Blog · Board and management reporting · Customer service
Customer sentiment summarizes the attitudes customers express in feedback. Learn how to compare supplied labels, reconcile the counts, and interpret changes without confusing association with causation.
Customer sentiment summarizes the attitudes customers express in feedback, measured here from labels your team has already supplied. Compare positive, neutral, and negative responses on one consistent scale, and show how much received feedback has a usable label. A score without its population and coverage can hide a change in who answered.
The business question is practical: did a service team's customers express a different attitude this month, and where should the team investigate? The answer starts with a de-identified feedback export, rather than an assumption that silence means satisfaction.
Choose the unit before calculating anything. A transaction-level survey gives frequent callers more opportunities to appear. A customer-level monthly measure gives each responding customer one place in that month's result. Both can be useful, but they answer different questions.
For the example below, the service operations team supplies one retained response per customer and service team per month. If several responses exist, the team keeps the latest completed response under a documented rule. Customer identifiers are de-identified, and the export contains supplied labels rather than raw comments.
Keep these fields: response ID, de-identified customer key, response date, service team, supplied label, and scoring-rule version. If you compare customer groups, also retain the supplied group and its effective date. A separate received-feedback count preserves responses whose labels are missing.
Keep the question and collection method stable. Pew Research Center's guidance on writing survey questions explains that wording, question order, and survey mode can affect responses and therefore comparisons over time. A new question can create a new series even when the spreadsheet column keeps its old name.
Positive share = Positive responses / Scored responses
Net sentiment points = (Positive responses − Negative responses) / Scored responses × 100
Scored coverage = Scored responses / Received responses
Scored responses equal positive plus neutral plus negative responses. Neutral responses remain in the denominator. Unscored responses do not enter either sentiment calculation, but they remain visible in coverage.
Net sentiment points are a locally defined summary here, not a universal industry standard. A zero result means equal positive and negative counts; it does not mean every respondent was neutral. If your supplier uses a different scale, document that scale and its mapping before comparing it with these measures.
Keep a satisfaction score or recommendation score separate unless its original question and scale support the same interpretation. For the broader files-first approach, see AI analytics for customer service.
A US distributor compares July and August 2026 feedback. These are illustrative counts, not a benchmark. Every percentage below is calculated from the supplied labels; percentages and point changes are rounded to one decimal.
| Period and team | Received | Positive | Neutral | Negative | Scored | Unscored |
|---|---|---|---|---|---|---|
| July, billing | 100 | 60 | 25 | 15 | 100 | 0 |
| July, delivery | 50 | 20 | 15 | 15 | 50 | 0 |
| July total | 150 | 80 | 40 | 30 | 150 | 0 |
| August, billing | 90 | 52 | 16 | 12 | 80 | 10 |
| August, delivery | 130 | 60 | 36 | 24 | 120 | 10 |
| August total | 220 | 112 | 52 | 36 | 200 | 20 |
The first check is an identity: August's 112 + 52 + 36 = 200 scored responses, and 200 + 20 = 220 received responses. Each team must pass the same check before its results enter the total.
| Period and team | Positive share | Net sentiment points | Scored coverage |
|---|---|---|---|
| July, billing | 60 / 100 = 60.0% | (60 − 15) / 100 × 100 = 45.0 | 100.0% |
| July, delivery | 20 / 50 = 40.0% | (20 − 15) / 50 × 100 = 10.0 | 100.0% |
| July total | 80 / 150 = 53.3% | (80 − 30) / 150 × 100 = 33.3 | 100.0% |
| August, billing | 52 / 80 = 65.0% | (52 − 12) / 80 × 100 = 50.0 | 80 / 90 = 88.9% |
| August, delivery | 60 / 120 = 50.0% | (60 − 24) / 120 × 100 = 30.0 | 120 / 130 = 92.3% |
| August total | 112 / 200 = 56.0% | (112 − 36) / 200 × 100 = 38.0 | 200 / 220 = 90.9% |
Positive share rises by 2.7 percentage points overall, and net sentiment rises by 4.7 points, calculated before rounding. Both teams improve, but the company result also reflects a larger share of delivery responses, the team with the lower positive share.
For a useful diagnostic, apply July's scored-response weights to August's team results. Billing had 100 of July's 150 scored responses, and delivery had 50. Holding those weights constant gives:
August positive share at July weights = (100 / 150 × 65.0%) + (50 / 150 × 50.0%) = 60.0%
That is 6.7 percentage points above July's 53.3%, calculated from unrounded values. It answers how August would look with July's team mix. The actual August result remains 56.0%; the adjusted figure is a separate diagnostic, not a replacement.
Coverage also fell. Ask why August has unscored responses before treating the improvement as complete evidence. Scored coverage is not survey response rate: calculating response rate also requires an appropriate invitation or eligible-customer denominator. Pew's research on low survey response rates shows why response rate alone is an unreliable indicator of bias. Inspect who responded, rather than declaring a result representative from one percentage.
The service operations owner should investigate delivery feedback first because its positive share remains below billing's. The owner of the feedback process should investigate missing labels separately. Neither action requires assuming that the score caused a commercial outcome.
If you join sentiment with revenue or renewals, use a de-identified customer key and feedback recorded before the outcome window. Count the customers matched and unmatched, and compare similar tenure or service groups. The net revenue retention worked example owns the revenue-retention calculation; sentiment supplies an additional grouping, not an alternative definition.
Likewise, a sales comparison should preserve the denominator in the quote conversion worked example. Different feedback scores among converted and unconverted accounts identify a question to investigate, not proof of an effect.
Bring a de-identified feedback file with supplied sentiment labels, dates, team names, and customer groups; add a compatible outcome export if needed. Covirage maps the columns, and its tools compute and check the group counts, shares, and supporting summaries; the external AI model explains the results with every figure traced to its rows. Create a free Covirage account and bring a sample of your own data.
See Covirage for customer service teams for the supporting use case.
No. Satisfaction usually answers a specific question about an experience or relationship. Sentiment is a broader description of expressed attitude. You can summarize either from supplied scores, but do not combine them unless the question, response scale, population, and scoring rules are comparable.
It can be a useful signal to investigate, but a difference in renewal rates does not establish that sentiment caused the difference. Compare feedback recorded before renewal, use comparable customer groups, and account for customers who did not respond before drawing conclusions.
Keep them in a separate unscored count. Exclude them from positive and negative share denominators, disclose the scored share of received feedback, and investigate changes in missingness. Assigning neutral labels to missing values makes the result look more complete without adding evidence.