What a satisfaction score can and cannot tell you.

Decide what the number is for

A satisfaction score answers one question: how did this person feel about this interaction, at the moment they were asked. It is not a measure of product quality, of loyalty, or of whether the underlying problem is fixed. Teams get into trouble when a single percentage is asked to stand in for all three.

Write down the decision the score is meant to inform before you start collecting it. Coaching an individual agent, choosing which contact reason to automate, and deciding whether to keep a supplier are three different questions, and they need different cuts of the same data.

Ask at the right moment

Surveys sent the instant a ticket is closed measure the closing message. Surveys sent days later measure whether the outcome held. Both are legitimate; mixing them in one average is not, because you no longer know which you are looking at.

Be careful with conversations that end in a refusal. A correct answer the customer did not want will score badly, and if those cases are a large share of a queue, the average says more about your policy than about your agents.

Know who actually answered

Response rates on support surveys are usually low, and the people who respond are rarely a neutral sample. Customers who were delighted and customers who were angry both have a reason to reply. Everyone in the middle is under-represented, which is why small samples swing so violently.

Record the response rate beside the score, always. A rise from 88% to 92% on forty responses is noise. The same movement on four thousand is a finding.

• Report the score, the response rate and the sample size together.
• Segment by channel, language and contact reason before drawing conclusions.
• Separate first-contact conversations from ones the customer had to chase.
• Exclude automated closes and duplicates from the denominator.

Read the comments, not just the average

The free-text field is the part with diagnostic value. A score tells you something is wrong; the comment tells you whether it was the wait, the answer, the tone or the policy behind it. Those four have entirely different remedies, and only one of them is about the agent.

Tag comments against the same taxonomy your helpdesk uses for contact reasons. When feedback and volume share a vocabulary, you can see whether the reasons customers contact you most are also the ones they are least happy about.

Act on it, or stop collecting it

A survey is a request for the customer’s time, and it carries an implied promise that the answer will be used. Collecting feedback that no one reviews is worse than collecting none, because it trains people that telling you is pointless.

Give the score an owner and a review rhythm. Each cycle, name one change that came from it. If several cycles pass with no change attributable to the feedback, the honest conclusion is that you are measuring the wrong thing, not that customers are hard to please.