Last year plus a bit
The budget for each site was built last autumn: last year's weekly sales, plus a growth percentage, adjusted for a refurb. It is now wrong for half the sites. One lost trade when a new office block delayed opening. Another is busier because a theatre reopened across the road. The GMs know this and forecast their own weeks from experience, in their heads or a notebook, and rota and order on that basis.
At head office, the budget is still the only formal forecast. So every week the variance report shows the same sites ahead or behind for the same known reasons, and nobody has a better expectation to work to.
Why forecasting stays stuck
- The annual budget is treated as the forecast, even when things have changed.
- Each site's history is in its own EPOS, not in one place a model can learn from.
- Bookings, which tell you much of the week in advance, are not joined to sales history.
- Local events, weather and school holidays are known but never recorded against trade.
- GMs forecast differently from each other, and nobody writes the forecast down.
What rough forecasts cost
Rotas built on a poor forecast overstaff quiet shifts and understaff busy ones. Orders built on it cause waste or run short. The central kitchen produces too much or too little. Head office cannot tell a site with a real problem from a site with an unrealistic budget. And GMs lose faith in the numbers they are judged against.
How we build site forecasts
- We bring together sales history by site, day and daypart from each EPOS into one store.
- Bookings from your reservation system are added, so the forecast knows how many covers are already committed.
- Known drivers are recorded against each site: school holidays, bank holidays, local events, sports fixtures, weather forecasts and changes such as a refurb or a new competitor.
- A forecasting model predicts sales and covers by site, day and daypart for the coming weeks, and learns from how well it did last time.
- GMs see the forecast for their site and can adjust it with a reason, such as a large party or a road closure. Both the model's figure and the adjusted figure are kept.
- The forecast feeds the rota and ordering views, and head office compares forecast, budget and actual in one report.
| Input | Why it matters |
|---|---|
| Trading history | The base pattern by day and daypart |
| Bookings on the system | Committed covers for the week ahead |
| Events and holidays | Known spikes and dips |
| Weather | Terraces, beer gardens, footfall |
| GM adjustments | Local knowledge the data cannot see |
No forecast is perfect, and we show how accurate it has been at each site so people know how much to trust it.
Planning the week with it
On Wednesday each GM sees next week's forecast by day and daypart, checks it against what they know, and adjusts one or two days. The rota and orders are built from that number. Head office sees the whole estate's forecast beside budget, and a site that is permanently behind budget for a known reason can be re-forecast rather than argued about every week.
Does this sound like your planning?
- The annual budget is the only forecast you have.
- GMs forecast from memory and never write it down.
- Bookings are not used to predict the week's sales.
- Weather and local events are never recorded against trade.
- The same budget variance comes up every week for known reasons.