Blog
One branch sells out at eight, another throws food away every night
A chain that orders the same quantities everywhere is paying twice for the same missing information.
Published
Half past eight, a branch in an office building in Sathorn. Three of the best-selling dishes ran out at eight. Staff have to tell arriving customers there is none left. Three in five walk out.
The same night, a branch in a suburban mall twenty kilometres away throws away ingredients for those same dishes, because it ordered to the standard quantity set centrally.
Both branches order from one set of numbers, despite serving different customers, at different hours, with different habits.
The invisible half of the cost
Waste is a countable cost, and usually the only one finance tracks.
Sales lost to running out are not countable, and are usually larger — because it is not only that meal. A customer who fails to get what they came for twice changes their behaviour without telling anyone.
Better forecasting is therefore a revenue question as much as a waste question, and those two pull in opposite directions. Which is exactly why safety stock settings should differ by branch and by item rather than being one number in a policy.
Data that is enough to start
Most chains already have receipt-level sales, which is enough. What adds real accuracy costs nothing, because it is external and freely available.
Weather moves outdoor seating and delivery orders noticeably. Public holidays and long weekends change the shape of the whole week, not just the day. Payday moves the average basket. Local exam calendars and festivals move branches near campuses a great deal.
A model that knows these outperforms a trailing four-week average clearly — most of all in unusual weeks, which are precisely the weeks an average misses worst.
What gets these systems abandoned
Forecasting at the wrong level. Predicting a branch's total sales does not help a branch manager order, because ordering happens by ingredient. The forecast has to reach the level where the decision is made.
No override. The manager knows things the data does not: an event beside the mall next week, the road outside closing for repairs. A system that will not let them adjust reads as head office not understanding the floor, and gets worked around.
No reason shown. A number that simply appears earns no trust. A number with a one-line explanation — higher than usual because it is a long weekend and rain is forecast — lets people decide alongside the system.
Where to start
Ten branches, the twenty best-selling items, and two months running in parallel with the existing method, changing nothing about how ordering is done.
Compare what the forecast would have produced against what actually happened. Running in parallel carries no risk and produces enough evidence to decide whether to expand or stop.
What GIPSIC does here
We build systems that join data from several sources and put it on a screen someone on the floor will actually use — including systems that have to work on a tablet, with messy hands, in a few seconds.
What we ask first is what your point-of-sale can export and how often, because that answer shapes what is possible far more than the choice of model does.
If you would like to talk about your chain, get in touch.
About the author
Film — Wisit. A businessman who still does his own BA work more often than he probably should, and writes a fair bit of code, front and back. Runs two or three small businesses. Follows technology and business obsessively, in Thailand and everywhere else. Off the clock: physics, astronomy, DIY, and anything to do with networks. Music always on, though he cannot sing. Plays instruments anyway, badly. Plays a lot of sport, racket sports above all. Not much of a traveller by himself, but happy to take Mint anywhere in the world. A man who fears — sorry, loves — his wife. One flaw: he barely touches video games.
Written with Claude Opus 5