There is no good survey response rate, and that is the useful answer

A good survey response rate runs anywhere from above 50% to around 3%, and which end you should expect depends on two things: how much your customer cares about what you sell, and whether something has just happened to them. Above 50% is normal shortly after someone buys a premium car. Around 3% is normal for a relationship survey that arrives on a schedule, referring to nothing. Both programmes are healthy. A single figure offered as "good" without naming the touchpoint describes neither end of that range.

Four of the five pages ranking for this question give a single number instead, and they do not agree with each other.

The definition, and two conventions that matter

Response rate is completed responses divided by delivered invitations. Send 2,000 email invitations, watch 100 bounce, receive 152 completions, and the rate is 152 over 1,900, which is 8%.

Two conventions decide whether that number means anything. Count delivered rather than sent, or a deliverability problem arrives on your dashboard disguised as an engagement problem and you spend a quarter fixing the wrong thing. And decide before launch whether partial completions count, then hold that rule still, because changing it mid-programme quietly bends the trend line you are trying to read.

Why the published numbers disagree

SurveyMonkey puts a good rate at 10% to 30%, with anything above 30% excellent, and attributes that to nothing. Pointerpro opens on a flat 33%, also unsourced. Delighted, the only page on the same results list publishing measured customer-experience data, reports 6% for email surveys from its own 2021 platform figures. And a 2022 meta-analysis of 1,071 academic online surveys lands on 44.1%.

The advice says good starts at 10% and excellence begins above 30%. The one measured CX figure on the same page is 6%. The academic literature reports 44.1%.

A CX lead runs a post-flight programme steady at 8%, reads that good starts at 10%, and walks into a quarterly review defending a number that is unremarkable for its touchpoint. Her counterpart at a premium vehicle brand reads the same page, sees his 25% sitting inside the good band, and never learns that comparable programmes run at twice that.

What decides a good survey response rate

Two things, and they multiply rather than add.

The first is how invested the customer is in the category. Buying a premium car is considered, expensive and identity-linked, and the buyer arrives at the survey still wanting to talk about it. A flight is a commodity the passenger mostly wants finished with. Survey craft does not close that gap.

The second is whether anything has just happened. A survey that arrives as a reply to a service visit yesterday, or a call twenty minutes ago, is part of a conversation. One that arrives on a schedule referring to nothing in particular is an interruption. Opinion decays quickly and the willingness to record it decays faster.

Multiply the two and you get the whole range. High engagement with no recent event produces a survey nobody has a reason to answer today. A well-timed survey about something the customer never cared about produces the same silence, for the opposite reason.

This is also why industry benchmark tables refuse to die and refuse to work. Industry is a real proxy for the first factor, and automotive genuinely does out-respond air travel. It carries no information about the second, so the same benchmark row gets quoted to a team surveying yesterday's purchase and a team running an annual relationship census.

Five touchpoints, counted

The figures below are from Numr client programmes, sorted by the moment the survey went out.

Each row is one delivered programme rather than an average across a category. The rows come from different brands, questionnaires, channels and years, field periods and sample sizes are client-specific and not published, and five programmes that differ in that many ways cannot isolate a variable between them. So read the table across as a ladder, not a benchmark: the question is which rung your touchpoint resembles.

Read that way, the top four rows carry the first half of the argument. All four are immediately post-event. A car has just been bought, a car has just been serviced, a call has just ended, a flight has just landed. Timing is doing its job in every one, and the spread still runs from above 50% down to around 8%, which at minimum means timing is not what separates them. Engagement is the candidate explanation, and across Numr's wider book it is the one that holds.

The second half is stronger, because one comparison in this set is controlled. The top row and the bottom row are the same automotive brand, the same customers and the same channel in the same quarter. Only the moment differs: shortly after a vehicle purchase, above 50%; on a generic relationship ping with nothing behind it, around 3%.

Same brand, same customers, same channel, same quarter. Change only the moment and the number moves by more than a factor of fifteen.

Between categories, engagement appears to set the spread, and that part is an inference from a pattern. Within a brand, the moment provably does.

What that does to the three published figures

Delighted's 6% is platform-wide email data blended across categories and moments. Mix rungs like these in ordinary proportions and you land in the mid single digits, which is where 6% sits, between a post-flight 8% and an event-free 3%. The 44.1% describes academic surveys, where respondents are recruited, often compensated, and answering on a subject they signed up to care about, which is both factors pushed favourable by design. And 10% to 30% is less wrong than unanchored: it happens to bracket the middle rungs while saying nothing about which rung you are standing on.

That reconciliation is an inference rather than a counted result. Treat it as a working explanation to test against your own first cycle.

The mistake this causes, and it is an expensive one

Picture a team looking at a relationship survey returning 3%. They have read that 30% is normal, so they conclude the questionnaire is the problem. They shorten it, rewrite the opening question, redesign the invitation email, and run an A/B test on the subject line. A quarter goes by and the number reaches 3.4%.

Nothing they did was wrong, and none of it touched the constraint. A questionnaire cannot manufacture a reason to reply. They were asking customers to comment on a relationship in the abstract, at a moment chosen by a calendar, about nothing that had recently happened to them.

The same survey attached to a service visit would have moved by a multiple rather than a rounding error.

The moment also decides who answers

A telephone survey arrives unrequested and captures whoever picks up, including customers who feel nothing much about you either way. A digital survey has to be chosen, and people choose to answer when they have an emotion attached to the answer. The strongly satisfied reply, the strongly dissatisfied reply, and the indifferent, who are usually the largest group you have, do not.

The middle is not missing from your customer base. It is missing from your response, which is why the distribution from a digital programme comes back bimodal with a thin middle.

Which is the practical reason a raw response rate tells you so little on its own. A 40% rate drawn entirely from customers who had a memorable experience can support a worse decision than a 12% rate that resembles your actual base. The percentage is the second question; who is inside it is the first.

What that costs you, in invitations

One number decides whether a sample is precise enough to act on, and it is not the rate. It is the count of responses. 385 responses gives a margin of error of about plus or minus 5% at 95% confidence, whatever proportion of invitations they represent.

Which turns the ladder into a budget line. To reach 385 responses you need at most about 770 invitations at the top rung, and around 12,810 at the bottom. The relationship-survey programme needs a contact list nearly seventeen times larger than the post-purchase one to reach the same confidence. That is the price of a badly chosen moment, and it is the only part of this argument that shows up in a procurement conversation.

One caveat, and it is the same one as before. No sample size repairs a sample skewed by who chose to reply, so the bigger list at the bottom of the ladder is buying precision rather than representativeness. Those 385 responses cost more and they are worth less.

The 385 itself carries conditions this page does not have room for. It assumes one proportion from an effectively infinite population, so a finite customer list needs fewer, a net score like NPS needs more, and every subgroup you cut has its own margin of error. How many survey responses do you actually need works through all of it.

The four levers, in order of impact

The rate is not fixed within its rung. Four levers move it, and the ordering is not close.

Speed comes first, by a wide margin. A survey that arrives within a couple of hours of the event catches the opinion while it still exists; after that you are surveying a memory. It is usually a question about your data plumbing rather than your questionnaire, which is why so many programmes leave it alone.

Language is second: offering the survey in more than one. The customer who would answer in their own language quietly abandons in yours, and never appears in the response data, so the size of this lever is invisible to the team deciding whether to pull it.

Channel is third. The effect is real and smaller than the two above, and it depends on where the touchpoint already lives. We publish no channel table for that reason.

Format is fourth. A chat survey pulls two to three percentage points more than a static form. Worth taking, not worth building a strategy on.

None of the four touches the first factor. Each either catches the opinion sooner or lowers the cost of giving it, and none makes a passenger care more about a flight. So the levers tune a programme within its rung rather than moving it up one.

Incentives are absent because the evidence conflicts. The 2022 meta-analysis above found no significant effect on response rate; Pointerpro's page tells readers incentives lift response by 10 to 15% and names no study behind the figure. Test before you spend.

Setting a target that means something

1. Name the moment before you name the number. Set targets per touchpoint, against that touchpoint's own baseline. One programme-wide target punishes whoever owns the low-stake moments and rewards whoever owns the showroom.
2. Take your first target from your own first cycle. You will have a real baseline after one send, and it will be worth more than any table on this page, including ours.
3. Work the levers in order. Speed, then language, then channel, then format.
4. Change one thing at a time. A rate that improved because you changed the method has not improved. It has become incomparable with everything measured before it.
5. Watch the shape of the response, not only its size. If your respondent mix stops resembling your customer mix, a rising rate is making the problem bigger rather than smaller.

So judge a rate twice: against the rung it resembles, to know whether the programme is broadly healthy, and against its own history, to know whether what you are doing is working. A cross-industry average compresses both factors into one figure and can do neither.

Frequently asked questions

What is a good survey response rate?

Anywhere from above 50% to around 3%, set by how invested the customer is in the category multiplied by whether a triggering event has just happened. Counted across Numr programmes, premium automotive surveyed shortly after purchase runs above 50%, post-flight airline around 8%, and relationship surveys with no triggering event around 3%. All three can be healthy programmes.

What is the average survey response rate?

Published averages run from 6% to 44.1% depending entirely on what was averaged. Delighted's 2021 platform data reports 6% for email customer surveys; a 2022 meta-analysis of 1,071 academic online surveys reports 44.1%. An average across those two populations would describe nothing that exists.

What is a typical survey response rate for a CX programme?

A post-call survey and an annual relationship survey differ by a factor of five, and two post-event surveys in different categories differ by six. If you need a planning number before your first send, find the closest match on both counts, similar customer involvement and a similar moment, then replace it with your own baseline after one cycle.

Is a 10% response rate bad?

10% is poor immediately after a contact centre call, unremarkable just after a flight, and strong for a relationship survey with no triggering event. There is no context-free failure threshold.

What is a statistically valid survey response rate?

There is no such thing, because no response rate makes a sample valid. Precision depends on how many responses you hold, not what proportion of invitations they came from: about 385 responses gives a margin of error of roughly plus or minus 5% at 95% confidence, whether they came from 770 invitations or 12,810. Two caveats matter more than the arithmetic. Sample-size formulas assume random sampling and cannot repair a sample skewed by who chose to reply. And NPS is a difference of two proportions, so it carries a wider margin of error than a single proportion at the same number of responses.

How is survey response rate calculated?

Completed responses divided by delivered invitations. Count delivered rather than sent, and fix your rule on partial completions before launch rather than after.

How do I increase my survey response rate?

Speed, language, channel, format, in that order. Send within a couple of hours of the event, offer more than one language, then look at channel, then at format. Change one thing at a time, or you will not know which one worked. And if the touchpoint is low-engagement with no triggering event, expect the levers to tune the rate rather than transform it.

Does channel change response rate?

Yes, and the relative ordering between channels is stable. We publish no channel table, because absolute figures are specific to the engagement and a table lifted out of its context becomes the next bad benchmark. Channel is a real lever and a smaller one than the moment.

---

Net Promoter Score and NPS are registered trademarks of Bain & Company, Fred Reichheld, and Satmetrix.