How do you measure customer sentiment?
Short answer
· 5 min read
What are the main ways to measure customer sentiment?
Most CX teams combine several methods because no single source covers everyone. Customer sentiment analysis usually draws on these inputs, each with its own strengths and blind spots.
Method | What it captures | What it misses |
|---|---|---|
Surveys with a free-text field | Direct, invited feedback tied to a specific moment | Only reaches the small share of customers who respond |
Sentiment scoring of support conversations (calls, chats, tickets) | Sentiment from every customer who actually contacted the company | Needs consistent scoring rules to stay comparable over time |
Reviews and app store ratings | Public, unprompted opinion with real stakes for the writer | Skews toward extreme experiences, not the average customer |
Social and community posts | Informal, often immediate reaction to an event or release | Hard to tie back to a specific account or resolution |
Net sentiment as a roll-up metric | One comparable number to track trend direction over time | Hides which topic or segment is driving the change |
Emotion and effort signals (frustration cues, repeat contacts, escalation language) | Early warning before a customer rates anything | Requires conversation-level access, not just survey scores |
Survey response rates are typically low: TruRating's 2026 benchmark report found a median digital customer survey response rate near 10%, which means most invited customers never leave feedback at all. Sentiment scoring of support conversations works differently: it reaches every customer who called, chatted, or filed a ticket, whether or not they answered a survey. Most teams need both: surveys for invited depth, sentiment scoring for full coverage. See sentiment analysis for a fuller definition.
How to measure customer sentiment step by step?
Turn raw feedback into a trackable metric by running the same process on every piece of feedback, then watching net sentiment move by topic instead of guessing from one company-wide score.
Collect every channel into one dataset. Calls, chats, tickets, reviews, surveys, and social posts need to live in the same place before scoring starts.
Score each conversation positive, negative, or neutral at the conversation level, using a consistent rule so scores stay comparable month to month.
Tag topic and journey stage on the same record. One message can carry two topics, for example a complaint that covers both billing and account access.
Calculate net sentiment per topic: share of positive mentions minus share of negative mentions. If billing generated 1,000 mentions in a month, 180 positive and 420 negative, that's 18% minus 42%, for a net sentiment of -24 on billing alone.
Set a baseline over a defined window, such as a rolling 90 days, before calling anything a trend.
Alert on movement past a threshold, broken out by topic and segment, not on the account-wide average.
Read the conversations behind the shift through root cause analysis before assigning a fix.
A single company-wide sentiment number hides exactly this kind of problem. An overall score can look healthy while billing sits at -24 underneath it, because strong sentiment on price or support cancels it out in the roll-up.
Where does automated sentiment scoring get it wrong?
Automated sentiment scoring misreads a predictable set of cases, so sentiment analysis accuracy needs a check, not blind trust. Three failure modes account for most of the errors.
Sarcasm. "Great, another password reset" reads positive on the surface words alone, and irony detection remains a known weak spot for text classifiers generally.
Mixed sentiment in one message. A customer can praise the agent and condemn the policy in the same ticket. Scoring the whole message produces one flat number that hides both signals, which is why sentiment by topic matters more than a single tag per conversation.
Very short replies. "Fine, I guess" or "ok thanks" carry too little text for a model to classify with confidence.
Run a monthly audit: pull a random sample of scored conversations, have two reviewers label the same sample independently, compare their agreement with each other and with the machine score, then route disagreements back into topic definitions and scoring rules.
Compare sentiment movement against CSAT, NPS, and churn for the same segment and period. Treat any relationship as correlation worth investigating, never as proof that sentiment caused the change.
How does OnClarity help with this?
OnClarity (www.onclarity.com) Voice of Customer reads 100% of conversations across calls, chats, tickets, reviews, surveys, and social, rather than a sampled slice. Sentiment and topics are applied to every conversation and tied to the causes behind them, so a dip in billing sentiment comes with the actual complaints driving it, not just a number. Copilot answers questions in plain language over the full dataset, and alerts flag a topic as soon as it starts trending, before it shows up in a quarterly report. OnClarity analyses survey results as part of that same dataset. OnClarity is SOC 2 Type II certified, GDPR and HIPAA Ready, and aligned with Saudi PDPL.
Frequently asked questions
Is customer sentiment the same as CSAT or NPS?
No. Customer sentiment comes from the language customers actually use in calls, chats, reviews, and free text, while CSAT and NPS are direct survey scores on a defined scale. CSAT measures satisfaction with one interaction, NPS measures likelihood to recommend, and sentiment can be scored on conversations where no survey was ever sent.
What counts as a good net sentiment score?
There's no universal benchmark, because a net sentiment score depends on the topic measured and how strictly a model classifies neutral text. A positive score means positive mentions outweigh negative ones, but the more useful read is the trend against your own baseline, broken out by topic, not a comparison to another company's number.
Do you still need human review if sentiment scoring is automated?
Yes. Sentiment analysis accuracy drops on sarcasm, mixed messages, and very short replies, so automated scores need a check. Pull a random sample of scored conversations each month, have two reviewers label the same items independently, and compare their agreement before trusting the topic rules behind it.
Key terms
CSAT
CSAT, or customer satisfaction score, is a post-contact rating averaged across responses and usually collected with a single question immediately after an interaction closes. Because only a small fraction of customers respond, CSAT reflects the experience of people motivated to answer rather than the whole contact population.
Sentiment analysis
Sentiment analysis scores the emotional tone of a conversation to flag risk and measure experience without waiting for a survey response. Applied across every contact rather than a sampled few, it covers the large majority of customers who never complete a satisfaction survey at all.
Net sentiment
Net sentiment is a single directional score for a topic or time period. It is calculated as positive mentions minus negative mentions, divided by total mentions. Unlike a per-item sentiment score, which rates one piece of feedback, net sentiment aggregates many. Buyers should know two limits. Neutral mentions are handled inconsistently across vendors. And the score is volume-sensitive, so a stable reading can mask rising positive and negative shifts that cancel each other out. Example: a CX team compares net sentiment on checkout-flow feedback before and after a release. OnClarity’s Voice of Customer supports voice of customer analytics by classifying feedback by topic and sentiment, so a net sentiment score can be traced back to the comments driving it.
Root cause analysis
Root cause analysis is the practice of tracing a symptom, such as a spike in complaints, back to the underlying process, policy, or product defect that produced it. The alternative, resolving each ticket in isolation, never stops the symptom from recurring. Established methods include 5 Whys (repeated questioning to reach the originating cause), fishbone (Ishikawa) diagrams (mapping possible causes by category), and Pareto analysis (isolating the small number of causes driving most of the volume). In CX, this starts with clustering tickets by theme, then tracing the largest theme to its source. OnClarity’s Voice of Customer automates this by surfacing root causes directly from feedback. Grubhub uses it to turn feedback into insights that shape its roadmap. The real test of success is recurrence: a fix that addresses the root cause should stop the issue from reappearing.
