Case study — the 1-minute version
Predicting how busy a restaurant will be, so it schedules the right staff
By Gurjinder Johal, full-stack developer. I built this sales forecast on my own, inside Shiftova, the scheduling app I make for restaurants. It's live and in daily use at a real restaurant.
~41%
lower error on holidays than the franchise's own formula
1,211
real days tested, using only what was known at the time
26% → 22%
smaller miss on the days around a holiday, from a pattern I found
What I did
- Built it. A forecast that learns from a restaurant's own sales history, so the app can suggest how many people to schedule.
- Tested it honestly. I replayed 1,211 real days, using only the sales known before each one, against the franchise's own formula and simple averages.
- Found what the averages hid. The biggest misses clustered around public holidays. Sales jump the day before a mid-week holiday and drop for two days after; Monday long weekends barely move. I built that in without changing any other day.
Compared with the same weekday in the surrounding weeks. Mid-week holidays move sales a lot; Monday long weekends barely do. Averaged together, the effect was hidden.
The honest part
It beats the franchise's formula by about 6% overall and ~41% on holidays. On ordinary days the edge over a simple average is small: most of what's left is random. With one restaurant's data, the holiday adjustment only switches on when the evidence is clear.
Skills used
About me
I'm a full-stack developer with a computer science degree and 10+ years building software and analytics for a restaurant franchise business. I also run Nechako Digital, the studio behind this site. I'm looking for a full-stack, backend or data role.