Running a CSAT survey on WhatsApp that people actually answer

Updated 2026-09-24 · 5 min read · By the Convo360 team

Why email surveys fail and WhatsApp ones do not

The standard support survey asks the customer to leave the conversation, open an email, click a link, load a page and choose a face. Each of those steps loses people, and the ones who survive all five are disproportionately the very angry and the very pleased. That is how you end up with a barbell of ones and fives that tells you nothing about the middle.

On WhatsApp there is no journey. The question arrives in the thread they were already in, and the answer is one character. Typing "4" takes a second and requires no context switch at all.

The practical result is a response rate several times higher and, more importantly, a sample that includes ordinary satisfied customers rather than only the extremes. A survey's usefulness depends far more on who answered than on what the average was.

There is a second reason email surveys mislead, which is timing. By the time a survey email is opened, often a day or two later, the customer is rating a memory rather than an experience. Memories drift towards the extremes: people remember being very annoyed or very pleased, and forget the competent middle entirely.

Timing, wording and length

Three decisions, and the defaults people choose are usually wrong.

• Timing. A few minutes after the ticket is resolved. Long enough not to interrupt, short enough that the customer remembers the interaction. Next-day surveys measure mood, not service.
• Wording. Short and human. "How did we do? Reply 1 to 5, where 5 is great." Avoid corporate phrasing, which reads as a form and gets ignored.
• Length. One question. The follow-up comment is a second message, sent only after they have answered the first.

That last point is the one teams resist and the one that matters most. Asking for a score and a reason together roughly halves the response rate, because it turns a one-second reply into a writing task.

Be careful about surveying every single ticket for your most frequent customers. A wholesale buyer who opens nine tickets a week does not want nine surveys, and the irritation shows up in the scores. Surveying a proportion of their tickets, or one a week, gives you the same signal without the fatigue.

Setting it up on Convo360

In Support Desk, open the Satisfaction panel and switch surveys on. You control the delay after resolution, the question itself, the optional follow-up question and the thank-you line.

The survey goes out on whichever channel the ticket came from, inside the same thread, so a WhatsApp customer is asked on WhatsApp and an email customer by email.

Reading the answer is more forgiving than most systems. A reply of "4", "4/5", "four" or four star characters all register as four. A reply that is clearly not a rating, such as "5 items are still missing", is treated as a new message and reopens the ticket rather than being recorded as a five. That distinction matters more than it sounds: naive parsing turns complaints into perfect scores.

Convo360 sends the survey only where there is an open window to send it in. On WhatsApp that means the twenty-four hour service window, so a survey never becomes a paid template message the customer did not expect. If the window has closed, the survey is skipped rather than sent, which keeps both the cost and the intrusion down.

The comment question is separate and optional in Convo360 for the same reason. It is sent only after a score arrives, so the customer who wanted to type one character is never asked to write a paragraph first.

Reading the score without fooling yourself

Three habits separate a useful CSAT programme from a vanity metric.

Response rate comes first because it governs everything else. A 4.6 average from eight per cent of customers is not a measurement, it is an anecdote with a decimal point.

The share of ones and twos is the number to act on. An average creeping from 4.5 to 4.3 is noise; a doubling of ones is a specific group of customers having a specific bad experience, and the tags on those tickets usually name it.

Watch for the survey that measures the wrong thing. If your resolution is 'we have raised this with the courier', the customer is rating a promise, not an outcome. Teams with a lot of dependent tickets often do better surveying after the customer confirms the problem is gone, rather than when the agent marks it resolved.

What to do with a bad score

A low rating is only worth collecting if something happens next. The minimum useful loop is that any score of one or two is looked at by a human within a day, and the ticket is reopened if the customer's problem plainly was not solved.

Resist the urge to argue with the customer about the rating. The purpose is not to change their mind, it is to find out what went wrong while they still remember. A short "Sorry that missed the mark, what should we have done differently?" gets useful answers surprisingly often.

Review scores by tag once a month. Individual ratings are noisy and personal; ratings grouped by what the ticket was about are where the pattern lives. If every low score carries the same tag, you have found a process problem rather than a people problem, and no amount of coaching will fix it.

Share the comments with the team, unedited, once a week. Scores are abstract and easy to argue with; a sentence in a customer's own words is not. The teams that improve fastest are usually the ones where everybody reads the raw feedback rather than a summary of it.

And keep one number in view alongside the score: reopen rate. A high satisfaction score with a rising reopen rate usually means customers are being polite about an issue that came straight back.

Step by step

1. Switch surveys onSupport Desk → Satisfaction. Enable the survey and set the delay after resolution, typically five to fifteen minutes.
2. Write one short question'How did we do? Reply 1 to 5, where 5 is great.' Keep it human and keep it to one line.
3. Add the follow-up separatelySet the optional comment question so it is sent only after the customer has given a score.
4. Watch response rate firstFor the first month, judge the setup by how many people answer rather than by the average score.
5. Review by agent and tag monthlyGroup scores by tag to separate process problems from people problems, and act on the share of 1s and 2s.

Example conversation

Frequently asked questions

Related reading

Available across India — see WhatsApp & AI calling automation by state and city.