Shift and site comparisons are the most requested analysis and the most frequently wrong.
What differs before anything else does
Work mix. Night shifts often handle replenishment; day shifts handle picking. Not comparable.
Volume, which affects efficiency in both directions.
Staffing level and experience, which is the largest single factor.
Equipment availability.
Congestion, which varies with how many people are on the floor.
Interruptions: deliveries, visitors, maintenance, all concentrated on days.
Normalise for these or the comparison measures the differences rather than the performance.
Doing it properly
Compare like work. Filter to a task type both shifts perform in volume.
Normalise per unit of work content, not per hour.
Account for staffing, as output per available hour rather than per shift.
Use a long period, several weeks, so ordinary variation does not dominate.
Check the spread as well as the average. A shift with the same mean and much wider spread has a different problem.
Compare each shift against itself over time, which is the more informative analysis and avoids the fairness problem entirely.
What real differences usually are
When a difference survives normalisation, the cause is usually structural.
Supervision quality and presence.
Experience distribution, where new starters concentrate on one shift.
Method differences that have drifted apart, which are worth capturing and standardising.
Equipment allocation.
Physical conditions: lighting, temperature, noise, which differ between day and night more than people expect.
Support availability. A night shift with no maintenance cover waits longer for every failure.
None of these is fixed by telling the slower shift to try harder.
Site comparisons
Harder still, and usually not worth attempting as a ranking.
Layout, building, mix and volume differ so much that a normalised comparison is close to impossible.
Compare practices rather than numbers. What does the better-performing site do differently, described concretely.
Compare each site against its own trend.
Use comparison to find things to copy, not to rank. Ranking sites produces defensive reporting and hidden problems.
Presenting a comparison
State what was normalised for and what was not.
Show the spread, not just the means.
Show whether the difference exceeds ordinary variation, which most do not.
Lead with the explanation, not the ranking. "The night shift handles a different mix" is more useful than a league table.
Do not name individuals, which converts an operational analysis into a personnel matter and stops the cooperation the analysis depends on.
When to refuse
A ranking request with no normalisation possible should be declined, with the reason.
Producing an unfair comparison because it was asked for damages the credibility of every subsequent analysis, and the person who requested it will quote the number without the caveats.
Comparing a shift against itself
The analysis that avoids the fairness problem entirely and is usually more informative.
Plot each shift's own series over time, with its normal range.
Look for changes within the shift, which are attributable to something that happened.
Compare the trend, not the level, across shifts.
A shift improving faster than another is a finding about practice rather than about people, and the practice can be copied.
This removes the mix and staffing normalisation problem completely, because each shift is its own control.
When to decline a ranking
Some comparison requests cannot be made fair, and producing them anyway is worse than refusing.
Where the mix differs fundamentally and cannot be normalised.
Where the group is small enough that individuals are identifiable.
Where the requester has already stated the conclusion.
Decline with the reason, and offer the alternative: each unit against its own trend, or a comparison of practices rather than numbers.
A number produced under pressure will be quoted without its caveats, and the credibility lost is not recovered on the next analysis.