Start with the objectives and the questions you asked
Before you open the data, restate the objectives from the survey plan. Then map each question to the objective it served. Questions that do not connect to a decision can be reported briefly or left out. If the study was never planned in writing, a short planning step first will make the analysis easier. For that, see planning customer satisfaction research.
Clean the data before analysis
Good analysis starts with a clean file.
- Keep the raw export unchanged and work on a copy.
- Remove duplicate responses and obvious test entries.
- Check for values outside the scale, such as a rating of 6 on a 1 to 5 question.
- Decide how to handle partial responses, and apply the rule the same way to every question.
- Record how many responses were removed and why. The report should state this.
Summarize rating questions
Report the full distribution of answers, not just the average. The Likert scale analysis guide explains which summaries suit rating scales. Two groups can share an average of 3.5 even though one group is mostly 3s and the other is split between 1s and 5s. The distribution shows which pattern you have.
For common measures:
- CSAT. Report the percentage of responses in the satisfied range, and state which answers count as satisfied. A 5-point scale that counts 4 and 5 as satisfied is typical, but state your choice.
- NPS. Report the share of promoters, passives, and detractors, then the net score. See how to calculate net promoter score.
- CES. State the scale and its direction, since wording varies. See customer effort score.
Always show the number of responses (n) behind each figure. A percentage based on 12 responses carries very different weight from one based on 1,200.
Compare groups carefully
Comparisons by product line, sales channel, customer tenure, or region often drive the most useful findings. Keep these points in mind:
- Small groups are unstable. A few extreme answers can move a score a lot. Report the count, and avoid firm conclusions from very small groups.
- Keep definitions fixed. If the question wording or scale changed between waves, a trend may reflect the change rather than customers.
- Check whether a difference is likely to be real. For formal comparisons, a significance test helps. A t-test can compare average ratings between two groups, and a chi-square test can compare proportions such as the share of satisfied customers. Ask someone with survey statistics experience to confirm the approach if the result is important.
- Avoid averaging averages. When groups differ in size, an overall average should weight each response, not each group equally, unless you explain the choice.
Code the open-ended comments
Open-ended answers explain the scores, but they need structure before they can be counted.
- Read a sample first. Read 50 to 100 comments, or as many as you can, and note recurring topics such as delivery delays, price, or staff helpfulness.
- Build a codebook. Write a short definition for each theme and an example comment. Keep the list manageable.
- Allow more than one code. A single comment can mention both price and support.
- Check agreement. Have two people code the same subset, compare results, and discuss the differences. Revise the codebook where the definitions were unclear.
- Count and quote. Report how often each theme appears and include one or two short quotes per theme. Quotes give the numbers a voice, but they should not replace the counts.
Automated sentiment and topic tools can handle large volumes. Check their output against a manual sample, because they can misread sarcasm, mixed comments, and short replies.
Look for patterns and likely causes
The most useful question is whether a theme connects to the scores. For example, do customers who mention delivery delays give lower ratings than those who do not? If so, the delay is a likely driver of dissatisfaction, though the survey alone does not prove it.
When you look at change over time, check whether a known event, such as a price change or a new release, lines up with the shift. Timing can suggest a cause, but it does not confirm one. State the reasoning and what else could explain the pattern.
Structure the report
A satisfaction research report usually works best with this order:
- Summary. Three to five key findings, each with its most important number. See how to write a market research executive summary.
- Objectives and method. What was asked, who was surveyed, when, how, and the number of responses and response rate.
- Sample. How respondents compare with the customer base, and any weighting used.
- Findings. One section per objective, with a chart or table, the n, and one or two quotes.
- Limitations. Nonresponse, small segments, self-reported answers, and the time window.
- Recommendations. Specific actions tied to findings, with an owner and a way to measure whether they worked.
- Appendix. The questionnaire, the codebook, and full tables.
Write the summary last, after the findings are settled. Keep recommendations separate from findings so readers can see which claims come from the data and which come from judgment.
Visuals and common reporting mistakes
Use one chart per key point, label the axes, and include n. Simple bar charts and tables usually communicate better than dense multi-series charts.
Common mistakes to avoid:
- Reporting only averages and hiding the distribution.
- Treating the share of comments about a theme as the share of all customers.
- Claiming a cause from a before-and-after comparison alone.
- Leaving out the limitations section.
- Including so many charts that the main findings are lost.
Next steps
Once the report is written, use the findings to decide which metrics to keep tracking. For the core metrics, see CSAT and KPI and customer satisfaction surveys. Decide how you will code the open-ended questions before fieldwork starts, so the codebook and analysis plan are ready when the responses arrive.