Half price on the same-day bed, and the bed did not fill

The decision

On 27 September our hostel in Oaxaca was 10.87% full. Five beds out of forty-six. The engine that rewrites our rates every morning had settled on a discount of -8% for the same-day band and -8% for one-to-three nights out. We overrode it by hand and set both bands to -50% through 4 October.

Half price on every bed a guest could book for tonight or for the next three nights. It is the move every operator reaches for in a dead week, and we had a specific reason to think it might work: near-term bookings had started moving again even though the nights were not filling.

Two nights have now settled. This is what they say, and what they do not.

What the four nights actually did

• 25 September, engine pricing: 19.57% occupancy, ADR MXN 225, MXN 2,026 for the night.
• 26 September, engine pricing: 17.39%, ADR MXN 191, MXN 1,531.
• 27 September, first night at -50%: 10.87%, ADR MXN 147, MXN 737.
• 28 September, second night at -50%: 21.74%, ADR MXN 155, MXN 1,554.

The two nights before the override averaged 18.48% occupancy and MXN 1,778 a night. The two nights under it averaged 16.31% occupancy and MXN 1,146 a night. Occupancy fell by about two points. Revenue fell by MXN 633 a night, which is a third of it.

Four nights is not a result. Two nights against two nights, in a house where one booking of three beds moves occupancy by six and a half points, is a story, not a measurement. We are publishing it anyway, because the interesting part is not the outcome. It is how hard this is to measure, and how confidently most of us conclude things from data exactly this thin.

Why the obvious comparison is worse than useless

The first thing anyone does is compare the week after against the week before. Ours looks like this: 15 to 21 September took MXN 23,515 at 42.24% average occupancy. 22 to 28 September took MXN 16,245 at 25.47%. Revenue per day fell from MXN 3,359 to MXN 2,321.

Read like that, the discount looks catastrophic. It is not that, and the reason matters more than the number. The override only covers the last two of those seven nights. The other five were priced by the engine at its usual -8%, and they fell anyway: 47.83%, 32.61%, 28.26%, 19.57%, 17.39%. On 15 September we were 89.13% full. Independence Day weekend ended, the market emptied, and the season moved a great deal further than the price did.

If you cut your rates during a demand collapse and then compare against the week before the collapse, you will convict the discount of something the calendar did. The reverse error is just as common: cut in the week before a festival and the discount takes credit for a house that was going to fill regardless.

The measurement that keeps coming back

Our engine keeps its own elasticity read, and it has been pointing the same way for a month. Across 28 logged days, nights priced at -8% on the near bands averaged 33.7% occupancy over 23 days, while nights priced at 0% averaged 42.0% over 5 days. That is 8.3 points of occupancy worse with the deeper discount. A fresh read this week came back at -7.8 points. Two independent reads, same sign.

The honest reading of that is not that discounting drives guests away. It is that the engine cuts price when the house is empty. The discount follows the empty night; it does not cause it. Any rules engine that lowers rates in response to weak occupancy will produce a log showing deeper discounts alongside lower occupancy, forever, no matter what the guests actually do.

Our manual override does not escape that trap either. We chose to cut in the emptiest week of the month, for the obvious reason. The only design that separates the two is alternating: some nights discounted, some not, chosen in advance, regardless of how the house looks that morning. We have exactly that configured as a four-day pattern. It has recorded fourteen consecutive days on one arm and has not produced a single day of the other since 10 September. The test we designed is not the test we have been running, and we did not notice for twenty days.

The one number pointing the other way

On the 29 September run, near-term pickup came in at 11 bed nights against a seven-day average of 2.1. A ratio of 5.13. Bookings are moving at five times their recent pace even though the nights have not filled yet.

That is the signal we will judge the override on when it closes on 4 October, and we decided that before the window ended rather than after, which is the only way a criterion means anything. Pickup leads occupancy by days. Judging a same-day discount on the same day it runs measures your booking window, not your price.

Which number decides it

Occupancy and rate both mislead on their own. Over the last 30 settled nights at our property, the correlation between nightly revenue and occupancy is +0.91. Between nightly revenue and ADR it is -0.11. The six fullest nights of the month averaged MXN 4,950. The six most expensive nights averaged MXN 3,322.

That is not an argument for filling at any price. It is a statement about where we are: at 42.46% average occupancy across 30 nights, the empty bed is the binding constraint, not the rate. A property running at 85% in high season would read the opposite sign off the same two columns, and should.

The metric that survives both cases is revenue per available bed: the night's revenue divided by every bed you own, sold or not. Ours went 44.04, 33.28, 16.02, 33.78 across those four nights. It is one number, it cannot be gamed by giving beds away, and it cannot be gamed by pricing yourself empty.

What we would tell an operator with 20 to 80 beds

• Before you cut, write down three things: which nights the cut applies to, the date you will decide, and the single number that decides it. Written after the fact, any of the three can be chosen to agree with what happened.
• Use revenue per available bed. Not occupancy, not ADR.
• Do not compare against last week. Compare against the nights you did not discount, inside the same demand conditions, if you have any. If you do not have any, say out loud that you are guessing.
• Judge on settled nights, not on the live dashboard. Our own brief reported the first override night at 46% occupancy. It settled at 21.74%. The field held a percentage and was read as a bed count, and 21 beds out of 46 is 45.7%, which is exactly the kind of plausible number nobody questions.
• Decide in advance what would make you stop. A discount with no exit condition is not a decision, it is a new rack rate.

The override closes on 4 October. We will publish what it closed at, including if it closes badly.

If this sounds like your property, see the three Nightfill tiers and start a free 14-day pilot on your own data: view pricing.