Where to Start

Scheduling improvement usually starts with a product demonstration. The order that works starts with two numbers.

The two numbers first

Notice actually achieved, averaged over the last three months, by site.

Employer-initiated changes per published shift, same period.

Both come from records you already have.

They are also the two numbers everything else improves, so they are the baseline.

Weeks one to four

Fix the availability data. A two-minute conversation with everyone, recording hard constraints separately from preferences, with a date.

Write down the capability requirement per shift type, and the matrix of who holds what.

Count the single points of failure.

No software required for any of it.

Weeks five to eight

Move publication earlier by one week, whatever the current figure is.

Measure what breaks. Usually less than expected.

Add a buffer of one person on the most variable shifts, sized from measured forecast error.

Count the changes that disappear as a result.

Weeks nine to twelve

Write the allocation rule for undesirable slots, and produce the current distribution.

Show it to the team, which is uncomfortable once and useful permanently.

Set up the rest constraint as a hard check before publication.

Then review whether a product is needed, which by this point is a much better informed question.

What to avoid

Buying before the availability data is fixed, because a product fed stale constraints produces the same rota faster.

Announcing a new approach before the first improvement is visible.

Starting at the easiest site, which tells you nothing.

Optimising cost before the constraints are in, which is the note on generated rotas.

What a quarter should produce

Notice improved by a week.

Employer changes down measurably.

Availability current and capability documented.

A stated allocation rule and a published distribution.

And a decision about software made on evidence rather than on a demonstration.

Fix availability before buying

The sequencing error that wastes a purchase.

A product fed stale constraints produces the same bad rota faster.

A two-minute conversation with everyone, recording hard constraints separately from preferences, with a date.

Then the software question is a real one, and the trial will tell you something rather than reproducing your existing data problem at speed.

Turn the principle into a test

For a concrete product reference, the official website can help turn the principle above into a test. Verify the current behaviour in a trial and judge the record it creates, not the wording on the page.