How to Create a CSAT Report That Drives Action
Two teams survey the same customers on the same day. One reports 82%, the other reports 4.1.
Both are correct. One used the percentage method and the other averaged the scores, and the source that defines the metric allows either.
That is the first thing to settle before you build the report. The formula is a choice, not a standard.
This guide covers what the report must contain and the two legitimate formulas. It then covers why the scale and the timing decide what you learn, and how to produce the report from a survey export.
What a CSAT report has to contain
The score is one line. The report is what makes the line usable.
Six things belong on the same view. The score, the formula used, the scale, the response count, the population surveyed, and the trigger point.
The trigger is the one most reports omit. A score collected after a support call and a score collected after 30 days of ownership are different measurements with the same name.
Add the distribution too. Keep the count at each point on the scale, because collapsing to a single figure throws away the shape permanently.
Response count carries more weight than people expect. With 40 responses, a handful of people moving one point swings the score by several points.
Two legitimate formulas, two different numbers
Qualtrics defines the metric and its calculation on its CSAT page, and it names both approaches.
The percentage method is the one it walks through. You "use the responses of 4 (satisfied) and 5 (very satisfied)," then apply the formula it publishes.
The formula reads as follows. Number of satisfied customers, divided by number of survey responses, multiplied by 100.
There is a stated reason for the top-two rule. The page notes that "using the two highest values on feedback surveys is the most accurate predictor of customer retention."
The averaging method is acknowledged on the same page. Results "can be averaged out to give a Composite Customer Satisfaction Score," although scores are "more usually expressed as a percentage scale."
| Percentage method | Average method | |
|---|---|---|
| Output | 0% to 100% | A number on your scale |
| Treats a 3 as | Not satisfied | Middling |
| Sensitive to | Movement across the 3-to-4 line | Movement anywhere |
| Comparable to published benchmarks | Usually yes | Rarely |
Pick one and write it on the report. The two are not convertible, and a team that switches mid-year has destroyed its own trend.
A related metric follows a different rule. NPS excludes its middle band from both percentages, while CSAT keeps everyone in the denominator. Our guide to creating an NPS report covers that comparison.
Where the scale decision bites
Qualtrics presents CSAT on a five-point scale, running from very unsatisfied to very satisfied.
The standard question wording is published too. It asks how the customer would "rate your overall satisfaction with the [goods/service] you received."
Plenty of tools ship a seven-point or ten-point default instead. That is not wrong, and it does change the arithmetic.
The top-two rule does not survive a scale change intact. Two boxes out of five is the top 40% of the scale, while two boxes out of ten is the top 20%.
So a ten-point survey scored with the top two boxes will report a lower CSAT than a five-point one, from identical sentiment. That is a measurement artefact, not a customer signal.
Two rules follow, and both are cheap to apply. Freeze the scale, and record it beside every score you publish.
If you must change the scale, treat it as a new series. Note the change date and stop comparing across it.
When you ask changes what you learn
Timing is a design decision that the same source treats as central.
Qualtrics describes CSAT as a "right here, right now" metric that "relates to a specific experience rather than an ongoing customer relationship."
That shapes when the question makes sense. For discrete interactions like a call to a contact centre, the page recommends asking immediately, while the conversation is fresh.
For something used over time, the advice reverses. A subscription or a durable item may need days or weeks before the customer can judge it.
Two practical consequences follow. A single blended CSAT across both kinds of touchpoint is close to meaningless, and the fix is to report by trigger.
The page also warns about frequency. Asking a repeat customer too often "can create its own problems," which shows up in your data as a falling response rate.
Watch that rate as a metric in its own right. A score that rises while responses fall is usually survivorship, not improvement.
How to do it manually
Option 1: Two counts and one division
Count responses at 4 and 5 with COUNTIFS, then count all responses with COUNTA on the rating column.
Using COUNTA on the rating column rather than the row count matters. Blank ratings are not zeroes, and letting them into the denominator drags the score down.
Divide once at the end and keep both counts visible. Anyone should be able to recompute your number from the two cells above it.
The ceiling is that you get one score for one period, with no view of which touchpoint or segment moved.
Option 2: The full distribution
Keep a count at every point on the scale, then derive both the percentage and the average from the same table.
Showing both is useful once, as a sanity check. If the two tell opposite stories, you have a bimodal distribution and a single figure will mislead whichever way you go.
Now the report can be sliced. By trigger, by product, by channel, by agent, or by tenure.
The finding usually lives in the slice. A healthy blended score often hides one channel sitting a full point below the rest.
The limit is joins and volume. Matching survey responses to account or ticket data is past the point where formulas stay comfortable.
Option 3: A definitions tab
Record the formula, the scale, the question wording, the trigger, and the population.
Question wording belongs on that list for a reason. A reworded question produces a different score, and it will be read as a change in sentiment.
Record the exclusions too. Internal testers, duplicate submissions, and responses outside the valid range each need a stated treatment.
The limitation is familiar. Writing a rule down does not apply it, and someone rebuilds the same filters next cycle.
The shared ceiling. All three assume the export carries a trigger or survey identifier. Without one, a blended score cannot be split back into the experiences that produced it.
Where the manual route slows down
The first report takes an hour. The fourth takes longer, because three things moved.
Survey tooling changes the scale during a redesign, and nobody connects the score shift to the form.
The population widens quietly. A survey sent to all contacts rather than recent purchasers brings in people with nothing recent to judge.
Then benchmarks get quoted. Qualtrics notes that benchmarking CSAT "isn't an exact science," pointing to the ACSI for industry figures while stressing that every business and product differs.
Its more useful line is simpler. If your score is moving from lower to higher, you are doing something right.
There is a fourth cost that appears under deadline. Somebody asks whether 82% is good, and answering needs the trend, the trigger split, and the response count you did not bring.
The tooling side has its own roundup, in AI software for survey data analysis.
How to build it with Powerdrill Bloom
Step 1: Upload your survey export
Upload the response export straight from your survey tool. Powerdrill Bloom profiles the columns on arrival, so blank ratings, out-of-range values, and duplicate respondent identifiers surface before any score is calculated.
Step 2: Describe the report in natural language
State the rules rather than building them. Name the rating column, the scale, the formula, the period, the population, and the trigger field.
Then ask the questions that catch the errors. Ask how many responses are blank or outside the scale. Ask for both the percentage and the average from the same data. Then ask for the score split by trigger and by segment.
Step 3: Export the chart, report, or deck
Take out the score with its full distribution, a trend across periods, or slides that carry the formula and response count beside the number.
Common mistakes
Not stating the formula. The percentage and the average are both legitimate and not comparable. Label which one you used.
Changing the scale mid-series. A five-point and a ten-point survey produce different scores from identical sentiment. Treat a change as a new series.
Counting blanks as zeroes. A non-response is not dissatisfaction. Exclude blanks from both the numerator and the denominator.
Blending trigger points. Post-call and post-30-day scores measure different things. Report them separately.
Reporting the score without the response count. Small samples swing hard. Put the count next to the number every time.
Reading a rising score with falling responses as improvement. That pattern is usually survivorship. Track response rate alongside.
Quoting another industry's benchmark as a target. Benchmarking is not an exact science here. Use your own trend as the reference.
Conclusion
Choose the formula, freeze the scale and the wording, report by trigger, and publish the score with its response count and distribution.
The score is not the deliverable. The trigger or segment sitting below the rest is what tells you where to spend next month.
If assembling that view each cycle eats your week, try Powerdrill Bloom on your survey export. See also our guide to turning messy survey data into clear charts, plus the voice of customer summarizer and CSV AI assistant pages.
Frequently asked questions
How is CSAT calculated?
The common method divides the count of 4 and 5 responses by total responses, then multiplies by 100. Averaging the raw scores is also accepted, and gives a different number.
Why use only the top two responses?
Qualtrics notes that using the two highest values is the most accurate predictor of customer retention. It also makes the result readable as a percentage.
What scale should a CSAT survey use?
A five-point scale from very unsatisfied to very satisfied is the published standard. Longer scales work, provided you keep the scale fixed and record it.
When should I send the survey?
Immediately after discrete interactions, and after days or weeks for products used over time. CSAT measures a specific experience rather than the overall relationship.
Is CSAT the same as NPS?
No. CSAT keeps every respondent in the denominator, while NPS subtracts detractor percentage from promoter percentage and excludes its middle band from both.