Guide
Part-time holiday entitlement: pro-rating that survives a tribunal
Updated
Pro-rating annual leave is one multiplication, and it still goes wrong regularly, because the number being pro-rated is often the wrong one to start with. Almost every error in this area is a full-time figure that already had bank holidays baked into it.
The calculation
Acas states the rule plainly: "someone who works part time is still entitled to 5.6 weeks' statutory paid holiday" and "the entitlement will be in proportion to the hours they work". Its worked example gives a three day week 16.8 days of paid holiday a year (3 x 5.6).
gov.uk gives the same example and the same arithmetic. Days worked in a normal week times 5.6, and nothing else enters the sum.
The error that survives audits
The common failure is to pro-rate 20 days rather than 28, because 20 was the number left after the employer's bank holidays were deducted from a full-time entitlement.
Pro-rating 20 days for a three day week gives 12 days. The statutory floor is 16.8. The gap is nearly five days a year, per worker, and it survives because the resulting number looks plausible.
The test is to pro-rate the statutory entitlement first, then deal with bank holidays. Doing it in the other order compounds a policy choice into a legal shortfall.
Weeks, not days, when the pattern is unusual
For a worker whose days vary in length, days are a poor unit. The entitlement is 5.6 weeks, and a week means a week of that worker's normal hours.
Where a pattern makes days-per-week meaningless, the safer route is to convert the entitlement to hours and track it in hours, which is what the law already requires for irregular hours workers.
The cap does not pro-rate
The 28 day ceiling is a ceiling on statutory entitlement, not a full-time figure that gets scaled down. A worker on a five day week reaches it exactly; a worker on six days does not get 33.6 days.
That means the cap has no effect at all on anybody working five days a week or fewer, which is almost all part-time staff. If a part-time calculation has invoked the cap, the calculation is wrong.