The average day is not the day that hurts
At the defaults there are 24 pool desks and 39 members competing for them, and each member comes in 55 percent of open days. The average day brings 21.5 people to 24 desks, which sounds like two and a half desks of slack. The pool still runs out on 16.3 percent of days — about three and a half days a month — because attendance is a distribution and the top tail of it is well above the mean.
The spread is what does it. With 39 members at 55 percent the standard deviation is 3.1 people, so a day two deviations above the mean brings 28 people, and days like that are not rare. This is the reason a sharing ratio cannot be set from a headcount and an occupancy percentage on the back of an envelope. The mean tells you the rent is covered; the tail tells you whether anybody has anywhere to sit.
The ratio is a bet that attendance is uncorrelated
The binomial says every member flips their own coin. Nothing in a real workspace works that way. Teams come in together. Tuesday and Wednesday are heavy and Friday is empty in most places that have measured it. A weather day empties the whole floor at once and a company all-hands fills it at once. Every one of those effects makes the true distribution wider than the binomial, and wider distributions breach a fixed desk count more often at the same mean.
That is why the peak weekday matters and the weekly average does not. At the defaults, lifting attendance 35 percent for the busiest weekday puts the expected turnout above the desk count outright, so that day is short more often than not, while the weekly average still looks like it has slack. The day the correlation bites is the day the model fails, and the model gives no warning first — it is fine for months and then a Tuesday goes wrong.
Service level is expensive at the top end
| Days you accept running out | What it buys |
|---|---|
| 1 in 5 | The most members per desk, and a reputation you will hear about |
| 1 in 20 | About one bad day a month; the setting most operators actually run |
| 1 in 100 | Costs a meaningful slice of the member roll to reach |
| Never | One desk per member, which is not hot desking |
Move the target field and watch the member limit fall. The curve is steep at the safe end, because pushing the breach probability down means chasing a tail that is thinning fast, and every step costs more members than the last. Deciding where on that curve to sit is a pricing and reputation judgement, not an arithmetic one, but the arithmetic tells you what each step costs.
What to do with the answer
Two levers move the picture and only one of them is the desk count. Adding desks is capital and rent; the load factor calculator puts a monthly number on each one and the desk layout calculator says whether the floor will take them. The other lever is flattening the week — pricing, scheduling or simply asking teams to spread out reduces the concentration figure, and a smaller concentration figure raises the member limit without touching the rent.
The same distribution turns up wherever a resource is deliberately oversold. No-shows and overbooking runs the identical binomial over an appointment book, where the bet is that people booked will not arrive rather than that people entitled will not come in. If your bottleneck is a queue for a shared machine rather than a shortage of desks, the queue model is the right shape instead, because there the arrivals wait rather than leave.
Questions people ask
What hot desk ratio should I use?
There is no general answer and anyone who gives you one is quoting a number from a different building. The ratio that works is entirely a function of your own attendance rate and how concentrated it is across the week, both of which you can measure from door data in a month. What this page can tell you is the ratio your desks support at the service level you choose. At the defaults that is about 1.46 members per pool desk for a one-in-twenty target, and it would be a different number at a different attendance rate.
Where do I get the attendance rate?
Door swipes, the booking system or the wifi association log over at least four consecutive weeks, divided by open days times members. Do not use a survey. People consistently report coming in more than they do, and because the answer is roughly proportional to this input, a five point overstatement moves the whole result. If you have weekday detail, use it twice: the overall rate goes in the attendance field and the gap between the busiest weekday and the average goes in the concentration field.
Why does the calculator show a chance of running out when the average day has spare desks?
Because attendance varies day to day and the variation is not small. With 39 members at 55 percent the day-to-day standard deviation is about three people, so days four, five and six above average happen regularly. The probability shown is the exact binomial sum over every outcome that exceeds the desk count, not the average plus a fudge. Slack on the average day and a comfortable pool are different things, and confusing them is the most common error in this arithmetic.
Does this account for people coming in together?
Not properly, and this is the honest limit of the model. The binomial assumes independence, and real attendance correlates through teams, meeting days, weather and holidays. Correlation widens the distribution, so the true chance of running out is higher than the binomial says, sometimes considerably. The busiest weekday input is a blunt correction for the largest and most predictable piece of that correlation. Treat the average day figure as a floor on how often you will be short, not an estimate of it.
Can I use this to decide how many people to let into a space?
No. It is a planning tool for desks and memberships, and it says nothing about how many people may lawfully occupy a space. Occupant load, egress and everything connected to them are set by the code your building department enforces and by the terms of your lease, and this page supplies none of those numbers and no rule about them. Whether you may license desks in space you lease is separately a matter for your lease, your landlord and a lawyer.