Somewhere in your sent folder is a reply that says something like "honestly, the only reason we nearly left was the export thing." Somebody answered it, the customer was happy, the conversation closed. That sentence is now unfindable. Nobody logged it, nobody counted it, and in two months the same complaint arrives from a different customer and gets handled as if it's new.
That's the real feedback problem at 2 to 15 people. It isn't that customers won't tell you things. They're telling you constantly, in the queue, and it evaporates because no step in the working day captures it. Fixing that is less about picking a survey tool than about deciding who owns the sentence after the ticket closes.
The short version
If you read nothing else, this is the whole system:
- Treat replies in your support queue as your first channel. They're already arriving and they cost nothing to collect.
- Add exactly one deliberate ask, not five. One question at the end of a resolved thread beats a survey link.
- Ask after resolution, never mid-thread, and never twice in 30 days to the same person.
- If you ask for public reviews, ask everybody at the same trigger point, and keep rewards out of it.
- Name one person who owns what comes back, with a deputy for their holidays.
- Run a 20 minute review every week that ends in a written decision, including the things you're declining.
Start with the feedback you're already getting
What counts as customer feedback?
Two kinds of feedback reach you, and the difference decides what you can do with each. Solicited feedback is what you get because you asked, which means you picked the topic and can compare answers over time. Unsolicited feedback is what arrives without an invitation: complaints, cancellation notes, a throwaway line in the middle of a support reply. It's messier to work with and usually the more urgent of the two, because somebody had to care enough to type it unprompted.
The channels, and what each one actually tells you
The survey is where most teams start. At this size it should be second, because the queue is already full of unsolicited feedback that nobody is reading as feedback.
| Channel | What it tells you that nothing else does | Effort to keep running | Who owns it | How it misleads you |
|---|---|---|---|---|
| Replies in your support queue | What broke, in the customer's words, with the context of a real problem attached | None. It's already arriving | Whoever answers the thread | Only represents people who bothered to contact you |
| A one question ask after resolution | Whether your fix actually landed for that person | Low. One line at the end of a reply | The person closing the conversation | Rates the last interaction, not the product |
| A short scheduled survey | What customers who never write in are thinking | Medium. Writing it, sending it, reading it | One named owner | Answered mostly by fans and by the furious |
| Public reviews | What prospects will read about you before they talk to you | Low to ask, ongoing to read | Founder or whoever owns marketing | Shaped heavily by who you asked and when |
| Cancellation reasons | Which problems actually cost money | Low. One required field on the way out | Whoever processes cancellations | People are polite while leaving. "Too expensive" often means "not worth it yet" |
| A prompt inside the product | What customers who never write in or answer email think | High to build, then almost none | Whoever owns the product surface | Catches people mid-task, so it favours quick reactions over considered ones |
| Unprompted mentions elsewhere | What customers say when they aren't talking to you | Medium. Somebody has to go and look | One person on a fixed weekday | Loudest voices, tiny sample |
Two of those rows are the ones that pay off first, and they're the two you already have: the queue, and one small ask bolted onto the end of it. A complaint is feedback too, though answering the complaint itself is a different job from logging what it tells you.
Pick two channels, not eight
Every channel you add has a running cost that shows up in your worst week, which is exactly when you stop doing it. So the test isn't "is this channel valuable." It's "will this still be happening in March."
So the real test is whether a channel will still be running in March. Three questions settle it:
- Can you read all of it in an hour? If the honest answer is no, you'll skim it, and skimmed feedback turns into whatever you already believed.
- Does it survive a bad week? Anything that needs a person to sit down and compose something is the first thing dropped when the queue backs up. Anything that's one line inside work you're doing anyway will survive.
- Does it reach people who don't contact you? The queue can't. That's the one real argument for a scheduled ask, and it's why one is worth running.
For most teams this size the two are the queue and one deliberate ask, usually the post-resolution question above. A voice of the customer program spanning six channels is the wrong shape here: you'd spend the effort on collection and have none left for acting, which is the only part customers ever notice.
Two things on that list aren't really channels for budgeting purposes, and it's worth being clear about which. A cancellation reason field is a one-time setup with no ongoing cost, so add it regardless. Review requests are a marketing decision rather than a feedback channel, and they come with rules of their own (next section). The channel that genuinely reaches customers who never write in and never answer email is a prompt inside the product, which is why it's in the table above; it needs engineering time you may not have, so treat it as the thing you add third, if at all.
Writing an ask people answer
The strongest ask a small team has is one question at the end of a thread that's already going well, sent by the person the customer has been talking to, about the thing they were just talking about. Concretely, it's one line under your closing paragraph:
"Before I close this off: was there anything about the way we handled this that we should do differently next time?"
That's answerable in one sentence, it asks about the handling rather than about feelings, and it says out loud that criticism is the useful answer, which is what stops people being polite at you. Compare it with "are you happy with our service?", which invites a courtesy yes, or "how much did you like the new export?", which tells the customer which answer you want. Pew Research Center's guidance on writing survey questions is blunt about why that matters: "even small wording differences can substantially affect the answers people provide." Two more of its points earn their place in even a five question form. The open versus closed choice is "one of the most significant decisions that can affect how people answer questions." And order effects are real, because "earlier questions can unintentionally provide context for the questions that follow," so don't put your satisfaction question directly after four questions about last week's outage.
When you do send a real questionnaire, keep it short and don't announce it. A Cochrane systematic review by Edwards and colleagues, updated in November 2023, pooled 88 randomised trials of ways to increase response to electronic questionnaires. Shorter e-questionnaires raised the odds of a response by about half (OR 1.51), a brief covering note rather than a detailed one tripled them (OR 3.26), and monetary incentives came close to doubling them (OR 1.88). The cheapest finding to act on is the least obvious: putting the word "Survey" in the subject line reduced the odds of a response (OR 0.81).
Scope that honestly before you bet on it. Those trials are mostly epidemiological and clinical research questionnaires rather than support surveys, and the heterogeneity behind the shorter-questionnaire figure is high (I2 = 94%), meaning the size of the effect varied a lot between trials. The direction is about as well established as anything here gets; the exact multipliers aren't.
Which score you put on the end of it is a separate question, and this page doesn't answer it. Whether you use CSAT, NPS or something you invented, and how to calculate and read it, is its own decision with its own arithmetic.
When to ask, and when to shut up
Timing does more for response rate than wording does, and it's mostly about which state the conversation is in.
Ask:
- Right after a resolution the customer has confirmed, in the same thread, from the same person.
- After a completed onboarding or a first successful use of the thing they bought.
- On cancellation, as one required field. It's the only moment you'll ever get that answer.
Don't ask:
- Mid-thread, while anything is still open. It reads as though you've stopped working on it.
- Straight after an escalation, even a successful one. Wait for the next ordinary interaction.
- Twice within 30 days to the same person. Pick a cap and enforce it in the tool, not in your memory.
- Immediately after a complaint. Give it a few days, or you're measuring how annoyed they still are. If they're still angry, that's a different conversation and a survey link will make it worse.
The frequency cap is the rule people skip and regret. Two asks in a fortnight from a company somebody already told what they think is how you teach customers to ignore you permanently.
Reviews and the rules you don't get to choose
Asking customers for public reviews is normal and fine. Two constraints on how you ask aren't matters of taste, and they don't fully agree with each other.
The FTC's Consumer Reviews and Testimonials Rule went into effect on October 21, 2024. Incentives themselves are still allowed: the FTC's guidance says the rule "does not prohibit giving incentives for reviews, as long as there isn't an express or implied requirement that the reviews have to express a particular sentiment." The part that catches people is the implied half: you "can't suggest to consumers that their reviews must be positive (or negative)" to get a promised incentive, the FTC adds, "even if you don't say so explicitly." So the test is whether the reward is conditioned on the rating, not whether both appear in the same email. "10% off if you leave us a 5 star review" is out. "10% off when you leave a review, whatever you say in it" is the same offer without the condition, and it's the version to send.
Asking only your happy customers is the other habit worth getting right, and the guidance here is narrower than it's usually reported to be. On soliciting reviews only from satisfied customers, the FTC says the rule "does not contain a specific prohibition against such conduct," while noting it could still violate the FTC Act. So it isn't automatically a violation of this rule, and it also isn't safe, which is a good reason to build the simpler habit: ask everybody at the same trigger point, or don't run review requests at all. You also get better information that way.
If the reviews you're chasing sit on a Google Business Profile, Google's own policy is blunter than the FTC's, and this is where the two diverge. Asking is fine: "to leave reviews, you can ask customers to visit a Google link or scan a QR code." But "offering incentives, like free or discounted goods or services, in exchange for customers to post reviews, change reviews, or remove negative reviews is considered fake & misleading content and is strictly prohibited," per Google's guidance on getting reviews. There's no sentiment condition in Google's version, so the unconditional discount that satisfies the FTC still breaks Google's policy. If you're asking for reviews there, ask without a reward attached.
Who owns feedback once it lands
Collection is a habit. Customer feedback management is the part that comes after, and it's five things: one person, one place, one weekly slot, one written decision and one reply. Four of them belong to the owner. The last belongs to whoever owns the conversation the feedback arrived in.
One owner, plus a deputy. Not "the team" and not "product." One person, named, whose weekly job is to read what came in, keep the counts current, run the review, and send the replies that close the loop. Name the deputy in the same sentence you name the owner, because the first week this breaks is a holiday week.
One place, chosen by mechanism. Feedback that arrives as a reply stays where it arrived and gets a tag, so the original thread and its context travel with it. Everything that arrives outside email, such as a review, a call note or something said in a meeting, gets one line in one sheet. Two places total. Anything more and you'll have three half-populated systems by spring.
Keep the tag set small enough to apply without thinking. Five works:
| Tag | What it means | Goes to | In what form | Closing the loop looks like |
|---|---|---|---|---|
bug |
It's broken and we agree it's broken | Whoever fixes things | Link to the thread plus what the customer was trying to do | "This is fixed as of today, and your original case was what found it" |
missing |
The product doesn't do a thing they need | Product owner | The customer's own words, plus how they work around it now | "We're doing this in the next release" or an honest no |
pricing |
Packaging or price blocked something | Founder | Which plan, what they wanted, what they compared it against | Usually no reply. It feeds the pricing decision |
confusing |
It works and they couldn't find it | Docs owner | The exact wording they used for the thing they couldn't find | "We rewrote that page, here's the bit you needed" |
praise |
Something worked notably well | Owner keeps the list | One line and who said it | Not a reply. It tells you what not to break |
Tag the conversation, not every message, and only tag when a customer told you something about the product or the service. A question you answered isn't feedback. The same question asked eleven times is, and it belongs to your documentation rather than to your roadmap.
A weekly slot, 20 minutes, with an agenda. Same time every week, in the calendar, owner runs it, with four boxes: five minutes reading the new tagged items, five minutes on the counts and which tag moved, five minutes picking exactly one thing to act on, five minutes writing down the decision. You need a threshold for that third box or the loudest recent ticket always wins. A workable default at this size: three separate customers on the same tag in a month makes it systemic and puts it in the running, while a single report stays a single report and gets handled in its own thread.
A decision record. One line per week in the same sheet: date, what you picked, and what you decided against. Recording the declines is what stops the same suggestion being re-litigated every month, and it's what you read from when a customer asks whatever happened to their idea.
Closing the loop, including when the answer is no. The person who owns the thread replies, not the owner of the system. "Great idea, I'll pass it to the team" closes nothing. "We're not building this, here's the workaround, and I'll email you if that changes" is a real answer. Doing this consistently is also the cheapest input to customer loyalty available to a small team, though loyalty work itself is a bigger job than feedback.
Mechanically, all of this depends on the tag and the owner being visible on the conversation to everybody, which is awkward when the support address is a mailbox several people log into. Tags live in one person's client, and the thread nobody claimed sits there looking claimed. That's what a shared inbox is for, and a shared inbox tool like TriageFlow keeps the tag, the assignee and the history on the conversation itself rather than in somebody's private labels.
Doing it with what you already pay for
You almost certainly don't need to buy anything to start. Your Google Workspace or Microsoft 365 subscription already includes a forms tool, and both have ceilings worth knowing before you build on them.
Google Forms will email the form for you: turn on "Notify People" when you share it. Two settings decide what you learn. Respondent identity isn't recorded by default, and Google's help page on sharing and publishing forms puts it plainly: "their usernames won't be recorded unless you turn on the 'Responses' setting to collect email addresses." And if you switch on "Limit to 1 response" to stop duplicate submissions, "to access and fill out the form, users must sign in to their Google Account." For a B2B audience that's usually fine. For consumers it's a large hole in your sample, so choose it knowing the cost.
Microsoft Forms takes up to 200 questions per form and 4,000 characters per question, which no sensible customer survey will trouble. The response cap is the one to check, because it depends on the account that created the form: work or school accounts allow up to 5,000,000 responses, while a personal Microsoft account allows 200 on the free tier and 1,000 paid (government tenants sit at 50,000). Build the form on somebody's personal login and a 900 customer send can run into a 200 response ceiling, so confirm whose account it's on before you send it.
The other ceiling is the send itself, and it's worth reading carefully because the two numbers people mix up are a per message cap and a per day cap. Gmail on Workspace allows 2,000 messages per day and 2,000 recipients per message, with a maximum of 500 external recipients in any one message. Every customer counts as external, so a list of 1,500 has to go out as at least three messages. All three can go the same morning: the daily ceilings are 3,000 external recipients and 2,000 unique external recipients. The exception is a free trial account, which caps at 500 messages a day and 500 unique external recipients, and Google notes those limits aren't raised during the trial.
When you do outgrow the free path, judge candidates on mechanics rather than feature lists:
- Does a response land back where the conversation already lives, or in a separate dashboard somebody has to remember to open?
- Can you export raw responses, including the free text, without a plan upgrade?
- Can it cap how often one customer gets asked, automatically?
- Can it tie a response to the specific conversation or order it followed?
- Can you delete one customer's responses on request, in a couple of clicks?
The deletion question is the one nobody checks during a trial and everybody needs eventually, so check it during the trial.
Frequently asked questions
What's the best way to collect customer feedback?
Whichever channel your customers are already using. For a small support team that's the reply thread, because response rate beats sample design at this size: an answer from somebody mid-conversation with you is worth more than a perfectly designed survey nobody opens.
How often should you ask customers for feedback?
Set a cap per customer and a trigger rule, then let the triggers do the work. A reasonable starting point is no more than one ask per customer per month, fired by events (resolution, onboarding finished, cancellation) rather than by the calendar. Published cadence advice ranges from "after every interaction" to "twice a year", which tells you there's no settled answer, so treat your own response rates as the authority.
What's a good survey response rate?
There isn't a benchmark here worth quoting. Most of the percentages in circulation are one platform's own customer data, sliced differently each time, so compare against your own last three asks instead. If you want to move the number, the Cochrane evidence above points at shortening the questionnaire and keeping "Survey" out of the subject line, with the caveat that those trials weren't support surveys. For how support benchmark numbers get made in general, we went through that on response times.
What tools do you use to collect customer feedback?
Whatever is already in your subscription, plus tags on your support address. The more useful question is when to replace that, and the trigger is rarely features: it's the day somebody is copying responses between two places by hand, or the day you can't answer "what did customers say about pricing last quarter" without opening a spreadsheet and a mailbox. Buy then, against the mechanical criteria in the section above, and check the data deletion one before you commit.
What are some best practices for collecting customer feedback?
The six numbered rules at the top of this page are the whole list, and the two that get skipped are the two that matter: cap how often you ask the same person, and put one name against what comes back. Everything else on this page is detail underneath those.
What is customer feedback management, and how is it different from collecting feedback?
Collecting is the asking; managing is everything that happens after the answer lands, and they fail differently. A collection problem looks like an empty inbox after a survey. A management problem looks like a full one: plenty of feedback, no owner, no record of what anybody decided, and the same suggestion arriving for the fourth quarter running. The five-part system above is the management half, and it's the half worth building first, because collecting more input you don't process just makes the pile bigger.