support KPI guide.

What survey wording works? Copy-paste templates

Ask a single, neutral, specific question, and put the optional comment second.

CSAT (1 to 5) "How satisfied were you with the help you received today? 1 = very unsatisfied, 5 = very satisfied."

CSAT for a specific order (suits a WhatsApp utility template) "You chatted with us about order {{1}}. How was your experience? Reply 1 to 5."

NPS "How likely are you to recommend {{company}} to a friend or colleague? 0 = not at all likely, 10 = extremely likely."

CES "{{company}} made it easy for me to handle my request. 1 = strongly disagree, 7 = strongly agree."

Follow-up comment (all three) "Thank you. What is the main reason for your score?"

Avoid leading words ("How great was..."), double-barrelled questions ("Was the agent fast and friendly?") and surveys that ask for a rating and a purchase in the same message.

How do you send satisfaction surveys over WhatsApp?

Inside the 24-hour customer service window you can send a survey as an ordinary message; outside it you must send an approved template. Meta documents the window: a user message starts a 24-hour timer, each new user message resets it, and when it closes you can only send pre-approved templates.

Step by step

1. Trigger it when the conversation is resolved, while the window is still open.
2. Choose the format. Reply buttons allow up to three choices, and list messages up to ten rows. A 1-to-5 rating fits a list, or three buttons for a simple good/neutral/poor scale. A 0-to-10 NPS has 11 values, so use a list for ranges (0 to 6, 7 to 8, 9 to 10) or a WhatsApp Flow. See interactive messages and WhatsApp Flows.
3. Capture the reply by its button or row ID, not its visible title, so you can change wording or translate without breaking reports.
4. Ask the comment next, only after the rating, as a free-text reply.
5. Store the result against the conversation, agent, queue and channel.

Which template category is a survey?

Meta describes utility templates as typically triggered by a user action or request and non-promotional. Its template guidance lists collecting feedback on previous orders, transactions or engagements as a utility use case, and says a general or generic survey or request for feedback will not be approved as utility. A feedback survey that includes promotional content is treated as marketing.

So tie a utility survey to a specific interaction (an order number, a visit), keep out offers and discount codes, and expect Meta to decide the category. Do not choose utility to avoid marketing charges; Meta may recategorise it, and charges depend on category and recipient market, so check Meta's current rate card and our WhatsApp pricing pages. Consent still applies: contact people only when they have opted in.

How do you send satisfaction surveys by SMS?

Send a short message with a single reply value, and use a sender type that can receive replies. Two-way SMS needs a number such as a long code or short code that accepts inbound texts; replies reach you by webhook or in an inbox. See the glossary entry for two-way SMS.

Example (160 characters is the single-segment limit for GSM-7 text; Arabic uses Unicode with a shorter limit, which our character counter shows):

"Thanks for visiting {{store}}. How was your service? Reply 1 (poor) to 5 (excellent). Reply STOP to opt out."

Handle bad input politely ("Please reply with a number from 1 to 5"), and always honour STOP. Check local timing and consent rules in SMS marketing rules by country; a service survey is not a marketing message in every country, but the safest practice is to send it only to customers who have an existing relationship with you.

How do you raise response rates and avoid bias?

Ask once, ask immediately, ask in one tap and in the customer's language.

• Timing: send within minutes of resolution, while the conversation is fresh.
• Length: one rating first; the comment is optional.
• Language: Arabic and English versions, with right-to-left text previewed on a real device.
• Channel match: survey on the channel the customer used.
• No double asking: one survey per conversation, with a cool-down per customer.
• Report the denominator: responses, surveys sent and response rate.
• Watch non-response bias: unhappy customers who already left will not answer; compare with reopen rate.
• Close the loop: route low scores to a supervisor for follow-up.

What mistakes should you avoid?

• Using NPS to score agents or single tickets.
• Changing the scale or the "satisfied" definition and comparing with older data.
• Publishing a score without response count.
• Sending the survey before the issue is actually resolved.
• Mixing a survey with a promotion, which changes the WhatsApp category and the customer's trust.
• Ignoring comments. The score tells you that something is wrong; the comment tells you what.

How OmniChannels.ai helps

OmniChannels.ai, powered by BroadNet Technologies, runs WhatsApp, SMS, Instagram, Messenger, Telegram and Email in one Omni Inbox, where agents tag conversations and reply with templates. WhatsApp templates support variables and buttons, with phone preview and approval tracking, and two-way SMS brings replies back into the platform. Chatbots can collect ratings and hand low scores to a human agent. Explore the customer support solution or book a demo.