CSAT stands for Customer Satisfaction Score. It measures how satisfied someone is with one specific thing — a support conversation, a delivery, an onboarding call, a purchase — rather than how they feel about your brand overall.

It's the simplest customer experience metric there is, which is both its strength and its limitation.
One question, asked right after the moment that matters:

How satisfied were you with this experience?

Answers usually come on a 1–5 scale, from "very dissatisfied" to "very satisfied." The score is the percentage of people who picked one of the top options.

Because CSAT is transactional, it tells you something NPS can't: not whether customers like you in general, but whether this interaction, on this day, in this channel went well. That makes it the front-line metric for support teams.

The CSAT formula

CSAT = (Number of positive responses ÷ Total responses) × 100

A "positive response" is conventionally the top one or two options on the scale — a 4 or 5 on a 1–5 scale. This is called top-two-box scoring.

A worked example

You send a post-chat survey to 400 customers. 220 respond:

118  —  Very satisfied
56 — Satisfied
24 — Neutral
12 — Dissatisfied
10 — Very dissatisfied

Positive responses (4 + 5) = 118 + 56 = 174

Total responses = 220

CSAT = (174 ÷ 220) × 100 = 79.1%

Your CSAT is 79%.

Notice what the formula does with the 24 neutral answers: it counts them against you. They sit in the denominator but not the numerator. That's deliberate — a customer who felt nothing is not a satisfied customer — but it's also why your scoring choices need to be documented and then left alone.

The decisions that change your number

Three choices will move your CSAT by several points without anything actually changing in your support quality:

1. Top-two-box vs top-box. Counting only 5s instead of 4s and 5s will typically drop your score 15–25 points. Neither is wrong; consistency is what matters.
2. How you treat neutrals. Some teams exclude 3s from the denominator entirely. This inflates the score and makes benchmarking meaningless.
3. Scale length. A 1–5 scale, a 1–10 scale, and a thumbs up/down all produce different distributions. Pick one and keep it.

Write these down in a metric definition doc. The most common cause of a "sudden CSAT improvement" is somebody quietly changing one of the three.

When and how to ask

Timing beats everything. Send the survey immediately after the interaction closes, while the experience is still fresh. A survey that arrives three days later measures memory, not experience.

Keep it to one question, with an optional comment box. The rating tells you what; the free-text field tells you why. The comments are where the actual operational insight lives — the ratings alone just give you a number to report.

Ask in the channel the conversation happened in. A chat conversation should end with an in-chat rating, not an email. Every channel switch costs you responses.

Watch your response rate. A CSAT built on a 3% response rate is close to noise, and it's biased: people who had an extreme experience — very good or very bad — are far more likely to answer than the satisfied majority. Track response rate alongside the score, and treat sharp CSAT movements on small samples with suspicion.

What counts as a good CSAT score?

The cross-industry average sits at roughly 76–78%, and it has been drifting slowly downward — the ACSI national average fell to 76.7 out of 100 in Q1 2026, with customer complaints up 16% over the same period. The driver isn't worsening service so much as rising expectations: customers' standards for response speed and resolution completeness are climbing faster than most support teams can match.

As a rough interpretation framework:

More than 85% - Excellent (top-quartile territory in most sectors)
75–85% - Healthy
70–75% - Acceptable, with room to work
Below 50% - A structural problem, not a coaching problem

Sector context matters more than the universal number. Ecommerce and retail typically run around 78–85%, helped by fast delivery, easy returns and responsive support. SaaS averages 78–80%. Telecom sits far lower, at 65–72%, because of billing complexity and support friction rather than lower customer expectations.

A single 80% can therefore be unremarkable or excellent depending entirely on where you compete. Which leads to the most useful advice about benchmarks: compare against your own sector, then against your own trend line. Steady improvement in your own number is worth more than hitting somebody else's average.

What actually moves CSAT

If you strip away the survey mechanics, satisfaction in support comes down to a short list of drivers:

1. First response time. This is the single strongest predictor of CSAT. Not resolution time — response time. Customers tolerate a complex issue taking a while; they don't tolerate silence.

2. First-contact resolution. Every handoff, every "let me check with the team," every repeat of the same question costs satisfaction points.

3. Effort. How many steps did the customer take? How many times did they re-explain their problem? Low effort correlates with high satisfaction more tightly than almost anything else.

4. Consistency. Two customers asking the same question should get the same answer. Variance between agents shows up in CSAT as unexplained volatility.

5. Accuracy. A fast wrong answer is worse than a slow right one. This matters enormously when you automate — a confident, incorrect reply is the fastest way to turn a neutral customer into a detractor.

6. Graceful escalation. Automation doesn't damage CSAT. Bad handoffs damage CSAT — being stuck in a loop, or having to start over with a human who has no context.

How VivoChat helps keep CSAT high

Look back at that list and most of it is about speed, consistency and knowing when to hand over. That's precisely what an AI support layer is good at — provided it's designed around the escalation path rather than around deflection targets.

Here's how VivoChat maps to each driver:

Response time drops to seconds. The AI assistant replies instantly, on every channel, at any hour. Given that first response time is the strongest single predictor of satisfaction, this is the highest-leverage change most teams can make — and it applies to the 2am message as much as the Tuesday-morning one.

Repetitive volume stops competing with complex cases. VivoChat's AI agents handle 50%+ of incoming inquiries — the order tracking, returns, shipping and product questions that fill an ecommerce inbox. Your team stops triaging WISMO and gets time to do the complex, emotionally loaded conversations properly. CSAT usually rises on both halves: automated answers get faster, and human answers get more attention.

Answers stay accurate because they're grounded. The assistant learns from your documentation, FAQs, help centre, Notion pages, PDFs and past conversations, so replies are based on your actual policies rather than plausible-sounding guesses. The analytics break accuracy down by knowledge source, which tells you exactly which part of your documentation is letting customers down.

It hands over instead of guessing. When the AI isn't confident, it routes the conversation to a human automatically. This is the design decision that protects CSAT more than any other: the failure mode of a confidently wrong bot is a detractor, and VivoChat is built to escalate rather than improvise.

Consistency by default. Tone and personality are configured once to match your brand, so every customer gets the same voice and the same answer to the same question — removing the agent-to-agent variance that shows up as CSAT volatility.

One inbox across channels. Website chat, email, WhatsApp, Telegram, Messenger and Instagram all land in the same place. Customers get to use the channel they prefer, which independently improves satisfaction, and your team never loses context in a channel switch.

Proactive order updates. Integrated with your shipping tools, VivoChat keeps customers informed about their orders before they have to ask. A WISMO ticket that never gets created can't be rated badly.

Multilingual coverage. Customers get answers in their own language rather than the one your night shift happens to speak.

And you can actually see the drivers. VivoChat's analytics track response time and resolution rates alongside AI accuracy, so when CSAT moves you can identify which underlying driver moved with it — rather than reporting a number nobody can explain.

Setup takes about 15 minutes, and the core platform is free for SMBs, so you can measure the effect on your own conversations before committing anything.

‍