Sentiment analysis describes the positive, negative or mixed tone expressed in a response. To understand what needs attention, analyze the topic and supporting evidence separately. A negative comment can identify a difficulty to investigate; it does not by itself establish what caused a later cancellation, absence or other outcome.
This lesson helps you build a feedback review that distinguishes tone, issue, severity and next action. Use it when a satisfaction trend or a collection of comments leaves your team asking what actually happened.
Separate four questions
- Tone: What positive, negative or mixed view did the respondent express?
- Topic: What experience, event or service feature were they describing?
- Consequence: What effect did they report, and what remains unknown?
- Response: What does the responsible team need to check or do next?
Do not use one sentiment label to answer all four. A politely worded account can describe a serious problem. An angry comment can concern a resolved issue. Positive and negative experiences can occur in the same response.
Use the words in their context
Consider a fictional customer check-in: “The adviser was excellent, but the device still does not work in the room where I need it.” Calling this response positive loses the unresolved difficulty. Calling it negative loses the praise and may misdirect a review toward the adviser.
A better record keeps the original comment, notes the positive view of the adviser, records the reported device problem and leaves the operational status to be checked. The team may need a support record or a follow-up conversation to learn whether the issue remains unresolved.
Keep the survey question, date, language and relevant event with the response. An answer to “What went well?” should not be interpreted as a balanced assessment of the whole experience. Preserve the original when using a translation, especially where negation or mixed sentiment could change the interpretation.
Define labels with examples and exceptions
| Field | Example value | Interpretation rule |
|---|---|---|
| Expressed sentiment | Mixed | The response contains both praise and criticism |
| Topic | Device usability | The text describes a difficulty using the device in a stated setting |
| Reported consequence | Cannot use the intended room | Retain this as the customer’s report, not an independently verified finding |
| Review status | Needs service review | An authorized reviewer checks current status and decides the response |
| Outcome | Unknown at this check-in | Do not label the customer as a future cancellation |
Allow “unclear” where the text does not support a confident interpretation. If your team needs a severity category, define it using the service context and responsible policy. Do not derive severity from emotional wording alone.
Keep individual history where appropriate
A sequence of check-ins can reveal whether an issue was first raised, repeated, addressed or described differently later. Where identified collection is appropriate, connect each observation to the continuing record and keep its date. Do not replace several dated responses with a single current sentiment label.
For employee or other anonymous feedback, do not create an individual history that conflicts with the collection promise. Analyze at the permitted group level and apply the relevant disclosure protections. The analytical question has to fit the data you are entitled to use.
In Sopact, this continuing context can help a reviewer read a comment alongside appropriate prior evidence rather than treating it as an isolated response. It supports investigation; it does not make a sentiment label a diagnosis or a prediction.
Review AI suggestions before relying on them
Ask an AI-assisted analysis to retain the supporting passage for each proposed topic and tone. Review examples with praise and criticism together, ambiguous wording, different languages and references to earlier events. Include responses the system did not flag.
When reviewers disagree, record why. A codebook may need a clearer rule, or the source may simply be ambiguous. Keep approved labels and their version with the reporting record. Re-running a model is not a substitute for preserving the decisions used in an earlier report.
Investigate patterns without calling them causes
Suppose a team observes that some customers who later cancelled had earlier mentioned setup difficulties. This is a question worth investigating. It is not yet evidence that setup difficulty caused cancellation or that a sentiment score predicts it.
Check the observation window, which customers had check-ins, whether the comment preceded the cancellation and how cancellation reasons were recorded. Consider customers with similar comments who remained, customers who cancelled without responding and changes in service or customer mix. Keep an explicit distinction between a reported reason and an inferred explanation.
When the evidence supports only a descriptive statement, use one: “Among customers with a recorded check-in before cancellation, these issues were mentioned.” State the reporting base and missing coverage. A validated predictive model or a causal claim requires additional work beyond descriptive sentiment coding.
Plan the recurring coding and reporting work
Keep tone and topic definitions under your team’s control as feedback grows. Apply revised codes across the authorized response set, preserve review decisions and connect themes with ratings and dated service evidence. This reduces repeated coding and matching work without treating sentiment as a prediction.
A workflow with repeated manual work
- Define from an initial sampleRead material and agree on the codebook.
- Apply it across the datasetCode responses and check the result.
- Revise a definitionReturn to affected material and recode it.
- Reconnect the numbersReconcile coded results with ratings and context, then rebuild the view.
The Sopact workflow
- Your team owns the definitionsDecide what each code means and improve it as you learn.
- Apply coding across the eligible dataAutomate application; people review quality and exceptions.
- Reprocess after a definition changesReapply the revised definition across the configured scope instead of recoding each response by hand.
- Ask across coded text and numbersKeep the response, rating and relevant record context connected; inspect the evidence behind the result.
This compares workflow patterns, not a claim that every research tool requires manual coding or separate files. Some already automate parts of this work; compare the complete cycle.
For this codebook-based workflow, the main saving is repeated application and reconnection—not the removal of human judgment. A changed definition can be reapplied across the configured data while reviewers concentrate on quality, exceptions and interpretation. Coded text stays connected to the relevant ratings and context.
Count the recurring work in ownership cost. Include setup, coding, recoding after revisions, source reconciliation, review and reporting, plus your actual platform and processing expenses. A worked scenario of four cycles of 4,000 responses illustrates 272 fewer annual staff hours; it is an assumption-based example, not a customer benchmark. Existing automation, review needs and implementation effort can substantially change the result.
Adjust the workload assumptions and compare total effort →
A reliable assistant should calculate from the selected records and let a reviewer open the supporting evidence. Check the data scope, definition, denominator and access permissions. Reproducible arithmetic does not make every AI interpretation correct.
Watch: Why Qualitative Analysis Stays Small — And How to Scale It
Watch this 2-minute 58-second explanation of repeated coding, revised definitions and connected analysis. Then estimate the work for your own collection cycle.
Make the dashboard useful to the people who act
Show topic counts, response coverage, dates and review status. Let authorized reviewers inspect the supporting text. Separate new concerns from already reviewed or resolved ones so repeated copies do not appear to be new incidents.
For a trend, check whether the question, timing and respondent mix stayed comparable. A shift in the proportion of negative comments may reflect a different collection moment or a different group of respondents. Explain those changes rather than presenting the line as a complete account of service quality.
Record the action owner and follow-up. An issue being flagged, routed, contacted and resolved are different stages. A dashboard should not count an assigned task as a completed improvement.
Your exercise
- Choose a fictional comment containing both positive and negative experiences.
- Record its tone and topics separately, with supporting words.
- List what the comment establishes and what needs another source.
- Choose the appropriate review owner and next action.
- Define a reporting base for a group summary.
- Write a trend statement that preserves coverage and avoids a causal claim.
Self-check: Could the team act on the finding without assuming that tone proves severity, a topic proves a cause or a repeated complaint predicts an outcome?
Frequently asked questions
Is sentiment the same as satisfaction?
No. Sentiment describes the expressed tone in a particular response. A satisfaction score is an answer to a particular measure. They can inform one another but should not be silently substituted.
Does negative sentiment predict churn?
A negative comment alone does not establish that a customer will cancel. Investigate the issue and relevant history; predictive claims require appropriate evidence and validation.
How should mixed sentiment be reported?
Retain the positive and negative aspects and their topics. A single label should not hide an unresolved difficulty or turn a positive view of one part of the service into approval of everything.