Customer satisfaction (CSAT) is the metric most directly tied to what a client cares about: did the customer leave the interaction happy? It is also one of the easiest metrics to misread. The calculation takes one line. Interpreting it well takes an understanding of how the score is built, what it hides, and how much noise is in it.

How the standard score is built

Most post-interaction surveys ask a single question, such as “How satisfied were you with your interaction today?”, on a 1–5 scale. The common CSAT calculation is top-two-box: the percentage of responses that are 4 or 5. Some programmes use only 5 (top-box), and some use a 1–10 scale with 8 and above counted as satisfied. Each convention produces different numbers for the same customers, so comparisons across clients only make sense when the scale and the cut-off are the same.

The bottom of the scale has its own name. DSAT (dissatisfaction) is usually the percentage of 1 and 2 ratings. A response of 3 is often treated as neutral and counted in neither.

Two agents, one score

Here are the survey results for two agents on the same process in the same month, each with 50 responses:

RatingAgent AAgent B
52820
4917
3511
221
161
CSAT (4+5)74%74%
DSAT (1+2)16%4%
Mean rating4.024.08

Both agents score 74%. Their mean ratings are nearly identical. Yet Agent A leaves eight customers in fifty clearly unhappy, while Agent B leaves two. Agent A has more delighted customers, and more angry ones. That pattern often means inconsistency: excellent on routine calls, poor on difficult ones, or occasionally slipping on a behaviour such as rushing to close. Agent B is consistently good without being outstanding. These two need completely different coaching, and a dashboard showing only CSAT would treat them as the same.

The practical lesson is to look at the distribution, or at least report CSAT and DSAT together. DSAT responses are also where the most useful verbatim comments usually sit.

How much of the score is noise

Fifty responses feels like a decent sample. Statistically, it is small. For a proportion around 74% with 50 responses, the approximate 95% margin of error is about ±12 percentage points. In plain terms, an agent whose “true” CSAT is 74% could easily show anything from the low 60s to the mid 80s in a given month purely by chance.

This has direct consequences for how CSAT is used on the floor. Ranking agents on monthly CSAT, or linking incentives to month-on-month changes of a few points, means rewarding and penalising random variation. Agents notice this, and it erodes trust in the whole quality system. Better approaches include using a rolling three-month window for individual scores, setting a minimum number of responses before an individual score is reported, and focusing individual coaching on DSAT verbatims rather than the percentage itself. At team or process level, with hundreds or thousands of responses, the noise shrinks and monthly trends become meaningful.

Who answers surveys

Response rates for post-call surveys are often low, frequently in single digits. The customers who reply are not a random sample. People with strong feelings, either very pleased or very annoyed, are more likely to respond than people who found the call perfectly fine. The survey channel matters too: an IVR survey immediately after the call captures a different group from an SMS link sent an hour later, and an email survey two days later reflects whether the problem was actually fixed.

Some practices quietly distort the sample. If agents can choose which customers get transferred to the post-call IVR survey, a few will skip the unhappy ones. Systems that offer the survey automatically to every eligible customer, or to a random selection, avoid this. Coaching agents to ask for a high rating (“please rate me 5”) is another common distortion, and many clients explicitly prohibit it.

Agent-driven versus process-driven dissatisfaction

A customer who is refused a refund because the return window expired may rate the interaction 1 even if the agent was courteous and accurate. Treating that as the agent’s DSAT is unfair, and it hides the real issue, which is the policy.

Many quality teams therefore review each DSAT and tag it by driver: agent behaviour, agent knowledge, process or policy, product fault, system issue, or repeat contact caused by an earlier failure. Over a month, the tags tell a story. If 60% of DSAT is policy-driven, coaching agents will not move the score much; a conversation with the client about the policy might. This tagging is also the raw material for closed-loop feedback programmes, where dissatisfied customers are called back and the underlying issue is fixed.

What CSAT does not measure

CSAT reflects the moment just after the interaction. It says little about whether the issue was actually resolved, or how much effort the customer had to put in across several contacts. That is why it is usually read alongside first-contact resolution and repeat-contact rate, and sometimes alongside customer effort score or Net Promoter Score, which capture different things. A process can post high CSAT from pleasant agents while the customer still needs to call back tomorrow. When CSAT and repeat-contact rate move in opposite directions, the repeat-contact number is usually the one to trust.

Written by the RP Strativa Team. Figures in the examples are illustrative; the margin of error uses the standard normal approximation for a proportion.

Leave a Reply

Trending