Most customer satisfaction surveys fail before anyone reads a single answer. A template gets copied from somewhere, a few questions get added because a manager asked for them, and what started as three questions arrives in the customer's inbox as eighteen. The response rate drops, the people who do reply are the angriest and the most loyal, and the average score that comes out the other end is used in a meeting as though it meant something.
The fix is not a better template. It is a shorter one, plus a decision made in advance about what happens to each answer. This guide gives the question set, the reasoning behind each question, the scale choice, the timing, and the part that templates never cover: what to do with a bad score at four in the afternoon on a Friday.
Decide what the survey is for before writing a question
There are two jobs a satisfaction survey can do, and they pull in opposite directions.
The first is measurement. The goal is a number that can be compared with last quarter's number, so the questions have to stay identical over time and go to a large enough slice of customers that the result is not noise. Changing the wording breaks the comparison. Adding a question this quarter and removing it next quarter makes the trend meaningless.
The second is recovery. The goal is to find the specific customer who had a bad experience while there is still time to fix it. Here the score barely matters. What matters is that a low score reaches a named person within hours, with enough context for that person to do something useful.
A single survey can serve both, but only if it is designed for the second job first. A measurement survey that nobody acts on is a reporting exercise. A recovery survey produces a usable trend as a side effect, because the scores are still being collected.
The practical consequence: before writing the questions, write down who receives a low score, how fast, and what they are expected to do. If that sentence cannot be finished, the survey is not ready to send. Teams that skip this step end up with a dashboard nobody opens and a folder of complaints nobody answered.
The template, question by question
Five questions. Two required, three optional. Nothing else earns a place in a survey sent to a paying customer.
Question 1: overall satisfaction
How satisfied were you with [the specific thing]? Very dissatisfied, dissatisfied, neutral, satisfied, very satisfied.
Name the thing precisely. "How satisfied were you with your recent order?" and "How satisfied were you with how the repair was handled?" produce different answers from the same customer on the same day. A survey that asks about the company in general gets a rating of the customer's mood.
Question 2: the open box
What is the main reason for that score? One text box, no character limit, optional.
This is the question that pays for the survey. The score tells you there is a problem and the text tells you what it is. Keep it directly after the score so the answer is anchored to it, and resist the urge to ask three open questions instead of one. People will write one paragraph. Asking for three gets one paragraph in the first box and blanks in the other two.
Question 3: effort
How easy was it to get what you needed? Very difficult through to very easy.
Satisfaction and effort measure different things, and effort is usually the more actionable of the two. A customer who got a good outcome after four emails will often rate satisfaction highly and effort poorly. That gap is where the process problems live.
Question 4: likelihood to recommend
How likely are you to recommend [company] to someone else? A 0 to 10 scale.
This is the question behind the metric most executives already know, which makes it useful for reporting even when the operational signal comes from questions 1 and 3. Include it if someone upstream needs it, and leave it out if not. It does not tell frontline staff anything they can act on.
Question 5: permission to follow up
Would it be all right to contact you about this response? Yes or no, with an email field that is prefilled if the sender is known.
Adding this one line changes the survey from an observation tool into a service tool. It also removes the awkwardness of replying to feedback from someone who thought they were answering anonymously.
Pick a scale and never change it
Scale choice matters less than consistency, but there are real differences worth knowing before the choice gets locked in for years.
| Scale | Reads well on | Good for | Cost |
|---|---|---|---|
| 5 point labelled | Phones, all languages | Operational surveys, high volume | Coarse. Small improvements hide inside a point |
| 7 point labelled | Desktop | Research, fewer respondents | Middle labels get read inconsistently |
| 0 to 10 numeric | Anywhere | Executive reporting, benchmarking | Numbers mean different things in different cultures |
| Thumbs up or down | Anywhere, in an email body | Fast pulse after a support reply | No middle. Loses the mild cases |
A labelled five point scale is the safest default for a survey attached to a transaction. Every option has words on it, so nobody has to guess whether 3 out of 5 counts as acceptable, and it renders on a narrow phone screen without wrapping.
Two rules apply whatever the choice. Label every point rather than only the ends, because an unlabelled midpoint is interpreted differently by different people and the difference does not average out. And avoid switching the direction of the scale between questions. A survey where one question runs from bad to good and the next runs from good to bad produces contaminated data that looks clean.
If the scale has to change, change it at a clean break, keep the old question running alongside the new one for a few weeks, and note the date in whatever report the numbers feed into. A trend line with an unmarked methodology change in the middle of it will be used to argue for the wrong thing.
Timing beats wording
When the survey arrives affects the answer more than how the questions are phrased.
For a transaction, send it while the experience is still specific. Same day for a support interaction, within two or three days for a delivery, and after the first real use for a product rather than at the moment of purchase. A survey that arrives three weeks later is asking about a memory.
For a relationship, a fixed cycle works better than an event trigger. Twice a year to the main contact at each account, asked at the same point in the cycle every time, produces a trend that can be compared. Quarterly is defensible for large accounts and starts to feel like harassment for small ones.
The limit that matters most is frequency per person, not per campaign. Set a rule, write it down, and enforce it in the sending tool rather than in a checklist: no more than one survey per customer per quarter, regardless of how many teams want to send one. Without that rule, a customer who buys twice and raises a support ticket gets three surveys in a month and stops opening any of them.
One more timing decision that gets forgotten: what happens when nobody replies. A single reminder, sent three or four days later to non responders only, typically recovers a meaningful share of responses. A second reminder annoys people for very little gain.
The part the template does not cover
Everything above is the easy half. The half that decides whether the survey was worth sending is what happens in the hours after a response arrives.
Three things have to be true. Each response needs an owner, so a low score is somebody's job rather than everybody's. Each response needs a status, so the difference between seen and handled is visible without asking. And the reply has to be sent from the same place the response is read, because the moment the work moves into a personal inbox, the record of what was said stops being shared.
A spreadsheet can hold the answers. It cannot hold those three things, which is why a spreadsheet based process collapses at around the point where two people are involved. The usual failure is not that feedback gets ignored on purpose. It is that a response gets read by two people who each assume the other replied, or gets answered twice with different information.
This is the specific gap a form tool with response management fills. The survey submission arrives with an owner and a stage attached, the reply goes out from the same screen as the answers, and the history of who changed what sits with the record rather than in someone's sent folder. The practical test when comparing tools is simple: open a single response and count how many other places you have to visit before the customer has an answer. Reviewing what the tools after the form actually do is more useful here than comparing question types, because every form builder can render a five point scale.
Repeat surveys create a second requirement. Responses from the same email address over two years should join up into one history, so the account that scored 2 last time and 4 this time can be seen as one story rather than two unrelated rows.
Read the results without fooling yourself
A few habits separate a survey that changes decisions from one that decorates them.
Report the distribution, not only the average. An average of 3.5 can be a room full of mild indifference or a split between delight and fury, and the two call for completely different responses. Showing the count at each point takes one extra column and prevents the most common misreading.
Watch who is missing. Response rates are rarely even across customer types, and the segment that stops answering is often the one about to leave. A falling response rate in one segment is a signal in itself, not a data quality problem to be corrected with more reminders.
Count the actions, not the responses. The number worth putting on a wall is how many low scores were contacted and resolved, and how long that took on average. A satisfaction score that goes up while nobody can name a single change made because of the survey usually means the sample changed, not the service.
And read the text. Twenty open answers read properly will point at the real problem faster than four thousand scores. Tagging them by theme as they arrive, rather than in a quarterly batch, is what turns comments into a queue somebody can work through. The questions people ask about running this cover the mechanics of tags and stages in more detail.
What to change first
Start with question 2, the open box, and the permission line in question 5, because those two together turn a score into a conversation you are allowed to have. Then name one person as the owner of every response below the midpoint, with a target of replying within one working day. When that is in place and the volume starts to hurt, see what one screen from response to reply looks like in Halict.
Q1. How many questions should a customer satisfaction survey have?
Two to five. One rating, one open box asking why, and optionally an effort question, a recommendation question and permission to follow up. Every question past the fifth measurably lowers completion, and the answers to questions six through twenty are almost never the ones that drive a decision.
Q2. What is a good response rate for a customer satisfaction survey?
It varies far too much by channel, industry and relationship for a single benchmark to be useful, so the number worth tracking is your own rate over time rather than a figure from a report. A rate that is stable and a rate that is falling tell you different things. Sudden drops usually mean a timing, deliverability or frequency problem rather than a change in sentiment.
Q3. Should the survey be anonymous?
Anonymous surveys get slightly more candid answers and make recovery impossible, so the usual compromise is to identify the respondent and ask permission before contacting them. If anonymity is promised, honour it completely, including not inferring identity from the account the link was sent to.
Q4. How often can the same customer be surveyed?
Set a hard limit of one survey per person per quarter and enforce it in the sending tool rather than by convention, since several teams will each want to send one. Transactional surveys after support interactions can be more frequent if they are a single question, but they still count against the limit for longer surveys.
Q5. Can one template be used for both a transaction survey and an annual relationship survey?
The rating question can be shared so the two remain comparable, but the rest should differ. A transaction survey names the specific event and stays under five questions, while a relationship survey asks about the overall working arrangement and can carry a few more. Keeping the core rating identical lets both feed the same trend.