These scores are lagged attitude snapshots and cannot prove a cause

What it is

Recommendation and satisfaction scores record a stated attitude, at the moment of the survey, about an experience that has already happened. They are a lagged attitudinal snapshot: the event comes first, the evaluation later, with memory, after-the-fact information, and willingness to answer in between. A later rise or fall shows only that people who were asked later made a different statement. It does not show that a release, a campaign, or a feature caused the attitude to move. Without a contrast and without assignment of the treatment in time, the missing link is between the wiggle and a cause.

Why it happens

Respondents retrieve a compressed memory, not the behavioral trace from the moment. A support clash, a news story, or a bill can rewrite the retrospective while the product sits still. The survey often appears after the journey ends, missing hesitation and abandonment along the way. Who stops to rate is already a filter. Aligning the curve with a release calendar, the eye claims the nearest change as the cause—a story from temporal proximity, not an identification strategy. A stated “would recommend” is also not later recommending: saying it and doing it are different acts. A snapshot can be a weather record. It cannot be a construction log.

Studying it

To claim that a treatment moved the score, the treatment has to be separated from concurrent season, marketing, survey rules, and sample composition: a contemporaneous control, a staggered rollout, or at least a pre-registered interrupted series, plus a check that the people eligible to answer stayed comparable. Place the score next to behavior in the same window (completion, repurchase, complaints, actual sharing) and see whether they move together; on a fork, prefer measurement or sample over product causation. Cross-correlation and “the first peak after launch” do not identify. Split descriptive trend and causal claim into two sections of the report; one line chart may not carry both languages.

Where it stops holding

As a monitoring alarm, a snapshot can say “the statement changed, go look.” An alarm is not an attribution. A post-task rating in the lab shortens the lag and still covers only people who stayed; it also does not make the lab task into market causation. If the treatment is a change to when or how the survey is asked, the movement measures the measurement process and is even less a product effect. Long brand trackers can describe slow drift; the cause set for drift is usually larger than a single release.

Applying it

• Score movement triggers an investigation list, not an external story that “X shipped, therefore this.”
• A causal sentence needs a designed contrast; with only a company-wide curve, write the conclusion as description.
• Draw survey rules, campaign calendar, and releases on one timeline; rule out measurement and activity before talking product.
• Check: remove the release-date labels and ask a colleague who was not there what caused the wiggle. If they cannot recover the official cause, keep that cause out of the release note.

Related

• Same group: Q3.17.1 Cultures occupy the same scale differently · Q3.17.2 An opaque industry benchmark is not a comparison · Q3.17.4 Intercepting at a high-score moment inflates the number
• Adjacent: Q3.03 Net Promoter Score · Q3.18 Experimental design and controls
• Search terms: lagged snapshot · attitudinal metric · causal attribution