Before using survey results to change your next workshop, write down how many people answered each question. A submitted form can contain skipped questions. One person can select several topics. Someone who never reached a question has not told you “none.” Those differences change the report you can defend.
This is a Part 2 workflow recipe in the Overskill build journey. You will make a small feedback report, then use it to describe the survey and reporting behavior your app needs. Start with the downloadable example or your existing survey export. If your current tools already produce a clear report, you can keep them.
Freeze the questions and the reporting window
The fictional example contains twelve invitation records, I01 through I12. Each invitation ID identifies one distinct fictional participant here. Real repeat invitations or responses need a declared respondent-identity rule before reporting unique people. Its reporting cutoff is November 9, 2026, at 17:00 UTC. Eight records include a final Submit event, one contains partial progress and three have no captured response. We do not know whether every invitation arrived or every recipient was eligible. Label these as invitation and submission counts, not a response-rate claim.
The first question asks whether the person attended. Seven answered yes, two answered no and three have no answer. Only attendees receive the usefulness question, rated from 1, “Not useful,” to 5, “Extremely useful.” The final question offers guided practice, more worked examples and a session recording. It allows several choices or an explicit “none.” Non-attendees can answer that final question too.
Keep the question text, choices, branch rule and version beside the data. “Would you like a recording?” and “Did you use the recording?” cannot share a result simply because both mention a recording. A changed question needs a new version and a deliberate comparison.
Download the fictional CSV, download its question schema, and download the completed and blank report worksheet. These are illustrative files, not an Overskill export format.
Give each missing answer its actual meaning
For the usefulness question, the twelve records divide as follows:
Scroll horizontally to compare all columns.
| Recorded evidence | People | What belongs in the report |
|---|---|---|
| Answered | 5 | Five ratings contribute to the result |
| Explicitly skipped | 1 | The question was shown and bypassed |
| Not reached | 1 | Captured progress stopped before the question |
| Not asked | 2 | The attendance branch excluded the question |
| Unknown | 3 | No response establishes exposure or an answer |
The example supplies those states explicitly. A blank cell alone cannot tell you which one occurred. Even a label such as “skipped” may cover more than one situation: SurveyMonkey documents that skip logic can bypass required questions and that such questions can appear as skipped in an individual response. Check what your own source records before translating its labels. SurveyMonkey’s explanation.
When the evidence is missing, keep “unknown.” Do not change it to zero, a deliberate skip or a negative opinion. If a record says someone did not attend but also contains an attendee-only rating, investigate that contradiction before completing the report. Preserve the source row so the correction has a reason.
Use a denominator for each question
The five usefulness ratings are 5, 4, 4, 3 and 2. Their sum is 18, so their mean is 3.6. The accurate label is “Mean usefulness among five answering attendees: 3.6 out of 5.” Dividing by seven attendees or twelve invitations would turn missing answers into a lower rating that nobody supplied.
Keep the distribution nearby: one person selected 5, two selected 4, one selected 3 and one selected 2. A mean alone hides that spread. The worksheet lists the contributing IDs, making the calculation inspectable without needing names or contact details.
Use the same inclusion rule for partial and submitted forms: include a supported captured answer. I07 answered the attendance question before stopping, so that answer still belongs in the attendance count. This example has no partial-session rating, but final Submit is not a substitute for evidence that a particular answer exists.
The topic question has seven answering people. Six selected positive choices and I04 explicitly chose none. Guided practice, worked examples and a recording each have three selectors. Each result is therefore 3 of 7 answering people, about 42.9%. The nine positive selections come from overlapping choices, not nine separate people. Checkbox percentages can exceed 100% when added together, as SurveyMonkey’s reporting guidance explains.
Count a repeated person–option pair once. Keep an explicit-none answer in the denominator, but reject a record that combines none with a positive selection. Neither a duplicate choice nor contradictory data should quietly improve a topic’s result.
Make a decision the answers support
The three topics tie. This report does not identify a winner, and it does not establish what all twelve invitees want. A useful next action is to offer a choice of topics rather than announce a preference the data has not shown. If comments were separately collected, review them for follow-up questions; this example contains no comments.
Write that decision beside the result. Include the question version, cutoff, inclusion rule and unanswered counts so a colleague can understand it later. The blank worksheet leaves space for the evidence still needed and the person responsible for checking it. For pre-build customer conversations, use the separate customer idea-validation guide.
Turn the report into an Overskill brief
Open the brief builder. Describe a workshop-feedback app that saves question versions, attendance answers, explicit progress and answer states, and unique topic choices. Ask for a staff report that shows each denominator and its contributing records. Keep individual answers private to authorized staff; publish only an aggregate the operator has explicitly chosen to share. These are requirements to build and test.
Edit the brief’s input fields, review the resulting prompt, then choose Start build. Copy brief is available as a secondary action. The link imports none of the downloaded files; supply the fictional setup separately. Normal account access and credit requirements apply when building.
Before using real responses, reproduce the example: twelve records, five ratings totaling 18, seven topic respondents and nine positive pairs. Test a blank, a repeated choice, a contradictory none answer and a non-attendee rating. Then check that a signed-out visitor cannot see individual answers. A useful report should make both its answer and its missing evidence clear.