Blog · How-to guides · Sales teams
Marketing research analytics compares structured survey responses with explicit bases and limitations. Work from a research workbook to checked counts, segment percentages and multiple-response tables without turning association into a causal claim.
Marketing research analytics turns structured survey responses into checked counts and segment comparisons. The central task is to define who belongs in each question's base, then show how the answers differ without overstating what the sample proves. Start with the research workbook and the exact wording and coding of each question.
Suppose a US business wants to compare satisfaction between existing and prospective customers who answered its research survey. This is a descriptive study of the respondents. The workbook does not establish that either segment represents every customer or prospect in the market.
Keep one row per respondent, an anonymous unique respondent identifier, a segment, question eligibility and the recorded answer. Preserve a separate copy of the original export for review. The analysis table should distinguish an unanswered question from an answer that explicitly means none, neutral or not applicable.
Use the wording and response codes supplied with the study. Pew Research Center's questionnaire guidance explains that response options and their order can affect answers. Comparisons across survey waves need that context, even when the column headings look identical.
This guide covers numeric and coded response tables. Commercial funnel outcomes are a different analysis; see B2B marketing analytics when the question is how campaign accounts turn into revenue.
All respondents in this example were eligible for a satisfaction question. Allowed answers are integer ratings from 1 through 5; a top-two-box result means a rating of 4 or 5. The source workbook contains the following reviewed counts.
| Segment | Eligible respondents | Valid ratings | Ratings of 4 or 5 | Other valid ratings | Missing ratings |
|---|---|---|---|---|---|
| Existing customers | 60 | 54 | 36 | 18 | 6 |
| Prospective customers | 40 | 32 | 16 | 16 | 8 |
| Total | 100 | 86 | 52 | 34 | 14 |
The first reconciliation is 86 valid ratings + 14 missing ratings = 100 eligible respondents. Within the valid base, 52 top-two-box ratings + 34 other valid ratings = 86. Both identities must hold before calculating percentages.
The eligible base is not automatically the survey invitation count. These records cover respondents included in the workbook. Without invitation and disposition records, they do not establish the study's overall response rate.
Put respondent ID in column A, segment in B, eligibility flag in C and satisfaction rating in D. Rows 2 through 101 contain the reviewed respondents. Before counting, validate ratings as allowed integers, remove duplicate IDs according to the study's documented rule, and keep an exception record for invalid codes.
For existing customers, the eligible base is:
=COUNTIFS(B2:B101,"Existing",C2:C101,1)
The valid base and top-two-box count are:
=COUNTIFS(B2:B101,"Existing",C2:C101,1,D2:D101,">=1",D2:D101,"<=5")
=COUNTIFS(B2:B101,"Existing",C2:C101,1,D2:D101,">=4",D2:D101,"<=5")
Microsoft's COUNTIFS documentation confirms that all specified criteria must be met for a row to count. Replacing the segment criterion with "Prospective" produces the other segment. The integer validation matters because the range criteria alone would also accept an unintended fractional rating.
If the top-two-box count is in J2 and the valid base in I2, calculate the rate as:
=IF(I2=0,"Not available",J2/I2)
Top-two-box percentage = Respondents rating 4 or 5 / Respondents with a valid rating
| Segment | Top-two-box count | Valid base | Calculation | Top-two-box percentage |
|---|---|---|---|---|
| Existing customers | 36 | 54 | 36 / 54 | 66.7% |
| Prospective customers | 16 | 32 | 16 / 32 | 50.0% |
| Total | 52 | 86 | 52 / 86 | 60.5% |
Percentages are rounded to one decimal after calculation. The combined result comes from combined counts, rather than averaging segment percentages. Using all eligible respondents would instead give 52 / 100 = 52.0%; that is a different, explicitly labeled measure, not a correction to the valid-response percentage.
Now consider a feature-interest question in the same workbook. Respondents could select either or both of the listed features. A question-level answered flag identifies 80 valid respondents; the remaining records must not silently become people who selected nothing.
| Feature | Respondents selecting feature | Valid respondent base | Calculation | Percentage of respondents |
|---|---|---|---|---|
| Faster reporting | 50 | 80 | 50 / 80 | 62.5% |
| Easier account comparison | 36 | 80 | 36 / 80 | 45.0% |
| Selections across options | 86 | 80 | 86 / 80 | 107.5% |
The total is 86 selections across 80 respondents. The percentage sum is allowed to exceed 100% because a person can appear in both option counts. It does not mean the workbook contains extra respondents.
For this question, use a reviewed answered flag in column G and a selected flag in column E for faster reporting. The following numerator and denominator preserve the respondent base:
=COUNTIFS(G2:G101,1,E2:E101,1)
=COUNTIFS(G2:G101,1)
State whether an explicit none option counts as an answered question. A blank cell alone cannot tell you whether the respondent skipped the question or selected no options.
Existing respondents have a higher recorded top-two-box percentage in this workbook. That result can guide a follow-up question or a sales conversation. It does not establish that becoming a customer increased satisfaction, or that the same difference exists in the entire market.
Pew Research Center's evaluation of online nonprobability surveys found that demographic resemblance alone did not predict accuracy. A well-organized workbook cannot repair selection bias by itself. Document how participants entered the study and ask the research owner whether population estimates, weights or uncertainty calculations are justified.
If weights are supplied, keep weighted percentages and unweighted respondent counts separately labeled. Do not introduce weights merely to make a segment distribution look more convenient. A research owner should approve the intended population and weighting method.
Use marketing KPIs when selecting a recurring commercial scorecard, and keep research findings identified as research. For a reviewable analysis, the principle in citations on every number applies: another person should be able to locate the rows and rebuild the result.
Covirage is not affiliated with Microsoft.
Bring a structured research workbook with respondent identifiers, segment codes, eligibility and numeric responses to Covirage, which maps the columns while its tools compute and check the figures. The external AI model explains the comparisons with every figure traced to its rows. Create a free Covirage account and bring a sample of your own data; creating the account takes about a minute.
See Covirage for sales teams for the supporting business context.
Marketing analytics often tracks recorded activity, spending and commercial outcomes. Marketing research analytics examines research observations, such as structured survey responses, to answer a defined question about customers or markets. The two can inform the same decision while using different records and populations.
Use the population described by the result. A percentage of valid answers uses respondents who supplied an allowed answer to that question. A percentage of all eligible respondents includes missing answers in its base. Label the choice and show the underlying counts.
In a multiple-response question, one respondent can select several options. Percentages calculated on respondents may therefore sum above the whole. Show the valid respondent base and label the question as multiple response. Do not confuse respondents with the number of selections.
A descriptive cross-tab shows how recorded responses differ across groups. It does not, by itself, isolate a cause. Differences may reflect who answered, how the question was asked or other group characteristics. Stronger causal claims require an appropriate study design and review.