At Yabune, we processed six months of measurement data from the building management system of a 254-room hotel in Budapest: 185 nights, from February to August 2026. The question was how much energy goes into rooms where nobody sleeps, and how much of it could be recovered.
In the end, the answer was a number. The more instructive part, though, was the moment we realised the analysis had almost started from a false premise. We presented the first results at the Smarthome Beach 2026 conference; this is the full analysis.
The trap: a keycard sensor is not an occupancy sensor
Every room has a keycard holder, and the system already records its state – a convenient, free proxy for “occupancy”. It seems obvious: if there is a card in the holder, someone is in the room.
The first run showed 91–98% occupancy. For a city hotel, that is not a credible six-month average – and if it were true, there would be nothing to analyse, because there would be hardly any spare capacity. The explanation is simple: during the day, housekeeping staff insert their cards too. A fifteen-minute room clean produces the same signal as a three-day stay, and in the daily totals the two are indistinguishable.
The fix was to restrict the signal to a night-time window: 00:00 to 06:00, because nobody cleans rooms at night. The result is 73–86% occupancy – a realistic range – and, even more convincingly, the distribution of night-time card presence is clearly bimodal. A room either signals for almost the entire window or not at all; intermediate values are rare. That is exactly the signal shape we expect from a genuine occupancy signal – and exactly what the daytime data smears out.
The first lesson, and the one that transfers best to other projects, is that validating proxy signals is not a formality. Having data is not enough: you also need to know whether the data really describes what you think it does. A sensor that “almost” measures what you want is worse than no sensor at all, because it gives you a confident, round, wrong result.
How many rooms are empty, and where?
With the validated signal, the picture looks like this:
- On average, 51 rooms – 20.1% of the room inventory – stood unsold every night. The maximum was 140, the minimum 7.
- Monday is the emptiest night (84 empty rooms on average), Sunday the fullest (28). A classic urban business-hotel pattern.
- On average, the empty rooms were scattered across 13.4 of the 18 wings.
- On none of the 185 nights was an entire floor empty. Completely empty wings were rare, and even then only among the smallest units of 4–8 rooms.
At first glance, that settles it: there is nothing to shut down, because there is no contiguous empty unit. But the volume is there. 36% of the time, there were enough empty rooms on at least three consecutive nights to fill an entire floor. There were enough for two floors on four occasions, 16 nights in total.
So the capacity exists – just in the wrong shape. It isn't missing; it's scattered.
What does a room use when nobody is sleeping in it?
Measured daily average power: an occupied room 78.6 W, an unsold room 51.8 W. Breakdown for the unsold room: TV 12.0 W, fan-coil fan 11.5 W, towel warmer 10.2 W, minibar fridge 8.0 W, lighting 8.7 W, sockets 1.9 W.
Roughly 63% of an empty room's consumption comes from devices nobody is using. That alone is food for thought, but it is not the real issue.
The key measurement: the cooling doesn't know about occupancy
We also measured the fan coils' actual cooling run time – not the fan's consumption, but how long each unit was actively cooling.
- In July, the fan coil in an unsold room was cooling 47.9% of the time – in an occupied room, 47.4%.
- In August: 54.2% versus 58.5%.
In summer, the system makes practically no distinction between let and empty rooms. Rooms that stay empty for days are kept at full comfort level.
A counterexample shows that this is not a hardware problem: for the towel warmers, the logic works (7.2% run time in an empty room, 15.1% in an occupied one). So the system is capable of it – the occupancy information reaches the controls – it simply was never configured for cooling. The fix is a settings change, not an investment.
The second lesson is that the biggest waste is rarely where you are looking. A TV's standby consumption is visible and easy to talk about. An air-conditioned empty room is an order of magnitude bigger, but it doesn't show up on the room's energy meter, because the work is done by the plant in the machine room.
How much could be saved?
Over the six months, the energy spent on unsold rooms came to roughly 24,600 kWh in total: 11,700 kWh of measured in-room electricity plus an estimated 12,900 kWh of HVAC plant load.
If, with the same number of rooms sold, bookings had been grouped into whole wings and the freed-up units had been put into a standby mode fixed for three nights – milder cooling setpoint, low fan speed, TV and minibar fully off, towel warmer disabled – then 8,405 kWh could have been saved. That is 34% of the energy spent on unsold rooms; annualised, roughly 16,600 kWh.
Why only with blocking? Because a scattered empty room can be sold at any moment, so it has to be kept ready: cooled, with a cold minibar. A closed-off, completely empty wing does not. There, a deeper setback is acceptable, because there is nothing to spoil.
This leads to the third lesson: this energy decision is not an engineering decision but a booking-management one. Room allocation is decided by revenue management and the front office, and today, typically, nobody there thinks in terms of energy cost. Compared with that, fine-tuning the temperature setpoint is negligible.
The timing makes this far from academic: the Hungarian exchange price for electricity (HUPX day-ahead) averaged 152 EUR/MWh in August 2026, against 80.5 EUR/MWh a year earlier – an increase of nearly ninety percent in one year.
What this number leaves out
The 8,405 kWh is not homogeneous. Half of it is a measured value (4,477 kWh): the consumption of the TV, minibar and towel warmer in an empty room can demonstrably be brought to zero. The other half is a modelled estimate (3,928 kWh, chiller side) resting on three assumptions: a fan-coil heat output of about 1.2 kW, a chiller COP of about 3.0, and standby mode halving the cooling run time. Our analysis puts the total between 6,400 and 10,400 kWh.
The most important limitation is that the analysis covers only the room-side, secondary circuit, plus the cooling energy actually delivered to the rooms. It does not include the primary side – the chiller's auxiliary equipment, the primary and secondary circulation pumps, air handling and ventilation, distribution losses, and the chiller's poor part-load efficiency. Taking a whole floor out of service would save on these too – by a realistic engineering estimate, roughly the same amount again. So the published figure is a lower bound, not a cautious midpoint.
We also did not examine the operational and commercial cost of blocking: the saleability of room categories, guest satisfaction, lost minibar revenue, and the quality of stock kept in a fridge switched off for days. These are real costs, and none of them appears in the balance above.
The fourth lesson is that you can't optimise what you don't measure. There is no sub-metering on the primary side, which is why the biggest item is also the most uncertain.
The most important conclusion, though, is that the biggest item – air-conditioning unsold rooms – can be tackled without blocking. There's no need to touch the booking process: it is enough to identify rooms that have been empty for several days from night-time keycard presence and give them a milder setpoint. The infrastructure needed is already in the walls.
We have written about hotel energy metering before: How to save money and analyze a hotel's energy consumption?
At Yabune, we design, build and operate smart home and building control systems, from family homes to hotels. If you run a hotel and want to see how much your empty rooms consume, tell us what you're planning: we'll look at what your system already measures and what is worth adding.