Customer Feedback Calls: Run a Phone Survey
Jaseem Abbas
- Published

A customer feedback call is a short phone call, made by a person or an automated agent, that asks a customer to rate a recent purchase, visit or support call and say why. Some Indian search results use "feedback call" for a job interview, so to be clear, this guide is about calling customers. It covers feedback calling as a phone survey: when to call, what to ask aloud, whom to sample and what to do with the answers.
Here is the problem it solves. A two-wheeler service chain in Tiruchirappalli sends an SMS link after every service visit. A few people reply, and most of those are angry. The chain switches to a 90-second call in Tamil or English, three days after the visit. The calls bring back reasons: a wash that was skipped, a bill that was explained badly, a wait of two hours. The chain, the city and the numbers in this guide are made up.
What is a customer feedback call?
A customer feedback call is a short phone call that asks a customer to rate a recent purchase, visit or support call and explain the rating. It is a phone survey with one customer at a time. In research it is called a telephone survey, and in a call centre it is also called a survey call or a post call survey.
The three names are often mixed up, so here is how we use them:
Feedback call, survey call or IVR survey?
A feedback call is the everyday name. A survey call is the same call with fixed questions you count month after month. An IVR survey runs through an IVR, usually a recording that asks the customer to press a key, and suits one number.
This guide is for an Indian customer experience or operations manager with a small calling floor. It does not cover job-interview feedback or agent coaching, for which see our guide to call audits in BPO.
When should you call instead of sending an SMS, WhatsApp or email?
Call when the answer is worth more than the cost of the call. That means a high-value or at-risk customer, a case where you need the reason and not just a score, or a customer who does not read English or use email. For thousands of low-value orders, a link by SMS or WhatsApp costs less per response.
The table below is our own framework. It is advice built from how a call works, not a tested result.
Situation: High-value or at-risk customer (renewal due, open complaint)
- Best channel
- Call
- Why
- You can ask why and save the account on the spot
Situation: You need the reason, not just a score
- Best channel
- Call
- Why
- A spoken answer to an open question tends to run longer than a typed box
Situation: Customers who do not read English or do not use email
- Best channel
- Call, in their language
- Why
- A spoken question needs no reading
Situation: Thousands of low-value orders
- Best channel
- SMS or WhatsApp link
- Why
- Cheap per response; a call per order costs too much
Situation: Rating right after a support call
- Best channel
- A question on the same call, or a short post-call IVR
- Why
- Memory is fresh and there is no second dial
Situation: A long questionnaire (10 or more questions)
- Best channel
- Online form
- Why
- Nobody stays on a call that long
A phone number reaches almost everyone. TRAI's press release 125/2026, dated 28 September 2026, reports 1,207.59 million active wireless subscribers in August 2026, so a call can reach customers who never read your email.
Reaching a number is not the same as being answered. Pew Research Center found that typical US telephone survey response rates fell to 7% in 2017 and 6% in 2018, and named robocalls and spam flagging of survey calls as likely causes. Those are public polls of strangers in the US, not a company calling its own customers, so do not read 6% as an Indian number. We could not find an Indian figure we could verify. The lesson still holds: people screen unknown numbers, and our guide on why business calls get marked as spam covers what you can do about it.
How do you write survey questions for the ear?
Write for listening, not reading. A listener cannot scroll back, so use few questions, short answer options and a scale whose meaning is said before the question. This is the biggest difference between a phone survey and a web form.
Here are seven rules. They are our advice, and the one on option order is backed by research.
- Three to five questions, under two minutes. Say the count at the start: "three quick questions".
- One scale per call. Say what the ends mean before the question, not after: "1 means very poor, 5 means excellent."
- Short answer options. If you read a list, remember that the last option heard tends to win.
- Ask the number first, then one open question. "What is the main reason for your score?"
- Accept spoken numbers in the customer's language. "Paanch", "five" and "excellent" can all mean the top box. Read the answer back to confirm it.
- Never lead. "You were happy with the service, right?" pushes the answer. Template 11 in our call center scripts guide says the same.
- No offer and no upsell on the same call. It changes the mood and, in India, may change how the call is classed (see the consent section below).
Why does the order of options matter on a call?
On a phone, people tend to choose the last option they hear. Holbrook, Krosnick, Moore and Tourangeau examined 548 experiments in telephone surveys by the Gallup Organization (Public Opinion Quarterly, 2007). They found that response-order effects were predominantly recency effects, and that they happened most with questions that were harder to understand and among respondents with the least education.
Two things follow. Keep questions plain, and if you must read a list, reverse its order for half of your callers so the effect cancels out across the sample. A scale of one to five read from "poor" to "excellent" is itself a list. That is why you say the meaning of the ends first and keep the options to a number the customer can hold in mind.
How many questions is too many?
More than five is too many for a call. Every extra question costs you callers who say "I am busy" halfway through, and they are the ones you needed. With ten questions, send an online form. This is our advice, and we have no study to cite for the exact limit.
How do you ask CSAT on a call?
Read a one to five scale aloud, say what the ends mean, and count the top two answers as satisfied. Customer satisfaction (CSAT) is the share of customers who rate the interaction as satisfied, usually a four or a five. Our call center metrics guide has the formula and a worked example, so we do not repeat it here.
On a call it sounds like this: "On a scale of one to five, where one means very poor and five means excellent, how would you rate the service you got on Monday?" Then stop and wait. Do not fill the silence with a hint.
Name the thing being rated ("your service visit on Monday", not "our service"), and repeat the number back ("so that is a four, thank you") to catch mishearing.
CSAT answers one question: how did that interaction go? It is a good fit for a call made soon after a visit or a support call.
How do you ask NPS on a call?
Ask: "On a scale of zero to ten, how likely are you to recommend us to a friend or colleague?" Net Promoter Score (NPS) is the percentage of promoters, who answer 9 or 10, minus the percentage of detractors, who answer 0 to 6. Passives, who answer 7 or 8, count in the base but not in either group.
The measure was popularised by Frederick F. Reichheld in "The One Number You Need to Grow", published in Harvard Business Review in December 2003. The article is behind a paywall, so we cite only its title, author and date, and take the question and the sums from Bain's own NPS page. You can find the article at Harvard Business Review.
Here is a worked example, made up. The chain in Tiruchirappalli gets 300 answers in a month.
- 150 promoters, which is 50%.
- 90 passives, which is 30%.
- 60 detractors, which is 20%.
- NPS = 50 − 20 = +30.
You can check that on a calculator. A zero-to-ten scale has eleven points, so it is harder to hold in the ear than one to five. Read it once, slowly, and say what zero and ten mean.
Which one should you pick? CSAT is for the interaction and NPS is for the relationship, as our metrics guide says. Do not ask both in full on every call. A short call can carry one scale and one "why".
What does a feedback call script sound like?
A feedback call script has seven parts: who is calling and why, a recording line, a permission ask with the count of questions, the rating, the reason, a read-back and a thank-you with what happens next. The script below does all seven in about 90 seconds. It is our own, made up, and your compliance team must approve any real one.
For the 4-line short form, see Template 11 in our call center scripts guide. This is the longer version, with the reasons behind each step.
A full script: service visit, CSAT and NPS
Agent: Vanakkam. Am I speaking with Mr Suresh Kumar?
Customer: Yes, speaking.
Agent: This is Divya from Anbu Motors, calling about your bike service on Monday.
This is a short customer survey, not a sales call. It may be recorded to keep a
record of your answers. Do you have two minutes?
Customer: Okay, quickly.
Agent: Thank you. I have three questions. First, on a scale of one to five, where one
means very poor and five means excellent, how would you rate your service visit?
Customer: Four.
Agent: A four, thank you. Second, on a scale of zero to ten, where zero means not at all
likely and ten means extremely likely, how likely are you to recommend Anbu
Motors to a friend or colleague?
Customer: Seven.
Agent: A seven. Last question. What is the main reason for your score?
Customer: The work was good but I waited two hours and nobody told me why.
Agent: Thank you. So the work was good, and the wait of two hours was the problem. I
will pass that to the service manager today. If you would like someone to call
you back about it, I can note that. Would you like that?
Customer: No, that is fine.
Agent: Thank you for your time, Mr Kumar. Have a good day.
Why it works: "not a sales call" keeps the call free of any offer. The recording line is advice, not a rule (see is call recording legal in India). The question count sets the end point, the scale meaning comes before the question, and one open question gives you a reason to act on. The read-back shows you listened, and the closing line says only what you will do.
Two lines in Hindi and Tamil
Customers answer best in the language they use with you. Two lines, in the Latin script, as an example:
- Hindi: "Aapka anubhav kaisa raha? Ek ka matlab bahut kharab, paanch ka matlab bahut accha."
- Tamil: "Ungal service anubavam eppadi irundhadhu? Onru endral migavum mosam, aindhu endral migavum nandru."
We have not had these two lines checked by a native speaker for a real script. Have a speaker of each language review any line before you use it, and do not translate word for word from English.
What if the customer is busy, or angry?
The customer is busy. Do not push. Say: "No problem. When would be a better time for a two-minute call?" Book a time, note it and call then. A callback you keep is worth more than a rushed answer.
The customer is angry. Do not read the next question. Say: "I am sorry about that. I want the right person to look at this. Can I pass your details to our service manager so that they call you?" Then hand it to a person. Our guide to handling angry customers on calls has the steps for the hard call.
IVR survey or a spoken survey: which should you use?
Use a keypad IVR survey for one number right after a call, and a spoken survey when you need reasons or when callers struggle with keypads. An IVR survey asks the customer to press a key. A spoken survey lets them talk. Our guide to IVR versus voicebots covers the difference between keypad menus and speech in general.
In our view, a keypad suits a fixed numeric answer, and spoken answers suit open questions. We have no study that proves it, so test both on a small group before you choose.
Right after the call, or days later?
Ask right after a support call, while the memory is fresh and there is no second dial. For a visit or delivery, a day or three later usually works better. The chain in Tiruchirappalli calls three days after a service visit, which is one chain's choice to test. Use the same gap for every customer so scores can be compared.
Who should you call, and how do you avoid a biased sample?
Pick customers at random from all closed tickets or visits, not the ones an agent chooses. Then count who did not answer and compare those who did with those who did not. A score from a biased sample can mislead you more than no score.
Three problems come up again and again:
- Agents choose the callers. If an agent picks who gets called, happy customers get picked. Let a random draw choose.
- Nonresponse bias. The very happy and the very angry tend to answer more often than the middle, as our call center metrics guide notes. Count the people who did not answer and report the response rate with the score. The American Association for Public Opinion Research publishes standard definitions for counting response rates, which is a useful reference for keeping the sum the same every month.
- Different groups answer differently. Compare the people who answered with the rest by city, product and ticket type. If one group is missing, your score is not about that group.
How big a sample do you need?
Enough that the margin of error is small enough to act on. The same normal approximation we use in the call audit guide works for CSAT. For a CSAT of 80%, the 95% margin of error is:
- 100 responses: 1.96 × √(0.8 × 0.2 ÷ 100) = 0.078, so about ±7.8 points.
- 300 responses: 1.96 × √(0.8 × 0.2 ÷ 300) = 0.045, so about ±4.5 points.
So at 100 responses, a CSAT of 80% could be anywhere from about 72% to 88%. A move from 80% to 83% sits inside that margin, so it may be noise. At 300 responses the margin is smaller, but a move of 3 points still sits inside it. Read the margin before you celebrate or panic over a small change. This assumes a random sample, which is why the first rule above matters.
The chain in Tiruchirappalli calls a random 300 of each month's visits. Pick a number that gives a margin you can live with, then keep it fixed.
When can you call, and what consent do you need in India?
We could not verify whether TRAI treats a satisfaction survey as a service call or as a promotional one. That is the first thing to say, and the question to put to your carrier and your compliance team. This is general information, not legal advice.
What we can point to:
- Service calls. Our guide to TRAI's new rules for business calls says a service call to your own customer, such as a delivery update, a warranty notice or an OTP, needs no explicit consent. Whether a survey counts is the part we could not confirm.
- An offer may change the class. Keep the call free of any offer or upsell. A pitch inside a survey can turn it into a promotional call. This is our reading, not a quoted rule.
- Number series. Service calls come from the 1600 or 1601 series, by sector. Use the series your provider assigns, as set out in our guide to 140, 160 and 1600 series numbers.
- Calling hours. We found no fixed window for promotional calls in the TRAI documents we read, and the 2026 memorandum we quote says there are no time restrictions for service and transactional calls. See what about calling hours and DND.
- Recording. Say that the call may be recorded, and see is call recording legal in India.
- Survey answers are personal data. Use them for the purpose you stated on the call. Our guide to DPDP rules for call centers explains what consent looks like on a call.
Which language should a feedback call use?
Use the language the customer used with you, and offer a switch. A rating needs no reading, but it must be understood, and the Gallup study found stronger order effects on harder questions.
- Start in the customer's language. If the service chat or bill was in Tamil, call in Tamil, and offer a switch ("Hindi or English?").
- Read numbers back. A number may be spoken in Hindi, Tamil or English in one sentence.
- Expect mixed answers. Our guide to Hinglish voice agents covers mixed-language speech.
Do not translate the English script word for word. Ask a native speaker to write it, then read it aloud once to test the scale.
How do you close the loop?
Closing the loop means the customer hears back and something changes. A rating that goes into a sheet and nowhere else tells the customer their time was wasted. These steps are our advice.
- Call back unhappy customers. A rating of one or two, or a detractor, gets a call from a person within a time you set, such as two working days.
- Share themes weekly. Group the "why" answers by cause, such as wait time or billing, and send the top three to the owners.
- Tell customers what changed, in the next survey call or on the bill.
How Bananaflow fits
Bananaflow is an AI voice agent platform. The features below can carry a short scripted call.
Campaigns run from a CSV with retries, calling time slots and a limit on calls at once, plus an A/B split between agent versions. The agent speaks English and 10 Indian languages and can switch language during a call, as listed on the languages page. After each call you get a transcript, a recording, extracted variables and an outcome code, and a QA step returns tags, sentiment and a summary. Reports show KPIs and trends and export to CSV. Call recordings are stored in India, and call data is not used to train or fine-tune models.
DND, TRAI and DPDP duties stay with your team, and your compliance team owns the script and the consent wording.
What should you remember about customer feedback calls?
Call when the answer is worth the call, ask few questions with the scale read aloud, and ask one open question about the reason. Sample at random, count who did not answer and read the margin of error. Keep the call free of any offer, check how your carrier classes a survey, and call back the unhappy customers. Reasons you act on are worth more than a score you only report.
FAQ
Frequently asked questions
What is a feedback call?
In this guide, a feedback call is a short phone call to a customer that asks them to rate a recent purchase, visit or support call and say why. The phrase is also used for a call about a job interview or for coaching a call centre agent. Those are different things and are not covered here.
How many questions should a phone survey have?
Three to five, and under two minutes. This is our advice, not a measured limit. Say the number at the start, so the customer knows when it will end. One rating question, one follow-up question that asks for the reason, and a thank-you is often enough.
What is a good response rate for a feedback call?
We could not find an Indian response rate we could verify, so we do not print one. Pew Research Center measured about 6% for typical US telephone polls of strangers in 2018, but a company calling its own customers is a different case. Measure your own rate the same way every month and compare like with like.
Is NPS or CSAT better on a call?
Use customer satisfaction (CSAT) for a single interaction, such as a service visit, and Net Promoter Score (NPS) for the relationship with the brand. Both fit in one short call if you ask the rating first and one open question about the reason after. Do not ask a long list of both.
Do I need consent for a feedback call in India?
We could not verify whether TRAI treats a satisfaction survey as a service call or as a promotional one. TRAI's rules say a service call to your own customer needs no explicit consent, but a call that carries an offer can turn promotional. Keep the call free of any offer and ask your carrier and your compliance team. This is general information, not legal advice.
Can a feedback call be automated?
Yes. An automated caller can read the questions and capture the answers, and a keypad menu can collect one number. Spoken answers suit open questions, because the customer can give a reason in their own words. Whatever you automate, offer a way to reach a person when the answer is a complaint.
Should I send an SMS or call for feedback?
Send an SMS or WhatsApp link when you have thousands of low-value orders and a score is enough. Call when the customer is high value or at risk, when you need the reason and not only the score, or when the customer does not read English or use email. This is our own framework, not a tested rule.




