CSAT Surveys: How to Measure Customer Satisfaction and Act on It
October 16, 2026
A support manager opens the monthly report and sees a CSAT of 81%. Last month it was 80%. The number is fine, so nobody asks a question. Meanwhile, a customer who gave a 1 out of 5 three weeks ago is still waiting for someone to call back, and the same root cause that upset them has just upset eleven more people. Nothing in the report shows any of this.
Nobody in that story did anything wrong. The survey went out, the responses came in and the dashboard updated. The failure is quieter: the score is treated as a result to report instead of a signal to act on. Most support teams collect customer satisfaction scores and almost none of them change anything because of them.
This article covers how to design a CSAT survey that customers actually answer, how to read the results without fooling yourself and how to build the loop that turns a low score into a recovered customer and a corrected process. The first sections are neutral toward any platform and apply even if your current tools are a spreadsheet and a shared mailbox. A later section explains how HaloCSM supports the program.
CSAT stands for customer satisfaction score. In its common form, a customer is asked one question right after an interaction, such as "How satisfied are you with the support you received?", and answers on a scale, usually from 1 to 5. The score is the percentage of respondents who chose a satisfied rating, which on a five-point scale means a 4 or a 5. If 80 of 100 respondents answered 4 or 5, the CSAT is 80%.
That simplicity is why CSAT is so widely used, and also why it is so often misread. It captures the customer's reaction to one interaction at one moment. It does not measure loyalty, it does not predict renewal on its own and it does not explain why the customer felt the way they did unless you ask.
Three scores tend to get mixed together, and each answers a different question. CSAT asks how satisfied the customer was with a specific interaction, which makes it the right tool for judging the quality of support after a ticket or a conversation. Net Promoter Score asks how likely the customer is to recommend you. It measures the relationship, and it is usually collected a few times a year rather than after every ticket. Customer Effort Score asks how easy it was to get the problem solved, which is useful for finding friction in the process rather than in the agent.
For a support operation, CSAT is the instrument worth running continuously, because it attaches a customer reaction to a specific case, agent, channel and category. That is the level at which the work happens, and therefore the level at which you can change it.
When we review support operations, the same four patterns explain most stalled programs. All four are design problems, not technology problems.
The first is a survey that goes to everyone and is read by no one. A large volume of responses with no owner produces a dashboard, not a decision. The second is the absence of any workflow for low scores. A customer leaves a 1 or a 2 out of 5 and nothing happens, so the customer concludes that the survey was decorative and stops answering the next one.
The third is using the score to rank individual agents. Once the number decides who gets recognized or disciplined, agents learn to manage it: they ask customers for a high rating, avoid difficult cases or close tickets before the survey fires. The data gets worse exactly when the organization starts to depend on it. The fourth is reading the score as one average. A company-wide CSAT of 80% can hide a strong channel and a failing one inside the same number, and the average gives the team no reason to look for the failing one.
The cost of these patterns is not only analytical. Each low score that goes unanswered is a customer who has told you, in writing, that something went wrong, and who then learned that telling you changes nothing. That is a retention risk that sits unmeasured behind a healthy-looking average.
It helps to know what a typical program looks like before setting a target. Survicate analyzed CSAT surveys from 2025 data: 1,259 surveys run by 370 companies, each with at least 30 responses. The median CSAT score was 80.0%, and the middle half of surveys fell between 64.8% and 90.0%. The median response rate was 7.5%, with the middle half between 3.6% and 17.1%.
Two details matter more than the headline. First, the way a survey is delivered changes both the score and the response rate. Survicate reported a median response rate of 20.2% for mobile surveys against 6.5% for widget surveys, and a median CSAT of 80.9% on mobile and 70.1% on widgets, compared with 86.2% for surveys shared by link. The exact numbers will differ for your customers, but the lesson is stable: how and where you ask shapes what you hear. Second, a score below 64.8% sits in the bottom quartile and a score above 90.0% in the top quartile, which gives you a sensible range in which to place yourself.
These figures come from the customer base of a survey software company. They do not describe support teams in Latin America or in your specific industry. Use them to calibrate expectations, and treat your own trend over time as the real measure. A team that moves from 72% to 78% over two quarters has learned more than a team that merely matches a median.
Response rate is the first test. When few customers answer, the ones who do are usually the very happy and the very angry, and the quiet middle that makes up most of your base stays invisible. A handful of design choices decide whether people respond.
Ask one question first. A single rating takes seconds, and everything beyond it is a bonus. Ask at the right moment: send the survey when the case is resolved, while the experience is fresh, and not days later when the customer has moved on. Keep the scale stable. Choose a five-point scale or a thumbs up and down and keep it for at least a year, so that the trend means something. Add one open question that the customer can skip, such as "What is the one thing we could have done better?", because a single sentence of free text produces more useful detail than a page of rating scales.
Limit how often the same customer is asked. Someone who opens six tickets in a week should not receive six surveys, and a rule that you agree on and respect protects both the response rate and the relationship.
A short set of questions works well in most operations: how satisfied the customer was with the support received, whether the issue was resolved with a yes or no, how easy it was to get help and what could have been done better. The resolved question is easy to skip and valuable to keep. A customer can be polite and satisfied with the agent while the problem is still open. Comparing the two answers is how you find cases that were closed on paper but not in practice, which connects directly to what we explained about first contact resolution.
Survey data is easy to misread, and the errors usually flatter the team. A few habits keep the analysis honest.
Look at the response rate before the score. A CSAT that rises while the response rate falls may only mean that unhappy customers stopped answering. Segment before you conclude: break the score down by channel, case category, customer segment and team, because an average of 80% often hides 90% on one channel and 65% on another. Read the comments attached to low scores first. The numbers tell you where to look and the words tell you what to fix.
Compare satisfaction with resolution. A satisfied customer whose case was reopened is a different signal from a satisfied customer whose case stayed closed. Watch the trend, not the single reading: one bad week after a product release is information, six bad weeks is a pattern. And be careful with small samples. A team of four agents with ten responses each cannot be ranked from CSAT alone.
It also helps to tie the score to operations. A drop in satisfaction often follows a change in the process rather than in the people. Customers who have to switch channels and repeat their story are a frequent cause, as we explained in why channel switching is costing you customers. Tickets that pass through several queues before reaching the right person are another, and the CSAT result is usually the first place that cost becomes visible to the customer.
A CSAT program has two loops. The inner loop deals with the individual customer who gave a low score. The outer loop deals with the pattern behind many low scores. Most teams run neither, and a simple version of each is enough to change the results. Both loops are processes your team runs. They do not depend on a particular tool, and they work with whatever platform you have.
For the inner loop, define the trigger in advance: any score of 1 or 2, or any answer of "not resolved", goes to a named person, usually a team lead or a senior agent. A task that belongs to the whole team belongs to no one. Agree on how quickly the customer will hear back, for example the same business day, and measure it. The contact itself should be personal: acknowledge the problem, ask what would have made it right and fix it if you can. A generic apology confirms the customer's suspicion that the survey was decorative. Finally, record the cause and the outcome on the case, so that the outer loop has data to work with.
For the outer loop, group the low scores by cause every month, using a short list of reasons such as slow response, wrong answer, bounced between teams, unclear instructions or product defect. Pick the one or two largest causes, because a team cannot act on fifteen findings. Assign an owner and a date for each fix. Some will be knowledge fixes, some process changes and some product feedback. Then tell customers and agents what changed, since customers who see their feedback produce a change are the ones who answer the next survey. The following month tells you whether the fix worked.
The outer loop often points to the same root causes. Missing or outdated answers send customers to an agent for things they could have solved alone, which is why knowledge base governance belongs on every service leader's agenda. Cases that stall between levels of support are another frequent cause, and a team that sees them in its survey comments has a concrete reason to fix its handoffs.
The fastest way to ruin a CSAT program is to turn it into a scoreboard. Three rules protect it. First, do not tie pay or discipline to individual CSAT scores. Use them in coaching conversations, where a low score on a specific case is a reason to review the case together. Second, never allow agents to ask customers for a good rating. It biases the data and it feels manipulative to the customer.
Third, review the cases where the agent did everything right and the score was still low. Those usually point to product, policy or process problems that the agent cannot fix, and the agent should not carry the blame for them. Used this way, CSAT becomes a tool the team trusts, and that trust is what keeps the response data honest.
CSAT is more persuasive when it sits beside operational and business measures. Pair it with first contact resolution, time to resolution, reopen rate, backlog and, where you can measure it, churn or renewal in the accounts that gave the lowest scores. We covered how to frame these for an executive audience in customer service KPIs your CEO actually cares about.
One useful view is to follow low scores into the account. If three of your ten largest customers gave low scores in a quarter, that is an account risk and not a support statistic. Routing that signal to account managers is where service data starts to influence revenue, and it is usually the argument that earns the program continued attention from leadership.
The measures that show the program is working are mostly not the score itself. Track the response rate, the share of low scores that received a personal follow-up within the agreed time, the number of root causes identified and fixed, and the CSAT trend by channel and category read alongside first contact resolution and reopen rate. Those measures show whether the organization is listening and acting, which is what customers will judge.
Halo is a partner of GB Advisors, and we implement it for clients across Latin America and the Caribbean. HaloCSM is its customer service management product, and it covers the mechanical part of the program: collecting the score at the right moment and showing it at the level where you can act.
HaloCSM includes configurable CSAT surveys that are triggered automatically when a ticket is closed. That removes the manual work of sending surveys and puts the question in front of the customer at the moment the experience is fresh. The results are tied to individual agents, teams and ticket categories, and they feed performance dashboards. That structure supports the segmentation recommended earlier: you can see the score for a team, a category or an agent instead of one company-wide average, and you can compare it with the rest of your service data in HaloCSM reporting and analytics.
HaloCSM works across the channels where customers actually reach you: email, call management, SMS, social media, chat, a self-service portal and a web widget. Omnichannel is a listed capability, which matters because satisfaction should be read per channel, as the benchmark data showed. HaloCSM also includes Customer SLA Management, Workflows, Rules, Notifications and Knowledge Management, so the service data behind a low score, such as SLAs and the knowledge articles available to the agent, sits in the same product as the survey result.
Halo AI adds auto-summarization of threads and calls, intent routing, priority prediction and sentiment flags, which gives teams another signal to read next to the survey score. HaloCSM also offers native bi-directional CRM synchronization and Sales and CRM capabilities, which supports the account-level view described above, where low scores are followed into the customers that matter most to revenue.
What the product does not replace is the discipline of the two loops. A survey that fires on closure gives you the score. The named owner, the response time, the monthly review of causes and the decision about which fix comes first are decisions your team makes, and they decide whether the number ever moves. This is the point where our advisory work usually starts: designing the survey, the follow-up routine and the review cadence first, and then configuring HaloCSM to support them.
Several features of the regional market shape how a CSAT program should be built and introduced. These are patterns we see in our advisory work, not universal rules, and they are worth testing with your own customers.
Customers in the region use messaging channels heavily. Many contact companies through chat and expect to be answered there, so a survey that sends them to a long web form will lose most of them. Keep it to a single question and make the rest easy to skip. Language and tone matter as well. Write the survey in plain, natural Spanish or Portuguese for each market, not as a literal translation, and check that regional expressions are understood by the people who will read them.
Politeness can inflate scores. In some markets customers are reluctant to give a harsh rating to a person who helped them, so the open comment and the resolved question are often more revealing than the number. Operations that span several countries need a local reading: a score that looks low in one country may reflect a different response style rather than worse service, so compare each market with its own history before comparing markets with each other.
Data protection rules also differ by country. Free-text comments can identify a person, so check what applies to survey responses in each market and involve legal or compliance teams before launch. Organizations that operate in regulated sectors will usually have a view on this already, and it is better to ask early than to redesign the survey after the first complaint.
Before you change any tool, work through a few honest questions about the program you run today.
If you answer no to more than two of these, the constraint is the process and not the platform. Fixing it first will make any tool you choose more useful.
"Our CSAT is already high, so the program is working." A high average with a falling response rate can mean that unhappy customers stopped answering. Check the response rate and the segments before celebrating, because the average is the number least likely to show a problem.
"Following up on every low score is too much work." It is less work than it sounds. With a median response rate of 7.5% in Survicate's benchmark and a median CSAT of 80.0%, the number of low scores for a given team in a week is usually small enough for one lead to handle. The effort is also the point: it is the part of the program that customers notice.
"Agents will game the score if we measure them." They will if the score decides pay or ranking. Use CSAT for coaching and case review, never ask customers for a rating and review the cases where the agent did everything right. That keeps the data honest.
"We already run NPS, so we do not need CSAT." The two answer different questions. NPS measures the relationship a few times a year, while CSAT attaches a reaction to a specific case, which is the only level at which a support team can act on it.
"We should wait until we have the right tool." Waiting is the more expensive choice. The two loops, the owner and the monthly review of causes can start with a short survey and a spreadsheet. The tool makes the program easier to run, but it does not create the habit.
A CSAT survey is not a technology project. It is a commitment to ask customers how they were treated and to act on what they say. The tool matters because it decides how easily customers can answer and how quickly your team can respond, but the program succeeds or fails on the follow-up.
Start small. Pick one channel, write a short survey, decide in advance who will contact every unhappy customer and agree on a monthly review of the causes. If that loop works, scale it, and let HaloCSM carry the automatic survey, the segmentation by agent, team and category and the reporting. If it does not work, no platform will fix it.
GB Advisors helps organizations across Latin America and the Caribbean design customer service programs and implement the platforms that support them, with a neutral assessment first and a recommendation second. Talk to our team about a short diagnostic of how your support operation listens to customers today, and we will help you design the first survey and the follow-up loop.
