Every advisor in the room is looking at the same star. None of them can see this.
The 117-Hotel Cohort.
259,647 guest reviews. 117 hotels. 8 countries. 4 platforms. 25 to 26 languages. Read once, adversarially tested, published once. This study reads the way a due-diligence report reads: a claim, a number, a denominator, and a stated limit, chapter after chapter.
Which one is losing guests.
Two assets, same rating on the listing site. Select a card.
At an identical 8/10, Luxury doesn't return at 16.2%, Midscale at 4.0%: a 4x spread hiding inside one number a due-diligence memo treats as fixed.
n=24,005 of 47,666 rating-8 reviews with a won't-return signal present (50% coverage) · Wilson 95% CI · full breakdown in Chapter 3
If you underwrote both at the same implied non-return rate, one of those numbers is already wrong, and it is not a coin flip which one. Guess right and you looked sharp for free. Guess wrong and nobody in the deal room finds out until the numbers come in. What would make you re-check a rating you had already priced in?
Eight chapters. Pick where you land.
A score is a settlement, not a verdict.
An average is built to resolve disagreement by erasing it: a hotel rated "overpriced" by its guests carries a 57.0% non-return signal against 2.7% for guests who call it worth it, the same cohort, the same rating band, roughly 21 times apart on the one thing that predicts whether they come back.
Sample details & methodology
If your underwriting model treats the published rating as a single number, it is already averaging away the one signal that predicts who comes back, and the retention line in your projections is the line this erases first. Where in your own model does that retention line currently sit: assumed, or checked?
This mechanism reproduces at national scale, too. See the Portugal 500 →
Luxury does not buy immunity. It buys better concealment.
The gap between the published score and the guest's actual intent is widest in the middle of the scale, not at the extremes, and it is widest of all at the top of the market, where the room rate and the reputational stakes are both highest.
A 4x spread between the tier that charges the most and the tier that charges the least, both wearing the identical published star. The full four-tier breakdown, Upper-Upscale and Upscale included, sits below.
Sample details & convergence note
For a Luxury asset, an 8/10 average should trigger a capex-sequencing conversation before the tier below it does, not after. On the Luxury asset you are underwriting right now, has that conversation already happened, or is the 8/10 still reading as safe?
"Overpriced" is not a rate problem. It is a friction problem.
The instinct is to cut the rate. The data shows the rate is rarely the thing the guest is actually pricing; it is a stand-in for an unresolved friction event the review is really describing.
Lower the rate
Among satisfied (8+) guests, overpriced perception predicts 6.3x more guests who won't return.
Find the friction
Among those guests who name an item, they cite ancillary charges (breakfast, parking, minibar), not the room rate.
Before a rate card gets touched, the ancillary-fee line items are the cheaper thing to review, and per this chapter, the more accurate one. Which line would you check first: the room rate, or what got added to it at checkout?
The loyal guest who has already left.
Once a repeat guest privately concludes the stay declined, in-stay recovery gestures do not appear to reopen the verdict by checkout. Correlational, not causal; the data cannot separate "the gesture failed" from "the guest had already decided."
Sample details & CI
The moment a loyal guest's language shifts from comparison to complaint is a closing-window signal, not a routine service-recovery ticket. Is that shift being tracked anywhere on your side as its own line, separate from whether a recovery gesture happened at all?
Non-return read from what guests write, not from reservation data: signal, not a booking record.
This isn't unique to this cohort. At national-index scale, 31 of 344 eligible hotels (9.0%, CI 6.4–12.5%) show the identical pattern: an eroding promoter trend under a star rating that never moves. See the Portugal 500 →
What the rival review is actually telling you.
A competitor-named review is usually read as noise, one disgruntled guest venting. Read as a structured signal instead, it is the highest-concentration, plain-text comp-set data a property already has and is not using.
Sample details
Every one of these numbers is reproducible back to the reviews that produced it.
The seller's broker will have a comp set on file. Does it include which comp-set hotel is actually beating this asset on food and facilities, by name, or does the memorandum stop at the aggregate score?
The fix that works and is almost never used.
This is not a "be nicer to guests" finding. It is a timing finding: the exact same problem, caught before the complaint instead of after it, produces a different guest by 43 points.
Sample details & CI
This is the cheapest lever in the whole study, it needs no capex and no rate change, and it is running at under 4% adoption everywhere in the cohort. On the asset in front of you, is anyone measuring how often staff catch a problem before the guest complains about it?
The DD Gap, across the whole cohort.
Chapter 1 showed this gap pooled across the whole cohort. This is the same mechanism run hotel by hotel: not an artifact of a handful of outliers dragging the average, a structural pattern present in 86% of the hotels with enough data to test it.
Sample details & methodology
86% of comparable hotels carry a gap large enough to matter. Before you sign, do you know this asset's own gap, and can anyone in the room say where that number would come from?
One rating. Two different assets underneath it.
The DD Gap, Study 01, Chapter 8 · 81 hotels, 86% clear a 40-point gap
Open as one-page exhibit →What to do with eight chapters.
Separate what's structural from what's fixable. The tier gap and the concentration pattern are patterns to monitor; the recovery-timing gap is a lever to pull this quarter.
Sequence by cost. The cheapest fix in this study costs nothing but attention: catch the complaint before it's filed.
Read this as a question generator, not a verdict. Every chapter here is a question you can put to an operator or a seller's broker before you commit capital. Ask it in the room and watch who has an answer ready.
Every number here re-runs from the same public base on request.
Eight chapters. One question the other side of the table hasn't prepared for.
Every angle run. Each survivor closed in a decision the rating alone cannot reach.
This is an independent practice in its first engagements: commission a read and you are among the first, and the work is priced and structured accordingly.
Know someone underwriting a hotel on its rating right now? Send them Chapter 1. Better they hear it from you than from the seller's broker.