Begin with the management question
Before opening a charting tool, name the decision. Do you need to prioritise a service problem, compare experiences across branches or understand what to investigate next? Different decisions need different questions, collection methods and levels of certainty.
AAPOR’s survey guidance emphasises matching the method to the research question, careful sampling and clear question design. It also distinguishes probability and nonprobability samples. Those distinctions matter when deciding how far a finding can be generalised. Read the primary guidance.
Existing survey data can still be useful when the original method was imperfect. The task is to describe what it can show, disclose what it cannot and avoid retrofitting certainty onto a dataset that was not designed to provide it.
Ask “out of whom?” before discussing a percentage
Consider an entirely synthetic example. A business invites 200 customers to give feedback. Fifty respond and 35 of those respondents select a positive answer. That is 70% of respondents, not evidence that 70% of all customers are satisfied. The nonrespondents may have different experiences.
The denominator can change from one question to the next. If ten people skip a particular item, that item has a different response base. Show the number of answers alongside each percentage and explain how missing and “not applicable” answers were treated.
A management report should also distinguish invitations from eligible people, completed responses and partial responses. If those counts are unavailable, say so. A large spreadsheet does not, by itself, resolve coverage or nonresponse concerns.
Check the structure before interpreting the pattern
Keep an original read-only copy of the data. Create a cleaning log that records duplicate-handling rules, inconsistent labels, impossible values and any exclusions. Agree the rules before looking for a preferred result.
Inspect the questionnaire as well as the data. A question that asks whether support is “fast and helpful” combines two concepts. A respondent may agree with one and disagree with the other. Similar-looking questions collected with different wording may not be directly comparable.
For branch comparisons, check that collection dates, question wording and respondent groups are sufficiently alike. A difference might reflect the composition of the respondents or the timing of the survey. Report those alternative explanations rather than treating the chart as a league table.
A pattern is a starting point for investigation
If customers with longer waiting times also report lower satisfaction, the pattern may be relevant. A descriptive analysis alone does not establish that changing waiting times will produce a particular satisfaction increase. Other factors may influence both measures.
Our suggested reporting structure separates observation, possible explanation and follow-up. For example: “Respondents in branch A reported longer waits”; “staffing or collection timing may contribute”; “check transaction records and observe the service process before setting a target.” The latter two statements are interpretations and proposed actions.
Advanced inference, causal evaluation and representative polling need a suitable study design and appropriate methodological expertise. A dashboard cannot supply those features after collection has finished.
Protect people while making findings useful
Employee free-text responses can identify individuals even after names are removed. Small groups and combinations of fields can also reveal identity. Agree aggregation, access and reporting rules before sharing results; there is no universal minimum group size that makes every dataset safe.
Kenya’s data-protection office describes principles including purpose limitation, data minimisation, accuracy and limited retention. These are relevant to planning a survey workflow; organisation-specific obligations need appropriate review. See ODPC’s principles and rights guidance.
For an initial discussion, provide a field dictionary and synthetic or properly de-identified examples. Agree a secure transfer method and data-handling terms before sending confidential records. Removing a name column alone is not a complete privacy assessment.
End with a report people can act on responsibly
A useful management pack includes the question, collection method, dates, response counts, cleaning decisions, charts and limitations. It should explain which findings warrant action now and which call for more information.
Assign each proposed action an owner and a review measure. If you repeat a survey, preserve comparable wording and collection methods or explain the change. Review action completion alongside the next result rather than treating another survey as an action in itself.
The best outcome is not the most colourful chart. It is a team that understands the evidence well enough to take a proportionate step and recognise when a more rigorous study is needed.
Sources & editorial note
- AAPOR — Best Practices for Survey Research. Primary methodological guidance on research questions, sampling and questionnaire design.
- ODPC — Rights of a Data Subject. Kenyan data-protection principles and rights; not a substitute for advice on a particular dataset.
Sources reviewed 25 October 2026. Frameworks, examples and suggested next steps are Sage Advisory editorial analysis. All business scenarios and numerical examples are illustrative. No client outcomes are represented. Published by the Sage Advisory editorial desk.