Client study · Customer experience · Mystery shopping
What predicts customer satisfaction?
A mystery-shopping programme tracked service quality over six waves in 2025 and 2026. Which parts of a visit move together with overall satisfaction, and do those patterns hold across channels, waves and times of day?
The programme
Repeated measurement, then analysis.
Client work, anonymised. The analysis covers three of the six waves.
My contribution
Consulting support in instrument development, data collection and management, analysis and reporting.
What the work covered
Journey stages, staff attitudes and behaviours, presentation and overall satisfaction, compared across waves and service channels. An earlier report on 102 visits found weaker execution at the initial welcome and farewell, and stronger explanation and solution discovery.
Drivers
What looked important alone, and together.
Pearson correlations first, then a multiple regression with all eight stages entered at once (741 observations, satisfaction rated 0–4). Importance is the LMG share of explained variance, with 1,000 bootstrap resamples as a robustness check.
Eight stages of the visit
Association with overall satisfaction, alone and combined
On its own: correlation with overall satisfaction (r)
Together: share of the model’s explained variance
Attitudes dominate
Staff attitudes had the strongest correlation (r = .83) and kept a large independent effect with the other stages held constant.
Overlap shows up
Staff behaviours correlated strongly (r = .63) but had no independent effect once attitudes were in the model. The two overlap.
Check-out turns negative
With the other stages held constant, check-out was negatively associated with satisfaction. This may reflect overlap with other stages, and it needs a closer look before it is read as an effect.
Priorities
Performance against importance.
Each stage is placed by its average score and by its share of the explained variance. The dashed lines are the averages across the eight stages.
Performance × importance
Where the stages sit
Protect and reinforce
Staff attitudes (30.4%) and explaining and answering questions (17.2%): the most important stages, performing around or above the average.
Room to improve
Understanding the visitor (12.6%): moderately important, with below-average performance.
Monitor
Welcome (9.0%) and check-out (8.0%): the lowest average scores, with a moderate share of explained variance.
Consolidated
Closing the discovery, staff behaviours and presentation and facilities (2.1%): performing well, with a smaller share of explained variance.
Welcome and check-out had the lowest average scores, but a moderate share of explained variance. Presentation and facilities scored well and added little, so it works as a baseline standard rather than a driver of satisfaction.
Attitudes
What staff attitudes are made of.
The composite was split into seven items and the model re-estimated. Explained variance stayed at R² = .71, so the composite was not hiding extra information.
Seven attitude items
Share of explained variance by item
The attitude effect was driven mainly by whether staff were available and effective, more than by the softer traits.
Stability
Did the pattern hold?
Interaction tests asked whether each stage’s association with satisfaction differed by channel, wave or time of day. The evidence was selective.
Across channels
The same drivers mattered in all three channels. The welcome effect was weaker in the partner channel, with no other clear channel differences.
Across waves
Staff attitudes were the strongest predictor in every wave. The association with explaining and answering questions was weak in the first wave and stronger in the later two.
Check-out over time
The negative association at check-out in the first wave was smaller in the two later waves. Closing the discovery stayed stable throughout.
Time of day
Satisfaction tended to be lower in the afternoon and evening (a marginal result), but the drivers did not change. Explaining and answering mattered more in the evening, and closing mattered more in the afternoon.
Limits
What the analysis does not show.
Stated up front, so the drivers are read as patterns, not as levers.
- These are associations in observed visits. The analysis does not show that improving a stage would raise satisfaction.
- The stages overlap strongly, so one coefficient read alone, such as the negative check-out one, can mislead. The LMG shares split the overlap but do not make it causal.
- Only three of the six waves are analysed here, so the later 2026 waves are not covered, and some of the channel, wave and time-of-day differences were marginal.
Where it sits in the method: Stage 02 · Build & validate · Stage 03 · Collect the data · Stage 04 · Analyse · Stage 05 · Deliver