CSAT and NPS Surveys on WhatsApp and Voice

Customer feedback programmes in the Gulf and the Levant fail for a boring reason. The survey goes out by email to a customer who answered the phone in Arabic, gets a response rate in the low single digits, and the result is reported as a score with no denominator. Nobody acts on it, and within two quarters the programme is a slide.

The fix is not a better survey tool. It is asking on the channel the customer already used, at the moment the conversation is still fresh, and reporting the answer with the number of people it came from attached. This guide covers how to do that, and what the platform gives you to do it with.

What you can ask

Five question types cover almost every contact centre programme without turning the survey into a form:

• Yes or no, for a single clear check such as whether the issue was resolved.
• CSAT on a 1 to 5 scale, for satisfaction with one interaction.
• NPS on a 0 to 10 scale, for likelihood to recommend.
• Consent, yes or no, recorded as a question type in its own right so the consent sits with the response rather than in someone's notes.
• Open text, for the sentence that explains the score.

Keep it short. On voice and on WhatsApp, one scored question plus one open question is usually the whole survey. Every extra question costs completions, and a completed two-question survey beats an abandoned six-question one in both response count and honesty.

Where you can ask it

The same published survey can be asked on five channels, and responses come back into the same analytics regardless of how they arrived.

One language note to plan around: the survey page renders in English and Arabic today, and other languages are covered on the voice side through IVR voice packs. This is narrower than the platform, which handles 11 languages, so check it against the languages your customers actually answer in before you launch a programme.

How a survey gets triggered

Three triggers exist, and the choice decides your response rate more than the wording does.

The post-call handoff is the strongest. The agent presses Send to survey on the agent desktop as the call ends, and the survey goes out on the configured channel while the conversation is still fresh in the customer's mind. It also gives the agent a small, healthy stake in the score.

The IVR node triggers the survey during the call, which suits inbound programmes where the customer is already on the line and a transfer to a short menu costs nothing.

The outbound survey campaign runs the survey as a campaign in its own right, with pacing and quotas, for when you need a representative sample of a customer base rather than a reaction to one interaction.

Locked versions, so your trend line stays honest

Published survey versions are locked. Editing a live survey creates a new version rather than changing the old one, so responses already collected keep the exact wording they were collected under. This is the difference between a trend you can present and a chart that quietly changed meaning in March when someone reworded question two. Six mobile-first page templates carry your tenant branding, with a live preview while you build.

Reading the numbers, with denominators

Survey analytics cover an overview, a breakdown, a per-question view, a trend and the drivers behind the score, with an export, and every view states its denominator. That last part is the whole argument. A CSAT of 4.6 means one thing from 400 responses and nothing at all from 7.

Deliveries are logged per channel, which tells you what was sent as well as what came back. That is how you separate two very different problems: a low score, and a low response rate caused by messages that never arrived. Scheduled reports go out daily, weekly or monthly, and survey usage is metered per tenant, which matters if you resell the platform and need to show each client their own numbers.

Recovery cases: what a bad score should trigger

A score is only worth collecting if something happens when it is low. Detractors and low scores are routed to a queue as recovery cases, each assigned to a person and closed with an outcome. The unhappy customer gets a call back rather than a place on a chart.

Set the rules before you launch: which score triggers a case, who owns the queue, how quickly a case must be opened, and what closing it requires. A recovery case that is opened three weeks later is an audit record, not a recovery. For collections and telecom programmes, a fast recovery call after a bad interaction is often worth more than the entire measurement exercise around it.

A programme that survives its first quarter

Run one survey per interaction type rather than one survey for the business. Ask on the channel the customer already used. Cap the sample so the same person is not asked every week. Report the score with its denominator and its response rate side by side, and report the recovery cases closed underneath, because that is the number that shows the programme changed something. Then review the open text monthly, since the sentence after the score is usually where the operational fix is hiding.

Frequently asked questions

Which question types are available?

Yes/no, CSAT on a 1 to 5 scale, NPS on a 0 to 10 scale, consent yes/no, and open text. Consent is a question type in its own right, so a recorded consent sits with the response rather than in an agent's notes.

Which channels can a survey be asked on?

Voice IVR by keypress or speech, the Layla AI voice runtime, WhatsApp, SMS, and a web page behind a short link. Responses from every channel land in the same analytics.

What languages does the survey page render in?

English and Arabic today, with other languages covered on the voice side through IVR voice packs. The platform itself handles 11 languages, so confirm the languages your customers will answer in before launching.

How is a survey sent after a call?

The agent presses Send to survey on the agent desktop when the call ends, and it goes out on the channel the survey is configured for. A survey can also be triggered from an IVR node during the call, or run as an outbound survey campaign with its own pacing and quotas.

Can I edit a survey that is already live?

You can edit, and the edit creates a new version. The published version stays locked, so responses already collected keep the wording they were collected under and the trend line stays comparable.

What happens when a customer gives a low score?

Detractors and low scores are routed to a queue as recovery cases. Each case is assigned to a person and closed with an outcome, so a bad score produces a follow-up you can audit rather than only a number on a chart.

Related reading

• Surveys and CX
• WhatsApp and SMS
• AI Voice Agent
• Reports and Insights