Most operations are not understaffed. They are understaffed between about 12 and 3, or during the checkout rush, or on the twilight shift, and adequately staffed the rest of the time - which averages out to looking fine.
The averaging is the problem, because customers do not arrive on average.
What the research actually measured
Vidya Mani, Saravanan Kesavan and Jayashankar Swaminathan took hourly data on traffic, sales and labour from 41 stores of one retail chain and modelled what the staffing level should have been each hour, using an approach borrowed from queueing theory.
All 41 stores were systematically understaffed during a three-hour daily peak. Not some of them, and not occasionally. Every store, every day, in the same shape.
The two causes they isolated are the interesting part, because both are fixable:
- Forecasting error - the demand curve used to build the rota did not match the demand that turned up
- Scheduling constraints - even where the forecast was right, the shift patterns available could not put people on the floor in the shape the day needed
The second one gets far less attention than it deserves. You can forecast perfectly and still be short at 1pm, because you only have eight-hour shifts and nobody wants to start at eleven.
Why nobody notices
Understaffing is quiet. Overstaffing is loud - it shows up as a labour percentage that someone asks you about on Monday morning.
Understaffing shows up as:
- A queue that people leave
- A room not ready at 3pm
- A table that turns once instead of twice
- A call that rings out
- A shift where everyone worked flat out and the numbers still looked mediocre
None of those arrive as a number with your name on it. The cost is entirely in transactions that did not happen, and things that did not happen do not appear in any report you already run.
Finding your own three hours
You almost certainly have the data. You are just not looking at it by hour.
Take four weeks and plot two lines against each other, per hour, per day of week:
- Demand - transactions, covers seated, check-ins, calls answered, rooms due out. Whichever one drives your work.
- Labour hours actually on the floor - worked, not scheduled. Use clock-in data if you have it, because the rota is a plan and the plan is not what happened.
Then look at the ratio, not the levels. You are looking for the window each day where demand per labour hour spikes well above the daily average. That window is where your money is going.
Two things usually surprise people the first time:
The peak is narrower than the shift patterns. It is often 90 minutes to three hours, and the rota is built in blocks of eight. That mismatch is the scheduling constraint the researchers measured.
The peak is not where the rota assumes. Lunch trade that moved half an hour later during the pandemic and never moved back. A checkout curve that is now driven by school runs. A breakfast service that shifted with the guest mix.
What to do about it in each sector
Retail - the peak is footfall, and the cost is conversion. A short mid-shift over the peak beats another full shift, and it is cheaper. Also check that till coverage and the shop floor are staffed against different curves, because they are.
Hotels - you have two, and they collide. Housekeeping is compressed between the last checkout and the first check-in, and reception peaks at check-in and at breakfast. Housekeeping cannot flex much once rooms are assigned, so the useful lever is assigning against the actual departure list rather than a flat rooms-per-attendant figure.
Bars and restaurants - the peak is the second sitting, and the constraint is usually the kitchen rather than the floor. Staff the pass against covers per fifteen minutes, not per service.
The cheapest fix is not more people
It is shift shapes that match the curve. A four-hour shift starting at eleven is not a worse job than an eight-hour one starting at nine - for a lot of people it is a considerably better one, and it is the shift that actually covers the peak.
Second cheapest is moving the work that does not have to happen at the peak. Deliveries, cleaning, stock, admin, handover. Every hour of non-customer work that sits inside the peak window is an hour of staffing you already paid for and cannot use.
Neither of those costs a single extra hour on the wage bill. They are just the rota having the same shape as the day.