Glossary · CSAT

What is CSAT?

CSAT (Customer Satisfaction Score) measures how satisfied a customer is with a specific interaction, product, or overall experience. Typically collected via post-interaction surveys asking customers to rate their satisfaction on a 1–5 or 1–10 scale, CSAT is reported as the percentage of respondents rating 4 or 5 (or 9 or 10), the “satisfied” or “highly satisfied” group.

CSAT is tactical (individual interaction quality) in contrast to NPS which is strategic (brand loyalty). AI-powered contact centers now measure CSAT on 100% of interactions through predictive analysis instead of the traditional 5–15% survey response rate.

CSAT formula

Standard survey-based CSAT:

CSAT = (Number of "satisfied" responses / Total responses) × 100

Where “satisfied” is typically a 4 or 5 on a 5-point scale, or 9 or 10 on a 10-point scale.

Some organizations use a straight average:

Mean CSAT = Sum of all ratings / Number of responses

Both are valid; stick with one methodology to enable time-series comparison.

Typical CSAT questions

Transactional (post-interaction):

“How satisfied were you with the support you received today?”
“Rate your recent experience with our team.”

Relationship (periodic):

“Overall, how satisfied are you with [Company]?”
“How likely are you to recommend us?” (this is actually NPS)

Specific:

“How satisfied were you with the product?”
“How satisfied are you with the delivery experience?”

One question, answered immediately after the event, yields the best data.

Why CSAT matters

Customer retention: unhappy customers leave. CSAT predicts churn.
Revenue: satisfied customers spend more, refer more, renew more.
Operational quality: low CSAT surfaces systemic issues.
Agent performance: CSAT segmented by agent identifies coaching needs.
Product feedback: CSAT by product category flags quality issues.

CSAT benchmarks

Typical CSAT by industry (percent satisfied, 4+ on a 5-point scale):

Industry	Typical CSAT
Software / SaaS	85–90%
Retail / ecommerce	80–90%
Hospitality	85–95%
Financial services	80–88%
Healthcare	75–85%
Telecom / cable	65–75% (classically lower)
Government	65–78%
Utilities	70–80%

Best-in-class organizations in each category exceed these benchmarks. Sustained CSAT above 90% in any industry is a real competitive moat.

The survey response problem

Traditional survey-based CSAT has structural problems:

Response rate is 5–15% on average. The 85–95% who don’t respond are silent.
Responders are not random. Extremes respond (very happy and very unhappy); the middle stays silent.
Survey fatigue. Response rates decline over time.
Recency bias. Customers remember recent experiences more vividly.
Channel bias. Email surveys skew older; in-app surveys skew younger; phone IVR surveys skew older + less tech-savvy.

The result is CSAT based on a biased minority voice.

AI-based CSAT (predictive CSAT)

Modern AI-powered contact centers measure CSAT on 100% of interactions without surveys. Signal inputs:

Sentiment analysis: tone of voice, language patterns
Frustration detection: specific markers (sighs, interruptions, complaint language)
Satisfaction language: thank-you phrases, positive affirmations
Interaction flow: escalations, callbacks, hold time
Resolution language: “that solves my problem” vs. “I’m still confused”

The AI produces a predictive CSAT score for every interaction that correlates with survey CSAT where surveys are available. Coverage jumps from 10% to 100%.

DialPhone’s AI Interaction Analytics delivers predictive CSAT on every interaction across voice and all digital channels.

Acting on CSAT data

The data is useful only if you act on it:

Low-CSAT calls flagged for supervisor review within minutes, not days
Callbacks scheduled for unhappy customers before they churn
Trend analysis: which issue types drive lowest CSAT?
Agent coaching: which reps have consistently low CSAT and why?
Feedback loops to product: which product/feature issues show up in low-CSAT calls?
Real-time intervention: AI flags mid-call frustration so supervisors can barge

CSAT vs. NPS vs. CES

Three common customer metrics:

Metric	Question	Measures	Best for
CSAT	How satisfied were you with this?	Specific interaction quality	Transactional, tactical
NPS	Would you recommend us?	Overall relationship / loyalty	Strategic, brand-level
CES (Customer Effort Score)	How easy was it?	Friction in the experience	Self-service, support optimization

Use CSAT for interaction-level measurement. Use NPS for relationship-level. Use CES when investigating friction.

Segmenting CSAT

Average CSAT hides problems. Segment by:

Agent: identify coaching and recognition targets
Issue type: find categorical satisfaction gaps
Channel: does voice outperform chat?
Product or feature: what’s driving low satisfaction?
Customer segment: new customers often have different CSAT patterns
Time of day or shift: catch staffing quality variance
Region: regional cultural differences in rating
Interaction duration: how does CSAT scale with wait time?

DialPhone CSAT features

Predictive CSAT on 100% of interactions via AI Interaction Analytics, no survey bias
Post-interaction survey automation: SMS or email surveys sent automatically after calls
Real-time frustration detection: supervisor alerts during live calls
CSAT segmentation: by agent, issue, channel, customer segment
Root-cause CSAT analysis: automatic correlation with call drivers
Integration with CRM: CSAT scores written to Salesforce / HubSpot contact records

Example

A 250-agent B2B SaaS customer success operation had 12% survey response rate and reported CSAT of 89%. After deploying DialPhone Professional with AI Interaction Analytics:

Predictive CSAT measured on 100% of interactions showed actual CSAT of 76%
The 24% non-responders were disproportionately frustrated
Root-cause analysis identified 3 onboarding failures causing 40% of low-CSAT calls
Fixing those 3 issues lifted measured CSAT from 76% to 84% in 90 days
Churn on affected customer segment dropped 18% year-over-year

The survey-only approach was masking a real problem. Measuring 100% uncovered it.

CSAT survey question templates that actually work

Most CSAT surveys fail because they ask too much, too late, or the wrong way. Below are six question templates that consistently produce usable data. Pick two or three — never more. Every additional question costs roughly 30–50% in completion rate.

Overall satisfaction (5-point scale) — “How satisfied were you with the support you received today?” Anchor the scale “Very dissatisfied” → “Very satisfied”. The 5-point Likert scale is the industry default because it produces cleaner distributions and is faster than 1–10. Use this as the primary CSAT score.
Customer Effort Score (CES) follow-up — “How easy was it to resolve your issue today?” Use a 1–5 scale anchored “Very difficult” → “Very easy”. CES catches friction that overall CSAT misses. A customer can rate a friendly agent 5/5 while still rating the process 2/5 — that gap is your roadmap.
Resolution confirmation — “Was your issue resolved today?” Binary yes/no. The operational reality check. CSAT scores look fine until you cross-tabulate them with this field; unresolved-but-satisfied customers churn anyway. Pair with First Call Resolution (FCR) tracking.
Free-text follow-up — “What’s one thing we could have done better?” Optional, single open field. Roughly 18–25% of respondents fill it in — the highest-signal qualitative data you’ll get. Run it through AI topic clustering monthly to surface emerging issues.
Agent-specific rating — “How would you rate the agent who helped you today?” Same 5-point scale. Keep this separate from overall satisfaction so you can disambiguate “great agent, broken product” from “broken agent, fine product.” Feed it into coaching workflows.
NPS overlap question — “How likely are you to recommend us to a colleague or peer?” 0–10 scale. Including a relationship-level NPS question on a transactional survey gives you one touchpoint that feeds both metrics. Only ask it on 1 in 5 surveys to avoid fatigue.

Survey length matters more than question design. Two questions complete at 60–75%. Five questions complete at 30–45%. Ten questions complete at under 20%.

When to trigger a CSAT survey (and when not to)

Survey timing is the biggest predictor of response rate and data quality. Trigger at the right moment and you collect representative data; trigger wrong and you collect noise.

Trigger CSAT after:

Resolved support tickets — closed status, customer-confirmed. Send within 5–15 minutes of closure.
Voice calls — within 5 minutes of disconnect via SMS or IVR. Surveys at 5 minutes post-call see 28–35% response rates. The same survey 24 hours later drops to 8–12%.
Post-purchase delivery confirmation — 24–48 hours after delivery, not order placement.
Onboarding milestones — day 7 and day 14 are the high-signal windows for B2B SaaS.

Don’t trigger CSAT after:

Unresolved tickets — measures frustration with the gap, not the service.
Repeat-issue tickets — the same customer rating the same agent the same week introduces severe bias.
Self-service article visits — no human interaction to rate. Use a thumbs-up/down widget instead.
Failed deliveries or billing disputes — these need a resolution loop first, then CSAT on the recovery.

The cleanest CSAT programs trigger on a narrow set of confirmed-resolution events with tight timing windows.

CSAT measurement methodology

The standard CSAT formula is the percentage of respondents rating the top two boxes (4 or 5 on a 5-point scale):

CSAT % = (count of 4 + 5 ratings / total responses) × 100

A 1,000-response survey with 820 ratings of 4 or 5 produces a CSAT of 82%. Some organizations report a mean score instead (sum of all ratings / count), but the top-two-box percentage is the industry convention and the only one comparable across vendors.

Industry benchmark ranges (B2B and consumer averages):

SaaS B2B: 70–85% is typical, 85%+ is excellent. Best-in-class is roughly 88–92%.
E-commerce: 75–85% typical, with logistics-heavy verticals at the low end and DTC brands at the high end.
Telecom / cable: 65–75% — structurally lower because of involuntary contracts and provisioning friction.
Healthcare: 75–85% — clinical satisfaction ratings, distinct from administrative ratings which run lower.
Banking and financial services: 70–80% typical.

Common anti-patterns that inflate or distort the number:

Reporting raw average instead of top-two-box inflates the headline by 5–10 points and makes scores incomparable to competitor data.
Using a 1–10 scale and reporting “satisfied” as 7+ — the industry norm is 1–5 with top-two-box at 4–5. Mixing scales kills time-series comparability.
Measuring at the wrong moment — surveying 7 days after a call captures the customer’s overall mood, not the interaction. The number drifts.
Selectively suppressing low-scoring responses (“the customer was clearly venting”) — this is data laundering and erodes the metric’s value within two quarters.

Lock the methodology once and don’t change it. Time-series trend matters more than the absolute number.

CSAT vs NPS vs CES — when to use each

The three customer-experience metrics solve different problems. Most mature contact centers track all three; the discipline is knowing which one to act on for which decision.

Metric	Question	Time horizon	Use case
CSAT	Were you satisfied with this interaction?	Short-term, transaction-specific	Post-call, post-ticket, post-purchase — tactical quality measurement
NPS	How likely are you to recommend us?	Long-term, relationship-level	Quarterly relationship pulse — strategic loyalty and growth indicator
CES	How easy was it to complete that task?	Short-term, friction-specific	Post-resolution, post-self-service — friction detection for support and product

CSAT tells you whether the interaction was good. NPS tells you whether the relationship is healthy enough to drive referrals. CES tells you whether the experience took too much customer effort regardless of outcome. A customer can rate a call 5/5 on CSAT, 8/10 on NPS, and 2/5 on CES — and that pattern is gold because it tells you the agent saved a broken process.

Use CSAT for tactical operational management (agent coaching, queue routing, ticket quality). Use NPS for executive-level strategy (board metric, growth modeling, customer-segment health). Use CES to find the specific bottlenecks driving churn risk that CSAT alone misses.

CSAT in 2026: what AI changes

Predictive CSAT — AI scoring of customer sentiment in real time from voice transcripts or chat text — is now standard in mid-market and enterprise contact centers. The shift is from sample-based, after-the-fact measurement to continuous, real-time scoring on every interaction.

Predictive CSAT models trained on labeled call transcripts now hit 78–85% correlation with actual survey responses. The model reads tone, language patterns (frustration markers, escalation language, thank-you phrases), and interaction flow (callbacks, holds, transfers) to produce a real-time score.

Practical applications in production:

Mid-call routing: low-predicted-CSAT calls auto-escalate to senior agents before the customer asks
Save-the-customer playbooks: when predicted CSAT drops under a threshold, the system triggers retention scripts or callback scheduling
Coaching at scale: managers review the 5% lowest-predicted-CSAT calls per agent per week instead of random sampling
Real-time alerts: supervisors get notifications when predicted CSAT crashes mid-call so they can barge or whisper

The caveat that gets ignored: predictive CSAT is a tactical layer, not a replacement for survey data. The model still needs labeled survey responses to train and recalibrate quarterly. Treating predictive scores as ground truth is how organizations end up with confidently wrong dashboards. Run both — surveys train the model, predictive scoring multiplies the coverage.

CSAT software and survey tools in 2026

The market splits into six categories. Pick the layer that matches your size and operational maturity — over-buying enterprise voice-of-customer (VoC) software for a 30-seat team is a common, expensive mistake.

Native in CCaaS platforms — DialPhone, Genesys, NICE CXone, and Five9 ship post-call SMS and IVR survey automation as a bundled feature. Surveys auto-trigger on call disconnect, write back to the agent record, and feed dashboards. Best for support orgs that live inside the contact center.

Standalone survey platforms — SurveyMonkey, Typeform, Qualtrics CoreXM. Roughly $25–99 per seat per month. Flexible logic, branding, multi-channel distribution. Best when you need surveys outside the contact center.

Voice-of-customer suites — Medallia, Qualtrics XM, InMoment. Enterprise-tier — $1,000+ per seat per year. Cross-channel CSAT, NPS, CES with text analytics and executive dashboards. Best for orgs with a dedicated CX team.

Help-desk native CSAT — Zendesk CSAT, Freshdesk CSAT, Intercom surveys. Bundled with the help desk at no extra cost. Triggers on ticket close, single-question post-interaction. Best for support orgs that route everything through a help desk.

Embedded in-product surveys — Pendo, Sprig, Userpilot. Triggered by in-app behavior. Best for SaaS product teams measuring feature satisfaction, not contact-center CSAT.

Free DIY — Google Forms, Microsoft Forms, Tally. Zero cost, no automation. Fine for sub-20-seat operations. Breaks down at any meaningful volume.

Recommendation by org size: SMBs (under 50 seats) use the native CCaaS or help-desk CSAT — don’t pay twice. Mid-market (50–500 seats) layers Qualtrics CoreXM or Medallia Concierge on top of CCaaS for cross-channel reporting. Enterprise (500+ seats) standardizes on Medallia or Qualtrics XM with the contact center feeding raw data in.

CSAT frequently asked questions

What’s a good CSAT score for B2B SaaS?

B2B SaaS CSAT typically lands between 70% and 85%, with anything above 85% counted as excellent and 88–92% considered best-in-class. The number depends heavily on segment — self-serve SMB customers rate lower than enterprise customers with named CSMs, and product-led companies generally rate lower than service-led ones. Don’t chase the headline number in isolation; track the trend over rolling 90-day windows and segment by customer tier, product line, and interaction channel before drawing conclusions.

How do I calculate CSAT?

The standard CSAT formula is (count of 4 and 5 ratings / total responses) × 100, expressed as a percentage. On a 1,000-response survey with 820 top-two-box ratings, CSAT is 82%. This top-two-box method is the industry convention and the only one comparable across vendors and benchmark studies. Avoid reporting raw averages (which inflate the number by 5–10 points) or mixing 5-point and 10-point scales in the same time series.

Is CSAT the same as customer satisfaction survey?

CSAT is the score; the customer satisfaction survey is the instrument that collects the data. The score is one number — a percentage — calculated from the survey responses using the top-two-box formula. The survey itself can include CSAT, NPS, CES, free-text questions, and demographic fields. Most teams use the terms interchangeably in casual conversation, but in technical reporting CSAT specifically refers to the score, not the instrument.

How long should a CSAT survey be?

Keep it to two or three questions maximum. Surveys with two questions complete at 60–75%; five-question surveys drop to 30–45%; ten-question surveys fall under 20% completion and the remaining respondents skew toward outliers with strong opinions. The cleanest design is one CSAT rating question, one optional free-text follow-up, and nothing else. Save the longer surveys for quarterly relationship pulses where you can email customers and offer an incentive for completion.

Should I use a 5-point or 10-point CSAT scale?

Use a 5-point scale anchored “Very dissatisfied” → “Very satisfied.” The 5-point Likert scale is the industry default because it produces cleaner response distributions, is faster for customers to answer, and matches the benchmark data you’ll be compared against. The 10-point scale belongs to NPS specifically (0–10) and shouldn’t be reused for CSAT. Mixing scales kills time-series comparability and makes vendor benchmarking meaningless.

Can AI predict CSAT before customers respond?

Yes — predictive CSAT models trained on labeled call transcripts now correlate at 78–85% with actual survey responses. The model reads tone, language patterns, and interaction flow signals (escalations, holds, frustration markers) to produce a real-time score on every interaction. This solves the coverage problem (surveys collect 5–15% response rates; AI scores 100%). The caveat is that predictive CSAT is a tactical layer on top of actual survey data, not a replacement — the surveys still train and recalibrate the model. Run both.

Contact center AI agent assist
Omnichannel contact center platforms
Net Promoter Score (NPS) — the relationship-level loyalty metric that pairs with CSAT
First Call Resolution (FCR) — the operational driver most correlated with CSAT
Average Handle Time (AHT) — the efficiency metric that must be balanced against CSAT
AI Contact Center pricing
AI Interaction Analytics