Say a merchandising director at a grocery chain decides on a Wednesday afternoon to start a weekend price promotion on Friday instead of Saturday. The change takes a few minutes in the planning tool. By Thursday morning a department supervisor in a store the director has never visited is redrawing the week, and a bakery associate who has worked Tuesday and Wednesday mornings for six weeks is told she now has Wednesday night and Friday evening. Nothing in the promotion's business case has a line for her week.

A randomized experiment at one apparel retailer suggests what steadier weeks are worth. Joan Williams, Susan Lambert and Saravanan Kesavan worked with Gap Inc. across 28 stores in the San Francisco Bay Area and Chicago, assigning 19 stores to a stable scheduling program and 9 to a comparison group between November 2015 and August 2016. The program increased median sales by 7 percent and labor productivity by 5 percent, and program stores brought in $6.20 more revenue per labor hour than comparison stores.1 The program was a bundle: a shift swapping app, a stable shift structure, a core scheduling team, a soft guarantee of 20 or more hours a week for a core group of associates, and extra staffing hours.1 Gap was both a research partner and one of the listed funders.1 Because the stores got the whole package, the design cannot separate which piece did the work, so I read it as evidence that the whole arrangement paid for itself in that setting, and nothing finer than that.

Where the churn started

The part of that report I would hand to a chief operating officer is about the source of the instability. Only 30 percent of the variability in weekly payroll hours was explained by week to week changes in store traffic.1 Store managers named three sources of instability that came from headquarters: inaccurate shipment information, last minute changes to promotions, and visits by corporate leaders.1 I read that as the managers' account of what moved their schedules, since the report gives no count of changes by cause. It still fits any operation with a central planning function. A change at the center takes minutes to make and a week to unwind at the edge, and the people doing the unwinding are the last to be asked.

What a changed shift costs the person working it

A smaller study from a grocery chain shows the cost at the scale of one worker. Guanyi Lu, Rex Du and David Peng analyzed 1,238,418 transactions by 126 cashiers at two stores of a Pacific Northwest grocer across 2016 and 2017, measuring items scanned per second. They put the benefit of working the same hours of the day at about 0.95 percent and of working the same days of the week at about 1.63 percent, and the data are observational, which the authors acknowledge a nonexperimental design cannot fully untangle.2 Among the 48 cashiers hired during the study, the estimates were larger, at 3.39 and 7.93 percent.2 The authors are plain about the limits: one grocery chain, one job, and a two year window focused on short term effects.2

Those are small numbers, and I would not build a staffing model on them. What I take from them is where the cost shows up. A slower scan rate appears on a productivity report, while the schedule change that preceded it appears on nobody's. I wrote earlier this month about what a reorganization spends in familiarity between colleagues, and I will not rerun that argument here. This is the same ledger at the scale of a single person's week.

Two weeks of notice, and then what

Daniel Schneider, David Arbelaez and Kristen Harknett published a study in Science Advances in June that examined Fair Workweek laws in five jurisdictions, using surveys of more than 87,000 hourly workers at 217 large retail and food service firms. The share of workers receiving at least two weeks' notice of their schedules rose 13 percentage points, exposure to a closing shift followed by an opening shift fell 8 points, last minute schedule changes fell, and the authors found no evidence that employers cut wages or benefits in response.3 That is real progress, and most of it is measured where a regulator can see it, in how early the worker hears and how much rest falls between shifts.

In the Gap experiment, both the program stores and the comparison stores already gave two weeks of notice and had eliminated on call shifts, and the 7 percent gain came on top of those practices.1 My read is that notice tells you when the worker learned about a change and says little about whether the change was necessary in the first place. Necessity gets decided in a planning meeting, usually without the worker's schedule on the table.

The cost of a late change is paid by the people who had no vote on it, in a currency the plan does not record.

A count before a survey

An operations leader who wants to test this at home can start with a count. Pick ten locations, pull every change made to a schedule after it was published during one quarter, and tag each change with what caused it, separating the ones forced by customer demand from the ones sent down from the center. Then compare revenue per labor hour at the locations with the fewest late changes against the ones with the most, though the comparison is rough because those locations differ in other ways.

The same test carries to any plan that moves people around by availability, such as a project portfolio staffed to keep everyone fully utilized. Neither study looked at that, so I am extending the reasoning and hold it loosely. The question I would put to the planners is who tallies what a change displaced downstream, since the people who decided it are rarely positioned to see.

The next time a promotion date moves, who will be the first to learn what that change cost, and how long will it take for that number to reach whoever moved the date?