How to Target a Survey by What Customers Said: 2026 Guide

Your quarterly NPS survey goes to the whole customer list. Thirty of those customers spent part of their last call complaining about SSO. They get the same five generic questions as everyone else, and their answers disappear into an average. This guide shows how to survey the customers who complained, and only them, using a spreadsheet and a free form tool. It also covers the point where tools such as BuildBetter take over the manual matching work. The method comes from how B2B product and CS teams at BuildBetter customers run decision-focused research.

What Is a Targeted Survey? (Surveying Customers Based on What They Actually Said)

A targeted survey is a survey sent only to the customers whose own words show they are relevant to a specific decision. You choose the audience by evidence, such as a complaint on a sales or CS call, a support ticket about a feature, or a competitor mention, rather than by plan tier, list membership or random sample. You then tailor the questions to that evidence.

The difference fits in one line. A list-based survey asks "what does everyone think?" A targeted survey asks "what exactly do the people who raised this problem need for us to decide?"

The raw material for targeting is the signal. A signal is one discrete thing a customer said or did that shows a need, complaint, request or risk. Common examples:

• A quote from a sales or CS call ("our IT team won't sign off without SAML")
• A support ticket about a feature
• A message in a shared Slack channel
• A churn or downgrade reason
• A usage drop on a feature you're about to change

Signal-based survey targeting reverses the usual workflow. Traditional surveys start from a list (all customers, one tier, a random sample) and then write questions. Targeted surveys start from a decision and the evidence behind it, then pick the audience.

This guide gives you a 7-step method you can run today with a spreadsheet and any survey form. It also includes a worked pricing example, the mistakes that break the method, and the point where manual work stops scaling. The scope is B2B product, customer success and founder-led teams with a known customer base. It does not cover anonymous consumer panels, where you have no prior conversations to target from.

Why Targeting by Customer Evidence Is Hard (and What Breaks)

Targeting by customer evidence is hard because the evidence lives in conversations, and survey tools only see contact lists. Six failure modes show up again and again:

• The evidence lives where surveys can't see it. Complaints sit in call recordings, ticket threads and Slack channels. Survey tools know email addresses and CRM fields such as plan, ARR and region. No CRM field means "complained about SSO on a call."
• Keyword search lies. Searching tickets for "price" returns customers discussing their own pricing pages, invoice typos and the word "priceless." Searching for "SSO" misses "Okta login is broken" and "our IT team won't approve this without SAML." A person has to judge each match. Counting hits is not enough.
• The loudest customers are not a random sample. Customers on frequent calls tend to be larger, more engaged accounts. Targeting by call evidence alone over-represents them and misses quiet churners entirely. This is selection bias. The 1936 Literary Digest poll collected about 2.4 million responses from car-owner and telephone lists and predicted Alf Landon would win. Roosevelt took about 61% of the vote and 46 of 48 states. A skewed sample stays skewed no matter how large it gets.
• Over-surveying. The same complaining champion gets the NPS survey, a feature survey and a CSM check-in in one month. Response quality drops and goodwill erodes. Research on survey fatigue (Porter, Whitcomb and Weitzer, 2004) found that a prior survey measurably lowered response to later ones. Pew Research Center telephone response rates fell from 36% in 1997 to 6% in 2018.
• Generic questions waste the targeting. If you pick the right 30 people and then ask "How satisfied are you with authentication?", you learn nothing their call didn't already tell you.
• No loop back to the original signal. Answers get analyzed on their own, so nobody checks whether the survey confirmed, contradicted or sharpened what the customer said on the call.

The Method: 7 Steps to Target a Survey by What Customers Said

This works with a spreadsheet, your call notes or transcripts, your ticket export, and any free form tool. No budget required.

Step 1: Start from the decision, not the topic

Write one sentence: "We will decide X by date Y, and the answer changes if we learn Z." For example: "By March 1 we decide whether SSO moves from Enterprise to Business tier. We change course if Business-tier accounts say SSO is a buying blocker rather than a nice-to-have." That framing borrows from the Kano model, which separates must-be features from attractive ones. If you can't write this sentence, you aren't ready to survey.

Step 2: Define the audience by signals, not attributes

List two to four signal types that make someone relevant:

• Raised topic X on a call in the last N days
• Opened a ticket about Y
• Mentioned a competitor in the context of X
• Gave X as a churn or downgrade reason

Set a lookback window. Sixty to ninety days is a sensible default for product decisions. Before searching, write down synonyms and phrasings for each topic. For SSO, that list includes SAML, SCIM, Okta, Azure AD, "IT won't approve" and "login is broken."

Step 3: Build the match list in a spreadsheet

Use these columns: account, contact, signal type, date, source (call, ticket, Slack), verbatim quote or link, and the matcher's initials. Pull from call notes or transcript search, ticket exports and relevant Slack channels. Add one row per signal, then dedupe to one row per person.

Step 4: Review each matched person against the evidence

Read the quote. Ask whether it actually shows the need or just mentions the word. Mark each row Keep, Drop or Unsure. Rule of thumb: if you can't explain in one sentence why this person is on the list, drop them. Most of the quality in this method comes from this step.

Step 5: Exclude and balance

Remove three groups:

• Anyone surveyed in the last 30 to 60 days
• Accounts with open escalations, where a survey would be tone-deaf
• Contacts who have left the account

Then deliberately add a small comparison group: accounts that match the segment (same tier, similar size) but have no signal. This group tells you whether the complaint is specific to the people who raised it or common across all your customers.

Step 6: Tailor the questions to the evidence

Reference what they said without quoting them back or making them feel watched. Keep it to three to six questions. Vendor research from SurveyMonkey found respondents spend far less time per question in long surveys than on the first question, so short surveys get more careful answers. A good set includes:

• One question that confirms the problem
• One that sizes it
• One that tests the option you're deciding between
• One open-ended "what would change your mind?"

Ask about past behavior and specifics rather than hypotheticals, as Rob Fitzpatrick argues in The Mom Test. Always offer "neither" or "something else," and randomize option order where your tool allows.

Optional tip: before sending, do a dry run. Have two teammates read the questions while pretending to be specific matched customers. They will catch leading or confusing wording quickly.

Step 7: Send through the right channel, then compare answers to the original signals

Use email for low-frequency or admin users and in-app for active users. For top accounts, send a personal note from the account owner. After responses close, add answer columns next to the original signal in the same spreadsheet. Code each answer as one of four outcomes:

• Confirmed: the answer matches the original signal
• Contradicted: the answer goes against it
• Sharpened: the answer is more specific than the call
• Unclear

The decision rests on the pattern of codes, not the average score.

Checklist recap

1. Write the decision sentence: decide X by Y, changes if we learn Z.
2. Define two to four signal types, a lookback window and a synonym list.
3. Build a one-row-per-person match list with the quote that matched.
4. Review every match: Keep, Drop or Unsure.
5. Exclude recent respondents, escalations and leavers. Add a no-signal comparison group.
6. Write three to six evidence-based questions with a "neither" option.
7. Send by the right channel, then code each answer against its original signal.

Worked Example: A Pricing-Change Survey Targeted at Accounts That Raised Pricing on Calls

This is an illustrative scenario with round, made-up numbers. It is not a customer case study. It shows how a pricing change customer survey plays out step by step.

The decision (Step 1)

A B2B SaaS team with 410 active accounts is considering moving from per-seat pricing to a platform fee plus usage. The decision is due at the end of the quarter. They will change course if affected accounts say predictability matters more than total cost.

Audience by signals (Step 2)

The signals are pricing, cost, budget, seats or renewal price raised on any sales or CS call in the last 90 days. They also include tickets tagged billing that discuss price, excluding invoice errors.

Build, review, exclude, balance (Steps 3–5)

• The first pass returns 58 accounts.
• Review drops 9 that mentioned price about their own customers or an invoice typo, leaving 49.
• Seven were surveyed in the last 45 days and 2 have open escalations, leaving 40 signal accounts.
• The team adds a comparison group of 20 similar-size accounts with no pricing signal, for a total send of 60.

Tailored questions (Step 6)

1. "When pricing came up recently, was the main concern the total amount, how it scales with seats, or not knowing next year's bill?"
2. "Roughly how many people use the product but don't need full access?"
3. Two pricing models shown side by side: "Which would you prefer at renewal, and why?"
4. "What would make a price change feel fair to your finance team?"

Results (Step 7)

Twenty-three of 40 signal accounts respond, plus 8 of 20 comparison accounts. The team codes each answer against the original call quotes:

• 15 Sharpened: "too expensive" on the call turned out to mean paying full seats for occasional viewers or contractors.
• 5 Confirmed: total cost is the issue.
• 3 Contradicted: the concern was resolved after a discount at renewal.
• Comparison group: 6 of 8 are indifferent to the pricing model.

The decision

The team does not do the full pricing-model overhaul. Instead it adds a low-cost viewer seat and keeps per-seat pricing. The overhaul would have disrupted the large majority of accounts that never raised pricing, to solve a problem a narrower change fixes.

A list-based survey would have produced a "pricing satisfaction" average across 410 accounts. That average would have looked mildly negative and pointed at the wrong fix.

Targeting raised response quality, and more importantly, it changed which problem the team thought it had.

One caution: 23 answers are enough to diagnose the kind of pricing concern (total, scaling, predictability). They are not enough to set price points. Van Westendorp, Gabor-Granger and conjoint studies need larger, more representative samples.

Common Mistakes When Targeting Surveys by Customer Feedback

Most targeted surveys fail because of judgment errors, not missing software. Some of these mistakes no tool solves, ours included.

• Surveying a topic instead of a decision. You get interesting answers and no action. Fix: write the Step 1 sentence or don't send. No tool fixes this.
• Treating complainers as the whole customer base. Signal-targeted audiences lean toward vocal, engaged accounts. Fix: always add a no-signal comparison group. Tooling makes matching faster, but it cannot make vocal customers representative. Selection bias in customer surveys is a judgment problem.
• Matching on keywords without reading the evidence. False positives dilute the audience and confuse respondents ("why are you asking me about SSO?"). Fix: have a person review every match.
• Quoting the customer's call back to them. It feels like surveillance. Fix: paraphrase the topic, not the sentence. Never reference something said in an internal Slack thread.
• Leading questions. "How much would a viewer seat help you?" assumes the answer. Krosnick's work on satisficing shows respondents often agree with whatever framing you give them. Fix: offer options, including "neither." A model can draft questions, but a person must check them for bias.
• Ignoring customers who never talk to you. Low-touch accounts generate no call signals. Fix: pair targeted surveys with a periodic broad pulse, and use usage data or churn reasons as signal sources.
• Over-surveying the same champions. Fix: keep a shared "last surveyed" column and a 30 to 60 day cooldown.
• Changing questions mid-flight. If half the respondents saw different wording, the results can't be compared. Fix: freeze questions once sent. If you must change them, start a new version.
• Not closing the loop. The people who complained and answered never hear what changed. Fix: send a short "here's what we did" note within weeks of the decision. This is the "closing the loop" practice from the Net Promoter System, applied to a narrower group.

When You Need Tooling: The Honest Threshold

List-based sends are fine for broad pulse surveys. Targeted ones win when you need a decision.

The manual method stops scaling at certain points. These are rules of thumb, not statistics:

• You have more than roughly 200 calls and tickets a month to search for matches.
• You run more than one or two targeted studies a month.
• Matching requires reading full transcripts rather than notes.
• More than one person builds match lists, and their judgments start to drift.
• You need answers to land next to the original call and ticket evidence automatically, instead of in a pasted spreadsheet column.

Below that threshold, stay manual. A spreadsheet and a free form tool are enough. Past that point, the bottleneck is finding and reviewing evidence, not sending the survey. Choose tools that solve that.

Tools for Signal-Based Survey Targeting

1. BuildBetter: Best for targeting surveys from call, ticket and Slack evidence

BuildBetter AI Surveys automates Steps 2 through 4 of the method. You describe the audience in plain language, such as "accounts that raised pricing on calls in the last 90 days." BuildBetter turns that into editable filters over your signals from calls, tickets and Slack. It shows each matched person with the evidence that matched them, and you approve the list before anyone receives anything. Manual email entry is available for one-off sends.

Delivery works by branded email, public link or verified in-app widget. It is the same survey with the same results across channels. AI follow-up questions ask at most one clarifying question when an answer is thin, and often decide not to ask one. There are no canned prompts. Questions are versioned and immutable once sent, which prevents the mid-flight-edit mistake. Every completed response is extracted into signals with citations to the exact answer. Those signals sit next to the calls and tickets that triggered targeting, which covers Step 7 without the spreadsheet.

Pricing, with worked numbers: 1 credit = $0.001. An invitation costs 4 credits, a completed response 8, an AI follow-up 1, and each extracted signal 4. A completed email response with two follow-ups and three signals is about 26 credits (~$0.026). Here is the cost of the worked pricing example:

• 60 invitations = 240 credits
• 31 responses = 248 credits
• About one follow-up each = ~31 credits
• Three signals each = ~372 credits

That totals roughly 890 credits, or under a dollar for the whole study. Activation shows your maximum credit exposure before sending. Previews and test responses are never billed, and seats are unlimited.

A few extras:

• Typeform and Pendo surveys sync in, and SurveyMonkey, SurveySparrow and Survey Planet integrations exist if you already send elsewhere.
• Survey-triggered workflows cover NPS escalation, adoption follow-up and QBR prep.
• BoB, the BuildBetter assistant, can draft a study (questions, audience, rewards) for you to review.
• The platform is SOC 2 Type II, HIPAA and GDPR compliant.

When the survey isn't enough: you can run AI-moderated video or audio interviews with the 5 to 10 respondents whose answers raise new questions. Synthetic Personas built from your real signals can pressure-test question wording before you send. They are for first reads and do not replace real customers. See how ustwo tested both.

"I've never used a tool that actually connects the dots across all my platforms. BuildBetter does that, and it makes me so happy." — Agnes Abel, Typeform

2. Spreadsheet + lightweight form builder (e.g., Tally or open-source Formbricks)

This combination works well for running the manual method at low volume. It costs little or nothing and is quick to set up. Matching and evidence review stay entirely manual, so this option fits teams below the threshold above.

Honest caveat: your main need may be sending one standardized survey to a very large enterprise list for trend tracking. In that case, a purpose-built enterprise survey distribution platform will fit better than signal-led targeting. For analyzing large volumes of open-ended answers after any send, see our guide to tools to analyze thousands of open-ended survey responses.

FAQ: Targeting Surveys by Customer Feedback

What is signal-based survey targeting?

Signal-based survey targeting means choosing survey recipients by evidence from their own words, such as a call complaint, support ticket or competitor mention, rather than by CRM attributes like plan or region. You then tailor the questions to that evidence.

How many people should a targeted survey go to?

Send to every person whose evidence survives review, typically dozens rather than thousands. Add a comparison group of similar accounts with no signal, about half the size of the signal group. Precision matters more than volume because the goal is a decision, not a benchmark.

Should targeted surveys replace NPS?

No. Keep a broad NPS or CSAT pulse for trend tracking and for hearing from quiet customers. Use targeted surveys, including a targeted NPS follow-up survey to detractors who named a specific issue, when a specific decision depends on what a specific group of customers needs.

How do I avoid making customers feel watched?

Reference the topic, not their exact words. Never cite internal conversations, and send from or with the account owner for key accounts. "You mentioned pricing came up at renewal" is fine; a pasted call quote is not.

How long should the lookback window be?

Sixty to ninety days works for most product and pricing decisions. Shorter windows miss slow-moving complaints. Longer ones include customers whose situation has already changed.

Can I do this without any software?

Yes. The whole method is three things: a spreadsheet of matched customers with the quote that matched them, a free form tool, and a column that codes each answer against the original signal. Software becomes worth it once matching evidence across calls and tickets takes longer than writing the survey.

Sources and Further Reading

• Wikipedia, Selection bias: why signal-only audiences need a comparison group
• Wikipedia, Survey methodology
• Wikipedia, Net promoter score: context for broad pulse vs. targeted surveys (Reichheld, "The One Number You Need to Grow," HBR, 2003)
• Pew Research Center, "Response rates in telephone surveys have resumed their decline" (2019)
• Porter, Whitcomb & Weitzer (2004), "Multiple surveys of students and survey fatigue," New Directions for Institutional Research
• Squire, P. (1988), "Why the 1936 Literary Digest Poll Failed," Public Opinion Quarterly
• Teresa Torres, Continuous Discovery Habits (2021); Rob Fitzpatrick, The Mom Test (2013)
• BuildBetter AI Surveys and BuildBetter docs
• Related reading: Tools to analyze thousands of open-ended survey responses

Make Churn Optional.

Your customers already told you what they need on calls, in tickets and in Slack. BuildBetter finds the people behind those signals and shows you the evidence for each one. It sends them a survey that asks the right question, then puts every answer next to the conversation that started it. You get a decision instead of an average.

Make churn optional. Book a demo.