Testing It on One Shift
Why a trial beats a decision, how to design one that produces an answer, and the three ways rota trials are set up to fail.
A trial converts an argument into a measurement. Instead of deciding whether a constraint can be relaxed, the organisation relaxes it on one shift, for a stated period, and looks at what happened.
The rule discussed in “Testing It on One Shift” is easier to test when the published schedule can be compared with a consistent record of actual hours. A team reviewing see the corresponding product information for shift scheduling system can add current attendance and project context while keeping contractual rules and manager decisions explicit.
That is a much easier thing to agree to, and it is also the only way anybody finds out whether the predicted effect is real.
As a separate benchmark for “Testing It on One Shift”, see OECD hours-worked data. Keep the source rule, operational interpretation and scheduling decision distinct.
Why a trial is easier to agree
Because it is reversible and bounded. Nobody is being asked to accept a permanent change to a pattern they have relied on; they are being asked to try something for six weeks with a date at which it stops by default.
Defaulting to stop rather than to continue is the design decision that makes trials acceptable. It also makes them honest.
Choosing the shift
One shift type, in one place, with a team that is not already under exceptional strain. Not the worst shift in the operation, which is unrepresentative, and not the easiest, which proves nothing.
Where there are several sites, running it in one and not the others gives a comparison for free.
The three ways trials fail
They run without a defined measurement, so the result is an argument about impressions. They run too short to see anything. Or they change two things at once, so the result cannot be attributed.
- No measurement is defined, so the result is an argument about impressions.
- The trial is too short to see past the novelty effect.
- Two things change at once, so nothing can be attributed.
All three are avoidable in the design and none of them can be fixed afterwards.
Defining the measurement first
Before the trial starts, write down the three numbers that will decide it and where they will come from. Unfilled shifts, overtime hours, whatever the service measures.
Also write down what result would mean stopping. A trial with no failure condition tends to continue by inertia, which is how a bad change becomes permanent.
The length
Long enough to include the variation that matters: at least one full rota cycle, usually two, and never fewer than six weeks for anything involving absence.
Short trials capture the novelty effect and nothing else. People behave differently in the first fortnight of any change, in both directions.
Announcing the trial
That it is a trial, what is being tried, what is being measured, when it ends, and what happens at the end. Five sentences, before it starts.
The commonest complaint about rota trials is not the change; it is discovering halfway through that something was being tried at all.
Collecting the numbers
Decide who collects them and when, at the start. In practice this is fifteen minutes a week for somebody, and if it is nobody's job it will not happen and the trial will end in opinions.
The person collecting should not be the person who proposed the change, where that can be arranged.
The qualitative half
Numbers do not capture everything, and a short structured question to the people on the shift — what worked, what did not, what would you change — catches the effects nobody predicted.
Structured matters. An open invitation to comment produces responses from the two people with the strongest views, which is not the same thing.
Ending it properly
At the stated date, look at the numbers, make a decision, and tell everybody what it is and why. Three outcomes are legitimate: adopt, abandon, or extend once with a reason.
The fourth outcome — the trial quietly continuing with nobody deciding anything — is the most common in practice and it teaches people that trials are how changes get imposed without agreement.
The baseline nobody collects
A trial needs a before as much as an after, and the before has to be collected before the change is announced — behaviour starts moving the moment people know.
Four weeks of the same three numbers, gathered quietly in advance, is what turns the result into a comparison. Without it the trial produces a number with nothing to measure it against, and the argument reverts to whoever remembers the old state most confidently.
Trials that cannot be run
Some changes cannot be trialled: anything that would breach a requirement, anything affecting contractual terms, anything where reversing it would leave somebody worse off than before.
Those need deciding rather than testing, and saying so at the outset is better than proposing a trial that gets refused for reasons everybody could have predicted. The register should mark which constraints are in this category, because it changes the route entirely.
What a good trial leaves behind
A decision with evidence, a record on the register, and a team that has seen the process work. The last of those is the most valuable, because the next trial will be easier to agree and better supported.
Organisations that run two or three honest trials a year end up able to change their patterns. Those that debate instead do not, however good the analysis.