A satisfaction score is only as good as the cut it can support, and most published scores are reported at a level of detail the sample was never built for.
Size against the smallest group you will act on
A single national figure needs far fewer responses than a comparison between twelve branches. If the plan is to move budget between branches, the branch is the unit and the sample has to be built for it. If the plan is only ever to brief the network as a whole, a much smaller sample answers the question.
The failure is invisible in the output. A branch score of 68 based on nine responses looks exactly like one based on four hundred, and the first will move ten points next quarter for no reason at all.
Report the base with the number
Every score published with a sample size and a response rate next to it. Where a cut is too small, the honest options are to suppress it or to label it clearly — not to publish it and rely on a footnote.
This is the discipline that most distinguishes usable research from research-shaped reporting, and it is the first thing dropped when a deck is being tidied.
Who is not in your sample
Response rates in this kind of work are commonly in the low tens of per cent, which means most of your customers are absent from the number. If the non-responders differ systematically from the responders — and dissatisfied customers frequently do not respond at all — the score is optimistic by an unknown margin.
That does not make the number useless. It makes the trend more trustworthy than the level, provided the method holds constant between waves. Trends survive a consistent bias; absolute claims do not.
Driver analysis, honestly
Driver analysis identifies which attributes move overall satisfaction. It requires enough responses and enough variation in the attributes to be meaningful. Run on a small sample with everything correlated, it produces a ranked list that looks authoritative and would reorder itself on a different sample.
Where the sample supports it, it is among the most useful outputs in the report. Where it does not, saying so is more valuable than producing it.
What the score is for
A satisfaction number is a monitoring instrument, not a diagnosis. It tells you something changed. It does not tell you what to do, and organisations that respond to a falling score with a general improvement initiative usually discover next quarter that it did not move.
The diagnosis comes from the verbatims, from the mystery shopping data, and from asking someone why. The score’s job is to tell you when to go and look — and how a programme is scoped decides whether it can.