Product research workflow
Define what you need to learn, build an editable survey, audit the instrument and sample, validate response CSVs, calculate NPS, CSAT, CES, PMF or SUS with the right denominator, and turn the result into a defensible next learning action.
CraftUp is not a hosted survey-distribution platform: it does not email respondents, recruit panels or host response collection. Export the definition or response template into your collection tool, then bring results back here for validation and scoring.
Share survey definition puts the editable survey definition in the URL. Anyone with the link can inspect it; it is not a hosted respondent survey.
Define what you need to learn, which decision this may inform, who should answer, and when. The survey score does not make the Product decision for you.
Editing a standardized metric question or scale marks it as modified. Modified instruments should not be presented as benchmark-comparable standard NPS/SUS/PMF without review.
CraftUp exports the survey definition; it does not host respondent collection or send this end screen to users.
No overall score. Findings stay inspectable by decision, sample, question, response, metric and interpretation risk.
No current finding.
No current finding.
No current finding.
No current finding.
No current finding.
No current finding.
No current finding.
No current finding.
Heuristics are review warnings, not proof that a question is unbiased or that a sample is representative.
The workflow is deliberately broader than “pick a template and get a number.” A score is only interpretable together with the question, scale, sample, timing, denominator and comparison context that produced it.
Step 1
Define the learning goal, Product decision, target respondent, eligibility rule, trigger, collection window and sampling method.
Step 2
Build or load a survey, then edit questions, options, required state, anchors, order and lightweight conditional logic.
Step 3
Review the Survey Decision-Quality Audit for wording, sample, burden, response-option, standard-metric and logic risks.
Step 4
Export the definition and response template or share the editable survey definition; collect responses in your chosen channel.
Step 5
Import validated responses or enter aggregate counts, inspect exclusions and denominators, calculate the metric, then choose the next evidence.
Use NPS for a recommendation/relationship signal, CSAT for satisfaction after a specific interaction, CES for perceived task ease or effort with explicit scale direction, PMF for the Sean Ellis essentiality heuristic in a qualified audience, SUS for a standardized 10-item usability score, Discovery for recent behavior/problem evidence, and Churn for retrospective exit signals that still need triangulation.
NPS = % promoters (9–10) − % detractors (0–6). CSAT defaults to the explicit 5-point top-two-box convention (4–5 / valid N). CES defaults to mean ease on the configured scale and polarity, with top-two ease shown separately. PMF = % Very disappointed among eligible responses, excluding N/A by default and evaluating the 40% heuristic before display rounding. SUS is scored per complete respondent: odd response−1, even 5−response, sum ×2.5; SUS is not a percentage.
The most expensive survey failures often happen before the calculator: an undefined audience, unclear trigger, leading or double-barreled wording, overlapping options, a modified standardized instrument presented as benchmark-comparable, silent response exclusions, or a trend comparison across different populations. The audit surfaces these as review warnings and hard integrity failures where appropriate; it does not claim regex heuristics can certify an unbiased survey.
Keep the question, scale, sample, timing, valid N and comparison context attached to the score. A survey result can motivate a follow-up interview, behavioral analysis, research triangulation, problem-validation pass or controlled experiment. It should not be converted directly into a roadmap verdict or causal claim.
Why: The questionnaire was written before stating the learning goal.
Fix: Write the uncertainty and Product decision first; remove questions that cannot change what you do next.
Why: The result is shown without its denominator.
Fix: Always show n by segment and label tiny slices directional rather than treating them as stable population estimates.
Why: The 0–100 score is being mistaken for a percentage and historical average for a requirement.
Fix: Show SUS score: 72, valid respondent N and benchmark context separately.
Why: Task ease was collapsed into a generic top-two satisfaction formula.
Fix: Store exact CES question, anchors, polarity and calculation; report mean ease and any top-two share separately.
Why: The 40% heuristic is treated as a verdict without qualifying the sample or checking retention.
Fix: Show the eligible denominator, qualification rule and retention/behavior follow-up.
Why: A fixed length threshold ignores matrices, long text and required effort.
Fix: Use a burden heuristic based on response effort and survey purpose, not a universal >12 rule.
Why: Question wording, scale, trigger or sample changed between waves.
Fix: Use the survey fingerprint/version contract and surface comparability warnings before showing a naive delta.
Why: Definition sharing was confused with hosted survey distribution.
Fix: Call it Share survey definition and export/copy the draft into your collection tool for response gathering.
No. CraftUp helps a Product team design, audit, score and interpret a lightweight survey. The share action serializes the editable survey definition; it does not create a respondent-facing hosted form, send email, recruit a panel or manage enterprise respondents. Export the survey definition or response CSV template and use your preferred collection channel or survey platform to gather responses.
NPS uses the 0–10 recommendation question and subtracts the percentage of detractors from promoters. CSAT uses the configured top-two-box satisfaction convention for a specific interaction. CES keeps the task-ease question, anchors, scale direction and calculation explicit; the default ease template reports mean ease and an optional top-two ease share rather than pretending CES is identical to CSAT.
No. A survey score describes responses from the observed sample under a particular question, scale, trigger and collection context. A later wave can change because the population, timing, wording, seasonality or external context changed. Use repeated comparable waves for trend evidence, behavioral analytics for observed behavior, and a controlled experiment when you need a stronger causal claim.
The Sean Ellis heuristic looks at the share of qualified respondents who say they would be very disappointed if they could no longer use the Product. CraftUp shows the eligible denominator and treats 40% as a strong signal for that surveyed audience, not proof that Product-market fit exists. Check who was surveyed, retention, segment behavior and qualitative explanations before making a scaling decision.
Standard SUS has ten 1–5 items. For each complete respondent, odd-numbered items contribute response minus 1 and even-numbered items contribute 5 minus response. Sum the ten contributions and multiply by 2.5. CraftUp then reports the mean across valid complete respondents. The resulting 0–100 SUS score is not a percentage; benchmark interpretation depends on relevant comparison context.
Question wording, response categories and earlier questions can influence later answers. The audit therefore flags obvious risks such as leading language, double-barreled questions, vague recall periods, overlapping numeric options and unclear anchors. These are transparent heuristics, not proof that a question is biased. For high-stakes research, pretest new wording and review the full survey flow with real respondents.
Yes. The scorer accepts CSV mapped through stable question IDs from the CraftUp response template. It handles quoted commas, escaped quotes, CRLF and extra columns, while surfacing invalid scale values, missing metric columns, duplicate response IDs and incomplete standardized responses. The import summary separates detected, valid, partial, invalid and excluded rows instead of silently coercing bad data into a score.
When an imported CSV includes a segment column, CraftUp can show the configured metric by segment together with that segment's valid n. Small slices are marked directional so a striking score is not presented without its denominator. Segment comparison is descriptive unless the sampling and analysis design supports a stronger inference; the tool does not invent representativeness from a high score or high response rate.
The survey definition and aggregate results workspace are saved separately in your browser. Definition sharing puts the survey draft into the URL, so anyone with that link can inspect the questions and research context. Imported respondent rows and open-text responses are not included in analytics and are not placed into the definition share URL. Use exports deliberately when data contains sensitive research or respondent information.
Choose the next evidence based on the result and its weaknesses. You may need to revise the questionnaire, collect a more relevant sample, run follow-up interviews, inspect behavioral analytics, triangulate several methods, validate the underlying problem or design an experiment. A survey is one evidence source: the correct next step is rarely simply “the score is good, so ship the requested Product change.”
If a score moved after a release, the survey alone cannot establish that the release caused the change. Define the causal hypothesis, metric and guardrails before choosing an experiment.
Build an A/B test planLast updated: 2026-09-18