Last updated: September 15, 2026
Customer Satisfaction Score (CSAT) is the one metric almost every support team reports and almost nobody manages well. It shows up on the dashboard, it gets read out in the Monday meeting, and then it quietly drifts up or down without anyone knowing why. This guide explains how CSAT is calculated, what a good score looks like in your industry, why it moves, and how AI-driven support teams push it up without adding headcount.
CSAT measures how satisfied a customer is with a specific interaction, purchase, or touchpoint. Unlike a relationship metric, it is transactional: you ask right after a support ticket closes, a delivery arrives, or an onboarding call ends. The question is usually a variation of "How satisfied were you with your experience today?" on a 1–5 scale.
The most common calculation, as described in Retently's CSAT guide, takes the number of "satisfied" and "very satisfied" responses (4s and 5s), divides by total responses, and multiplies by 100:
CSAT (%) = (Satisfied + Very Satisfied responses ÷ Total responses) × 100
If 20 customers answer and 10 of them pick a 4 or 5, your CSAT is 50%. Some teams average the raw scores instead, but the percentage method is the one most benchmarks use, so stick with it if you want to compare yourself to anyone.
CSAT works because it is short and immediate. Send it within minutes of the resolved ticket, not in a weekly digest. Retently cites SurveyMonkey research showing that a respondent spends about 75 seconds on a one-question survey; every extra question you add lowers the odds they finish it. One question, one scale, an optional comment box. That is the whole survey.
The honest answer is "it depends on who you are competing against." Retently's 2026 benchmark data puts anything above 70% in the "good" zone and anything over 90% in "excellent" territory, but the spread across industries is enormous:
| Industry | Average CSAT (2026) | What "good" looks like |
|---|---|---|
| Consulting | 83 | 85+ |
| Financial services | 81 | 83+ |
| Ecommerce & retail | 77 | 80+ |
| B2B software & SaaS | High 70s | 80+ |
| Education | 68 | 72+ |
| Healthcare | 57 | 65+ |
Source: Retently CSAT Benchmarks 2026. The "good" column is a practical target of a few points above the industry average, not a published threshold.
Zoom out and the picture is sobering. The American Customer Satisfaction Index for Q1 2026 came in at 76.7 out of 100, essentially the same level as in 2013, despite what ACSI estimates as more than $100 billion a year spent on CX initiatives. The same release notes that customer complaints surged 16% in the quarter. Satisfaction is not stagnating because companies stopped caring; it is stagnating because most CSAT programs measure without acting.
Key takeaway: a "good" CSAT is a few points above your industry's average and trending up. A flat 82% in consulting is a warning sign; a rising 70% in healthcare is a success story. Benchmark against your peers, then compete against last quarter's you.
These three metrics get lumped together, but they answer different questions and you need more than one.
Transactional and immediate. Best for diagnosing individual channels, agents, and ticket types. Weak at predicting long-term loyalty because a customer can be satisfied with a refund and still leave.
Relational and periodic. Net Promoter Score tells you how the overall relationship is trending, usually surveyed quarterly. It is the metric your board cares about, but it is too slow and too broad to tell a support lead what to fix this week.
Customer Effort Score measures friction. It is the best early-warning signal for churn among the three, because customers who had to repeat themselves, get transferred, or wait for a callback rarely stay, even when the issue was eventually solved.
A practical stack: CSAT after every resolved interaction, CES on the handful of journeys where effort is the known problem (onboarding, billing disputes, returns), and NPS once a quarter. If you can only run one, run CSAT, because it is the only one that produces a daily signal you can act on.
Most CSAT declines trace back to four causes, and only one of them is "the agent was rude."
Customers judge the interaction from the moment they reach out, not from when an agent picks up. A long queue poisons the survey before anyone has said a word, which is why first response time is usually the first metric to fix when CSAT slips.
Repeat contacts are the silent CSAT killer. The customer may rate the final agent a 5, but they rate the experience a 2. Improving first contact resolution tends to lift CSAT more than any coaching program.
Forcing a WhatsApp customer onto email, or a phone customer into a web form, adds effort at exactly the wrong moment. CSAT by channel is one of the most useful cuts you can build.
Here is the uncomfortable part. Zendesk's 2026 statistics roundup cites Coveo research showing that 56% of consumers rarely complain about a bad experience; they quietly switch to a competitor. Those customers never fill out your survey. A stable CSAT can therefore coexist with rising churn, because the unhappiest people have already left the sample. Track survey response rate alongside the score, and treat a falling response rate as a red flag in its own right.
The commercial stakes are not abstract. The same Zendesk roundup reports that 73% of consumers will switch to a competitor after multiple bad experiences, and more than half will do so after just one, while three in four say they spend more with businesses that deliver a good experience.
The cheapest CSAT win is removing the wait. An AI agent that picks up every WhatsApp, web chat, and email conversation in seconds, resolves the routine ones outright, and hands the rest to a human with full context changes the tone of the entire interaction. Customers are ready for it: Zendesk's data shows 51% of consumers prefer a bot over a human when they want immediate service, and nearly eight in ten say AI bots are helpful for simple issues.
A 1 or 2 rating is not a data point; it is a customer asking to be rescued. Route every detractor response to a human owner automatically, with the transcript attached, and measure how many are contacted within a day. Teams that do this consistently see the customer's second rating land far above the first, and they learn more from ten recovery calls than from a thousand 5-star ratings.
Read the comment field. Tag it by root cause (billing confusion, delivery delay, missing feature, agent tone). You will almost always find that three causes explain most of the low scores. Fix those, remeasure, repeat. This is the same discipline behind good voice-of-customer analysis, just applied to a single metric.
Agents cannot control whether a customer is angry about a policy. They can control whether they acknowledged it, set expectations, and followed up. Coach on those behaviors and CSAT follows. Coach on the raw score and you get agents begging for 5s, which inflates the number while destroying its usefulness.
The highest CSAT scores come from interactions the customer never had to start. Order delay notifications, renewal reminders, and "we noticed you got stuck, here's how to fix it" messages all count as satisfaction moments. Proactive support converts complaint volume into goodwill.
Example: a LatAm retailer running support on WhatsApp saw CSAT sitting in the low 70s with most detractor comments about "waiting for a reply." After deploying an AI agent to handle order status, returns, and store hours instantly, the human team focused on the cases that needed judgment. The score climbed because the wait disappeared for the majority and the difficult cases got more attention, not less.
AI touches CSAT in three places. First, at the front line, an AI agent removes the wait and resolves routine requests around the clock, which addresses the most common cause of low scores. Second, in analysis, AI reads every comment, tags root causes, and surfaces the three issues worth fixing this week instead of leaving that to whoever has time on Friday. Third, in recovery, it routes detractors to the right human instantly, drafts the follow-up, and tracks whether the loop got closed.
This is exactly what Eva, Darwin AI's customer experience agent, is built for: she answers on WhatsApp, email, and web chat in seconds, resolves the routine questions end to end, escalates with full context when a human is needed, and runs the CSAT survey and follow-up automatically. The team stops firefighting queues and starts managing satisfaction.
What AI does not do is fix a bad policy, a broken product, or an unclear invoice. It makes those problems visible faster. Whether CSAT rises after that depends on the same thing it always has: whether someone acts on what the customers said.
Answer every customer in seconds and watch CSAT follow
Eva resolves routine support conversations on WhatsApp, email, and chat, runs your CSAT survey, and routes every detractor to a human, automatically.
Meet EvaAccording to Retently's 2026 benchmarks, anything above 70% is generally considered good and above 90% is excellent, but industry averages range from the mid-50s in healthcare to the low 80s in consulting and financial services. Compare yourself to your own sector and to your own trend.
CSAT measures satisfaction with a specific interaction and is surveyed immediately after it. NPS measures the overall relationship and likelihood to recommend, and is usually surveyed quarterly. Most teams need both: CSAT to fix this week's problems, NPS to track the relationship.
After every resolved interaction, within minutes. Keep it to one question and an optional comment. Longer surveys lower completion; Retently cites SurveyMonkey data showing respondents spend roughly 75 seconds on a one-question survey and progressively less per question as surveys grow.
Yes, mainly by eliminating wait time and repeat contacts, which are the two largest drivers of low scores. Zendesk's research shows 51% of consumers prefer a bot when they want immediate service, and nearly eight in ten say AI bots are helpful for simple issues. AI also speeds up comment analysis and detractor follow-up.
Survivor bias. Coveo research cited by Zendesk finds 56% of consumers rarely complain about a bad experience and simply switch. Those customers never answer your survey, so the score stays flat while the base shrinks. Track response rate and churn next to CSAT.