Who Quits When Schedules Get Stable?
The most operationally useful secondary result in the Gap experiment is about the experience level of the people who leave, not the headline turnover rate.
Reviewed against primary sources on July 19, 2026 by the Soon operations research team
The evidence in one line
In the Gap stable scheduling experiment, a practitioner report states that average tenure of employees who quit treatment stores fell 10.8 months against a 25.5-month base, while overall retention was unchanged (Williams et al., 2018). Read plainly, stable scheduling changed who left rather than how many left: the people walking out had less experience, and experienced staff stayed. That report was not peer reviewed, so treat the tenure figures as a mechanism worth testing rather than a settled effect.
The retention rate hid the actual result
Most retention reporting collapses a workforce into a single number: the share of people who stayed. In the Gap experiment that number did not move, and a manager reading only that line would conclude the scheduling changes did nothing for staffing. The practitioner report points at a different column. Average tenure of employees who quit treatment stores fell 10.8 months against a 25.5-month base, while overall retention was unchanged (Williams et al., 2018).
A drop in the average tenure of leavers, with the same number of leavers, has one plain reading. The composition of the exit queue shifted toward newer employees, which means fewer experienced employees were in it. Turnover was not eliminated, it was redistributed onto the part of the workforce that is fastest to replace and least costly to lose. For an operations manager that is a different fact from a lower turnover rate, and arguably a more useful one, because losing a long-tenured employee and losing a recent hire are not the same event on the floor.
This is the reframe worth carrying into your own reporting. A headline retention percentage treats every departure as an identical unit, so it is structurally incapable of showing you a change in who is leaving. If the mechanism described in the Gap report is real, it is invisible to the metric most operators watch.
Grade this as a mechanism, not as a headline finding
The tenure figures come from the practitioner report on the Gap experiment (Williams et al., 2018), which was not peer reviewed. They have not been separately verified against the tables in the peer-reviewed paper that reported the experiment's main results. That places them a clear rung below the headline finding from the same study, a 5.1% productivity increase (Kesavan et al., 2022), which passed peer review and is stated in the published record. Same experiment, two different tiers of evidence.
The practical rule that follows is to keep the source attached whenever the tenure numbers travel. It is accurate to say the practitioner report describes a shift in who left. It is not accurate to present a 10.8-month drop as an established effect of stable scheduling. That distinction stops being a technicality the moment you ask a finance team to fund a rollout on the strength of it.
There is also a reason to expect secondary results to be softer in general. Headline outcomes are the ones a study is powered and pre-specified to detect, while a compositional finding like this one is the kind of detail that surfaces during analysis and gets reported in a practitioner write-up. That does not make it wrong. It makes it a hypothesis with unusually good provenance rather than a measured effect you can quote as settled.
The sales mix points the same direction, which is suggestive and no more
One further detail from the practitioner report belongs next to the tenure finding. The sales lift in the experiment came from better conversion and higher basket values rather than more foot traffic (Williams et al., 2018). Traffic is largely outside a store's control. Conversion and basket size are what the people on the floor do with the traffic that arrives.
That mix is consistent with more experienced staff being present, which is what the tenure finding would predict. Consistency is not confirmation. Two results from the same report that agree with each other are still two results from the same report, and a tidy story that fits the available data is exactly the kind of thing that later fails to hold up. Treat the alignment as a reason to take the mechanism seriously enough to test where you work, not as evidence that it has been established.
How to check this in your own operation
Tenure at exit is not a hard metric to build, and most scheduling and payroll systems already hold the data. For every departure in a period, record how long that person had been employed, then track the average alongside the headline turnover rate. Break it out by site, because a chain-level average will smooth away exactly the variation you are trying to see.
Run together, the two numbers say more than either does alone. A flat turnover rate with rising average tenure at exit means you are losing more of your experienced people, which is bad news the headline rate will never show you. A flat turnover rate with falling average tenure at exit is the pattern the Gap practitioner report describes, and it is worth detecting before you write off a scheduling pilot as having produced nothing.
What this means for your schedule
- Track average tenure at exit alongside your turnover rate, because a flat rate can conceal a large shift in who is leaving.
- Break the tenure-at-exit number out by site, since a chain-level average hides the locations where experienced staff are walking.
- Attach the source every time the 10.8-month drop against a 25.5-month base travels, naming it as a finding from the Gap practitioner report (Williams et al., 2018).
- Judge a scheduling pilot on the composition of departures, not only on whether the headline retention number moved.
- Read the tenure-at-exit trend next to your sales mix, since conversion and basket size are where the presence of experienced staff tends to show up first.
The business case
The retention question executives usually ask, did turnover fall, may be the wrong question. In the Gap experiment the practitioner report describes overall retention as unchanged while average tenure of employees who quit treatment stores fell 10.8 months against a 25.5-month base.
Replacement cost and lost capability concentrate in experienced staff, so a steady headline rate paired with a less experienced exit queue is a better staffing outcome than the rate alone suggests.
Weight it accordingly: the tenure figures come from the practitioner report (Williams et al., 2018), so present them as a mechanism to test rather than as a forecast of what a rollout will deliver.
Frequently asked questions
- Did stable scheduling reduce turnover in the Gap experiment?
- Not according to the practitioner report on the experiment, which describes overall retention as unchanged. What shifted was the makeup of the people leaving: average tenure of employees who quit treatment stores fell 10.8 months against a 25.5-month base (Williams et al., 2018). Fewer experienced employees were in the exit queue even though the queue stayed about the same length.
- How much weight should the Gap tenure finding carry?
- Less than the peer-reviewed results from the same experiment. The 10.8-month drop in average tenure at exit, against a 25.5-month base, comes from a practitioner report on the Gap experiment (Williams et al., 2018) rather than from the published tables. Use it as a mechanism worth testing in your own operation, and name the source whenever you quote the figure.
- Why does the experience level of leavers matter more than the count?
- Because a departure is not a fixed unit of damage. Losing a long-tenured employee removes product knowledge, customer familiarity, and the ability to run a shift without close supervision, none of which a new hire replaces quickly. That is why the Gap practitioner report pattern, average tenure of employees who quit treatment stores falling 10.8 months against a 25.5-month base with overall retention unchanged (Williams et al., 2018), reads as a better staffing outcome than the flat headline rate suggests.
- Does anything else in the Gap experiment support the shift in who quits?
- One further detail points the same direction. The practitioner report on the Gap experiment records average tenure of employees who quit treatment stores falling 10.8 months against a 25.5-month base, and it also attributes the sales lift to better conversion and higher basket values rather than more foot traffic (Williams et al., 2018). Conversion and basket size are what staff on the floor do with the traffic that arrives, so the two findings fit together. Both come from the same report, so the agreement is not independent confirmation, and it is a reason to test the mechanism where you work rather than to treat it as settled.
Sources
Every figure on this page is drawn from a cited primary source and checked against the original publication.
Williams, J. C., Lambert, S. J., Kesavan, S., Fugiel, P. J., et al. (2018). Stable Scheduling Increases Productivity and Sales: The Stable Scheduling Study. [Report]. Center for WorkLife Law, UC Hastings College of the Law. https://worklifelaw.org/publications/Stable-Scheduling-Study-Report.pdf
Practitioner report on the same randomized experiment (not peer reviewed)
Kesavan, S., Lambert, S. J., Williams, J. C., & Pendem, P. K. (2022). Doing Well by Doing Good: Improving Retail Store Performance with Responsible Scheduling Practices at the Gap, Inc. Management Science, 68(11), 7818โ7836. https://doi.org/10.1287/mnsc.2021.4291
Randomized controlled field experiment (28 stores, roughly 150,000 shifts, about 1,500 employees)
None of the studies cited here evaluated Soon.They examine scheduling practices, shift patterns, and working hours as studied by independent researchers, so their findings describe what those practices are associated with, not what any particular software produces.
This article summarizes published research for scheduling and operations decisions. It is not medical advice. Individual health questions belong with a qualified clinician.
Keep reading
What Happened When a Retailer Randomized Stable Scheduling?
A randomized controlled field experiment in scheduling research: what it measured, what it earned, and where it stops.
Read →How Do You Judge a Scheduling Study?
A short field guide to reading scheduling research, worked through this library's own sources, including the claims we checked and cut.
Read →How steeply does hardship rise with schedule unpredictability?
As schedule unpredictability climbs, predicted material hardship climbs with it across four separate domains, and it does so after pay is accounted for.
Read →