If you want to know whether a multi-site hospitality group is well-run, you do not look at the management deck. You look at Saturday night. Specifically, you look at one Saturday night, in February, when the weather is doing something unhelpful and there's a derby on. The operating culture, the tech stack, the people decisions — they all express themselves in the same place at 7pm.
5pm: the calls start
Sarah is the operations director at a 12-site managed pub group. She does not need to look at her phone to know that by 5.07pm she will have heard from at least two GMs. The first call is always the same: 'two no-shows, one of them a key holder.' The second is usually 'we're slammed earlier than expected.' Both calls were predictable on Tuesday and now have to be solved in twenty minutes. The site that built its rota off last week's footfall is the site that runs short. The site that bumped headcount on a hunch is the site that costs Sarah margin she cannot get back.
6pm: the WhatsApp chaos
The cover request lands in the staff group, where it competes with two memes, a holiday photo from a chef on annual leave, and a rolling argument about a shift swap from last weekend. By 6.30pm somebody has agreed to come in. By 7pm the GM is on the floor with a clipboard trying to remember what the new starter is allowed to pour. None of this is in payroll. None of this is in the rota. The next time a labour vs sales report runs, there will be a £400 hole nobody can quite explain.
“The site that built its rota off last week's footfall is the site that runs short. The site that bumped headcount on a hunch is the site that costs Sarah margin she cannot get back.”
7pm: the queue at the door
Two sites are flying. Three sites are quiet. One site — the one with the new GM, the one nearest the football ground — is in trouble. The bar is two-deep, food is twenty minutes behind, and a regular has just left a one-star review on the way out. Sarah's phone, which has been quiet since 6.45, lights up with a message that says 'we need help.' The labour cost report on Monday will not show this — the site was, technically, on plan. What the report cannot show is that the plan was wrong.
20 minutes
to solve a four-hour problem
The window an area manager has between 'we're slammed' and the kitchen falling behind. Forecasting is what closes that window before it opens.
8pm: the price of last-minute cover
Last-minute cover costs more than scheduled cover. It is not just the overtime — it is the inefficiency of someone arriving mid-service, the GM time spent finding them, and the morale dent on the team that did show up on time. Across a year, in a 12-site group, last-minute cover is the single most under-tracked source of margin leakage. It does not appear on any dashboard because it has not, historically, been measurable. Almost everybody knows roughly what it costs. Almost nobody can put a number on it.
11pm: the post-mortem
By 11pm Sarah has the same conversation she has every Saturday. 'What went wrong at site seven?' 'Who built that rota?' 'Why did we not see this coming?' The answer, every time, is some combination of 'the forecast was wrong' and 'we did not have anyone on standby for that catchment'. Both are true. Both are also fixable, but not by trying harder, and not by hiring more staff.
A different Saturday: what AI forecasting changes
Imagine the same Saturday with a forecast that knew on Tuesday that site seven would be 22% above its four-week average because of the football fixture and the weather. The auto-built rota carries an extra runner, a second bartender, and a closing-shift KP. The standby list — three people in catchment range who have flagged availability in the staff app — went out on Wednesday. By Friday the GM had confirmed cover before service started. The two sites that are quiet have shed three staff each, who are paid for the hours they were rostered and offered a swap into next week. Site seven is busy, but not breaking. Sarah's phone is quiet at 7pm.
None of this is futuristic. It is what good forecasting plus good rostering plus a usable staff app looks like in a single product. The hard part is not the AI. The hard part is the ten-thousand-decision week being made consistently, across twelve sites, by managers who would rather be on the floor than in a back office. The forecast is the input. The compliant, ready-to-publish rota is the output. What the operator gets back is Saturday night.
See it in your own numbers
Book a 30-minute walkthrough. We'll model what better forecasting would have meant for your last quarter, against your own POS history.