The Hidden Cost of the Four-Week Rolling Average

Walk into any hospitality operations meeting in the country and you'll hear the same opening line: 'last four weeks averaged 18% labour, so we're holding 18% next week.' It is the most common forecasting method in UK hospitality, the one every spreadsheet defaults to, and the one almost no one stops to question. It is also, on a typical year for a typical 12-site operator, costing the equivalent of a week's labour bill.

Why operators lean on the four-week average

There is a reason the four-week rolling average is so durable. It is a sensible reaction to noise. A single Saturday tells you almost nothing — there was a derby on, or a heatwave, or an England match. Averaging four of them smooths the spikes and gives you something that looks like a number. The trouble is that smoothing is the only thing it does. It assumes that the next four weeks will look statistically like the last four, and in hospitality, they almost never do.

Demand in a multi-site group is shaped by at least three forces that the four-week average has no opinion on at all. The weather changes, sometimes by 10°C inside seven days. The local fixture list shifts — Brighton home games, Saturday racing at Lingfield, a Coldplay residency at Wembley. And the underlying business itself drifts: a new pub opens up the road, a marketing campaign lifts mid-week dinner covers, a refurb closes Sundays. None of this lives in last month's numbers, but all of it lives in next week's.

“The four-week average is a sensible reaction to noise. Smoothing is the only thing it does.”

What it actually misses

The damage from a smoothed forecast is not symmetrical. Overstaffing a quiet Tuesday costs roughly the price of two extra runners on shift. Understaffing a busy Saturday costs you a queue at the door, two negative reviews, three abandoned tables, and — for the operations director — a phone call at 7.42pm that ruins their weekend. In a 12-site group, you only need this to happen at one venue per week to lose every margin gain you found from the other eleven.

The maths is unforgiving. A site doing £35k a week at 28% labour is spending around £9,800 on staff. A 7% forecasting error in either direction — undershooting on a busy weekend, overshooting through a quiet Wednesday — moves £685 a week per site. Across 12 sites and 52 weeks, you are inside touching distance of half a million pounds of operating margin moving on the back of a number nobody actually believes.

Where the four blind spots actually live

What a real forecast does instead

A genuine demand forecast does not start by assuming the past was representative. It starts by separating the signal from the noise — what was the underlying baseline, what was the day-of-week effect, what was driven by external factors like weather or events, and what was random variance. It then projects each component forward based on what is actually happening next week, not what averaged out last month. The result is a forecast that responds when you tell it the local team are at home, when the temperature is set to drop ten degrees on Friday, when school holidays start on the 24th.

The honest endgame

Better forecasting is not a magic wand. The best models in the world will not save you from a leaky front-of-house culture, an unreliable POS feed, or a refurb that nobody on the management team forecast for. But for a 5-to-50 site operator running on the four-week average — quietly, reluctantly, because nothing else has been good enough to switch to — the gap between that smoothed past and a forecast built on real signals is measured in tens of thousands a quarter. It is the cheapest margin point on the floor.