AI Deep Dives

Short-staffed Fridays: how to predict staffing from your till data

By Steve McLean · 2 November 2026 · 6 minute read

It's 7.40 on a Friday. The bar is three deep, two tables are waiting for menus, and someone has rung in sick. Meanwhile on Tuesday afternoon, four people stood around polishing glasses. Most rotas are written from memory and last week's rota. Your till has a far better record of when you're busy. This is how to use it.

In short

Why the rota gets Fridays wrong

A rota written from memory treats "Friday" as one thing. But Friday at 5pm and Friday at 8pm are very different shifts. So are a payday Friday and the Friday after a bank holiday. If you staff the whole evening for an average, you're short at the peak and over at the edges.

Being short costs you twice. Customers wait, order less and leave. And the staff who are there have a hard night, which is how good people end up leaving.

What your till already knows

Almost every till can export its sales as a CSV file. You need the sales by hour, not just the daily total.

In your till exportWhat it tells you
Sales by hourWhen you're actually busy, hour by hour
Transactions or covers by hourHow many people you're serving, which drives workload more than the money does
Day of the weekWhich days follow the same pattern, and which don't
Area or till pointWhether the pressure is on the bar, the floor or the kitchen

Add two things the till can't see: your bookings for the week ahead, and anything unusual coming up (a match on TV, a local event, school holidays).

Build a simple forecast in an hour

  1. Export eight weeks of hourly sales. Enough to see a pattern, recent enough to still be true.
  2. Average each hour of each weekday. All eight Fridays at 5pm, all eight Fridays at 6pm, and so on. Leave out any week that was clearly odd, like a closure or a heatwave.
  3. Adjust for what you know is coming. More bookings than usual, a big event, payday. Write down why you changed a number, so you can check later if you were right.
  4. Turn sales into staff with the sum below.
  5. Put the forecast next to your rota. Look for hours where they disagree.

The sum

Staff needed in an hour = forecast sales that hour ÷ sales one person can comfortably handle in an hour.

The second number is yours to set. Take a few recent shifts that ran well, with no long waits and no one idle, and divide sales by staff hours worked. That's your "sales per staff hour". It will differ for the bar, the floor and the kitchen, so work out each if you can.

Worked example (illustrative numbers)Forecast salesStaff needed (at £150 per staff hour)On the rotaGap
Friday 5pm–6pm£45035+2 over
Friday 6pm–7pm£75055Right
Friday 7pm–8pm£1,05075−2 short
Friday 8pm–9pm£1,20086−2 short
Friday 9pm–10pm£60046+2 over

These are example numbers to show the sum, not an average or a real business. Your own till will give you the real figures.

Look at what the example shows. The total hours on the rota (27) are almost the same as the hours needed (27). The problem isn't how many hours you pay for. It's where they sit. Moving hours from the edges of the evening into the 7pm to 9pm peak fixes most of it, at no extra cost.

What an hour of staff time costs now

Every hour on the rota has a real cost. The current rates:

CostRateFrom
National Living Wage (21 and over)£12.71 an hour1 April 2026
Rate for 18 to 20 year olds£10.85 an hour1 April 2026
Employer National Insurance15% on earnings over £5,000 a yearTax year 2026 to 2027

So an hour at the National Living Wage costs about £14.62 once employer National Insurance is added (£12.71 × 1.15), for someone earning above the threshold. Holiday pay and pension come on top. Two spare people for an hour, twice a night, five nights a week, adds up quickly. But two missing people at 8pm on a Friday can cost more in lost sales and tired staff. The forecast helps you avoid both.

Check it every week

A forecast you never check just becomes a new guess. Each week, put forecast sales next to actual sales for each hour. Where they were far apart, ask why: weather, a big booking, a quiet match. Nory, a firm that sells scheduling software, suggests comparing the last eight weeks of forecast and actual by site and time of day to find where planning needs the most work. That's sound advice whatever tools you use.

If you run more than one site, do it per site. A group average hides the site that's struggling. "Sales per staff hour" and "wages as a share of sales" are fair ways to compare sites of different sizes.

Doing it without the spreadsheet

The method above works. But it takes an hour a week per site, and on a busy week it doesn't get done. Horizon, Lagoona's operations team member, does the reading for you. It reads each site's till, rota, bookings and stock figures every day, from a CSV upload or a daily file sent to its own secure address, and warns you about trouble before it lands, with what to do about it.

It needs a few weeks of till and rota history before it can warn about staffing. Most tills and rota apps can export that on day one. It suggests; your managers decide the rota. Its dashboard, Early warnings, shows the problems it spotted, how far ahead, which site needs you next, and a record of how often its warnings were right.

Horizon is built for groups with several venues or sites, and the method in this post works just as well for one. Horizon is £669 a month per location, running costs included, with no contract.

Try Horizon live

Choose Horizon and ask "Are we overstaffed anywhere?" to see an answer for the demo business. Nothing real is booked, sent or stored.

Try Horizon
Sources