For the decisions you can't undo.
The hotel to worry about isn't the one whose rating looks a little too good. It's the one its own repeat guests, guest against guest, love more than the world does. Two hotels can carry the identical rating for opposite reasons, and the instinct every buyer walks in with points at the wrong one.
Screen Portugal's own curated best 500 against six of the hardest-to-fake risk signals in the record, and hotels the public rating flatters show no elevated risk at all. Hotels the rating quietly underrates, the ones its own guests would rank higher, flag red about three times more often than average. Two separate methods, built independently, land on the same hotels.
For the asset you're underwriting right now: does its own guest record say it's flattered by its rating, or quietly underrated by it, and has anyone actually checked?
Property-Pull Index, national index. The star score above can't tell you which kind of asset you're looking at: it explains barely a tenth of the difference. The asset you're underwriting right now: is its rate of returning guests the property's own, or the region's? Corpus: 500 hotels, Portugal, 2026-07-10 index cut.
Full intervals + method
Full range across all 493 qualifying hotels: 67.2–99.1%, median 92.7%. Narrow-band cut (rating 8.9–9.1, n=86): spread is 97% as wide as the full national range. National correlation: Pearson r=0.334, R²=0.112, Spearman rho=0.319 (p=4.0e-13, n=493). Minimum-inclusion floor: 15 attributed return-intent mentions per hotel; national base rate 91.4% property-attributed [Wilson 95% CI 91.2–91.6%], empirical-Bayes shrinkage k=29.4.
You know your hotel. You don't know why you're losing to the one down the street.
Revinate or ReviewPro already show you a topic score against the handful of rivals you picked. What they don't hand you: the guest-by-guest resolution rate behind that score, which specific friction is costing you the guest who walked, or a peer set built from who your own guests actually compare you to, not just the five you chose. The read does. No cooperation from the rival needed.
Show the numbers
You already own two of these tools. Neither one can walk into the room with you on the day it matters.
STR tells you the market: how the category is pricing and filling rooms. Reputation-management platforms, ReviewPro and TrustYou among them, are bought by a hotel to manage its own rating from the inside, with the operator's cooperation. Neither is built for the person deciding on an asset they do not yet own, or the advisor who cannot get inside the operator's dashboard on a pre-acquisition target. This is a finished, outsider read of any hotel, including ones you do not operate, built once for the person deciding on the asset, not a subscription the hotel buys to watch itself.
Neither tool you already pay for can read a target before you've bid on it, or benchmark an asset you're exiting against the buyer's own comp set. If the read has to wait for operator cooperation, it isn't available on the day the decision gets made.
Pre-acquisition DD. STR shows you market pricing and occupancy, not the target's own guest experience. Reputation platforms need the seller's cooperation and a dashboard login, so they're not available before a bid. This reads any target off public reviews alone, no seller cooperation needed.
Same story on exit positioning, operator reviews, and renovation cases: none of it clears until someone reads the guests, not just the market or the operator's own dashboard.
SEE ALL FOUR DECISION MOMENTS
| Decision moment | STR | Reputation platforms (ReviewPro, TrustYou) |
Hedonic Intelligence |
|---|---|---|---|
| Pre-acquisition DD | STR Market pricing and comp-set occupancy only, no read on the target's own guest experience | ReviewPro / TrustYou Requires the seller's cooperation and dashboard login; not available before a bid | Hedonic Intelligence ✓ Any target, public reviews only, no seller cooperation needed |
| Exit positioning | STR Shows where the asset sits on rate and occupancy, not on guest sentiment | ReviewPro / TrustYou Own account only, not accessible to a non-operator on someone else's asset; and its comp set is whichever rivals the operator already picked, not the buyer's | Hedonic Intelligence ✓ Reads the asset and its rivals the same way a buyer's advisor will |
| Operator-performance review | STR Market context only; cannot isolate what a specific operator is doing to guest loyalty | ReviewPro / TrustYou Reads the operator's own dashboard, built for the operator, not an independent second read | Hedonic Intelligence ✓ An outside, independent read of the same guest record |
| Renovation business case | STR No guest-level detail on which friction a capex line would resolve | ReviewPro / TrustYou Surfaces flagged reviews, not a ranked, sized business case | Hedonic Intelligence ✓ Names and sizes the specific friction, ranked by exposure |
This does not replace your PIP consultant or your technical DD. It is the independent second read that checks their number against the guest record before the room votes on it. Bring it to the next PIP negotiation →
On the asset you're managing right now: if the operator's capex ask and the guest record disagreed, which one would you currently find out about first?
Two hotels can carry the identical 9.00 rating and need six times as much money spent on the building. The listing page will never tell you which one you're looking at.
Hold the star rating fixed at 9.00 and look only at how much of each hotel's guest complaints an engineer would actually have to fix, not a training memo, and the two hotels closest to an identical score in the whole national index span a 5.7x range. Read it against your own PIP line items before the next allocation gets signed off.
INTERVALS + METHOD
Across the full 8.90–9.10 rating band, 75 hotels clearing the volume floor range 4.9% to 62.7% (p10=18.6%, p90=47.1%). The tightest pair, within 0.03 rating points of an identical 9.00 (n=20 hotels at that cut), spans 10.9% to 62.7%, a 5.7x range at essentially zero rating gap. National baseline 30.9% lands almost exactly on the band's own median (30.5%): the story is entirely in the spread, not a shifted center.
For the two assets closest to each other on your own shortlist's rating: do you actually know which one is carrying the bigger repair bill, or are you assuming the ratings already told you?
Same screen, all 500 hotels in the national index, not just the ones already on your desk. See the full method on the Portugal 500 →
Three categories a skeptic assumes a review-mining tool cannot reach.
Concentration risk, where guest perception is leaking to a named rival, and an exposure line no operator report will ever volunteer: the same investigation runs on all three, each closing in a verdict, not a dashboard metric.
The exposure line your operator's monthly report will never volunteer: how much of your guest experience rides on one person who could hand in notice tomorrow.
At national-index scale, across 389 of Portugal's 500 best hotels, one named staff member is behind a median 10.1 to 12.5% of that hotel's warmest reviews. Reviews that name that person by name run 32.6 points warmer than reviews that don't, even comparing reviews of the same length. No HR system, no PMS, and no operator report tracks review silence; a hire-away event changes how warmly guests write about the stay with no line item anywhere in what you're sent. This is a risk-register entry, not a compliment.
389 of 500 hotels clearing a review-volume floor. Length-matched lift CI 32.2–33.0pp. Never paired with a specific hotel or brand.
Don't run a blanket service-quality initiative for this. Triage the 1 in 100 guests actually causing it.
1.1% of guests generate 53.8% of all will-not-return signals, a 49x concentration, since 2023. An averaged satisfaction score is mathematically designed to smooth exactly this kind of spike away, which is why the operator's own dashboard reads it as noise. The verdict: this is a named, individually-triaged recovery process, not a broad training spend.
Read from what guests write, not from reservation data; it signals intent, never confirms a booking. Denominator: all reviews 2023 and later, n=205,660; the concentrated group is 2,173 guests.
Guests are already voting on the kitchen, not the lobby, four times out of five.
Where a rival is named as better, food loses 81.8% of the time (108 of 132), value 79.1% (155 of 196), facilities 74.2%, service 64.3%, location 51.6% (fewest reviews behind this one). This text sits inside a review scored high enough that no alert fires; a star-rating scan reads it as one more 8/10. It is the highest-concentration, lowest-noise guest-perception signal in the corpus: where to point a capex or ops review first in the room, not a substitute for that review.
Denominator: guests who name a rival as better on that dimension, 2024 and later. Every one of these numbers is reproducible back to the reviews that produced it.
Of these three, which one would your operator's own report have flagged for you before you asked?
Built to answer one question: what does the record show that the rating cannot. Not a reputation score. A decision file.
You already check the building. Nobody has been checking the guests.
A building report tells you about the asset. The rating tells you the aggregate. Neither tells you whether guests are quietly deciding not to come back, or which specific friction is the reason.
Before you bid. The difference between a hotel whose guests feel ripped off and one whose guests feel like they got a deal is enormous, and a rating alone won't tell you which side of that line the property you're underwriting sits on. Most hotels in the study clear a 40-point gap between the two verdicts. See the full DD Gap chapter →
Mean gap 53.8pp per hotel [95% CI 51.3–56.4pp]; 86% of 81 qualifying hotels clear the 40-point threshold.
Before you fund the renovation. You'd assume the wear that matters most is at the bottom of the market. It isn't: the guests most likely to quietly stop coming back are staying at the top, the segment carrying the biggest checks, and the size of the property tells you nothing about which ones are hiding it.
Before you renew or replace the operator. It isn't your average guest driving the losses you can't explain. Roughly one guest in a hundred is carrying half of them alone, and an averaged score will never show you which one.
Of the three decisions in front of you right now, bid, renovation, or operator renewal, which one are you closest to making on the rating alone?
Cut the rate and you'll still lose them. "Overpriced" is a friction verdict wearing a price complaint's clothes.
The instinct when a guest calls the rate unjustified is to lower it. That's usually the wrong fix. Even guests who are otherwise happy with the stay are roughly six times more likely to say they won't be back once they call it overpriced, and most of them aren't really complaining about the room rate: three in five point at one specific charge instead, breakfast, parking, the minibar. That's a conversation with operations about that one line item, not a conversation with revenue management about the whole rate.
Among guests who were otherwise satisfied with the stay, calling it overpriced makes them about six times more likely to say they won't be back: 18.8% versus 3.0% for guests who felt they got a fair deal. If this were really about the price, cutting the rate would fix it. It doesn't seem to.
Among the guests who won't be back and say why, three in five point at one specific charge, breakfast, parking, the minibar, not the room rate itself. The fix is a conversation with operations about that one charge, not a conversation with revenue management about the whole rate.
Method + denominator
Cohort: 259,647 reviews, 117 hotels. Denominator: 8+ guests who said they wouldn't return, with a named ancillary item, 60.2% CI 56.4–64.0% (Wilson 95%).
The next time a guest calls your rate unjustified, is it the room rate you'll look at, or the line item underneath it?
The repeat guest who wrote they loved it. And then wrote it was worse.
A repeat guest who says this stay was worse than before is telling you something real. Roughly seven in ten of them also say they won't be back, more than thirty times the rate of a repeat guest who doesn't report a decline. This is a pattern, not proof of cause: once a guest has reached that verdict, a nice gesture during the stay alone doesn't seem to change their mind.
Among repeat guests who scored the stay exactly 8 out of 10 but said it was worse than before, close to half say they won't be back. The window to act is before that verdict forms, not after the review appears.
Decline-reporting repeat guests signaling non-return: 71.7% [68.3–75.0], vs 2.1% for non-declining repeats, a corrected 34x. Repeat-guest reviews rated exactly 8/10 with an explicit decline signal: non-return 46.7% [39.2–54.3], n=165.
This isn't just an artifact of the smaller study group either. At national-index scale, 31 of 344 eligible hotels, 9.0%, show this exact pattern: guests are quietly turning against the hotel while its star rating never moves.
Do you know which of your repeat guests already quietly filed that verdict?
One completed. Now a second, four times the scale.
The investigation started with 117 hotels. It has just published its first national release across Portugal's 500 best hotels, read the same way.
The 117-Hotel Cohort
PublishedEight chapters on what a hotel's rating conceals about who is quietly leaving, each with its number, its mechanism, and its stated limit.
The Portugal 500
PublishedEvery hotel Portugal is proud of, read the same way, four times the scale of Study 01. Adversarially tested the same way, before anything shipped.
Anatomy of a read
A walk through what a paid, one-hotel engagement produces, illustrated with cohort data. Not a finding. A blueprint.
The finding nobody else brings to the room. Delivered as one brief that sits on the shelf next to HVS or JLL.
SHOW THE MATH
An ADR movement typically flows 70 to 80% to GOP and 55 to 60% to NOI (industry rule of thumb, not a measured figure). Asset value moves with NOI through the exit multiple. Size the retention envelope on your own asset: your ADR delta × your room count × your occupancy × the 55–60% NOI flow-through above is the annual NOI this kind of finding typically has in play, before the exit multiple.
- One-off per engagement: no subscription, no onboarding, no recurring commitment. One hotel, 14 days. Terms, scope and price agreed on a short scoping call first.
- A finished PDF per property: where it lands against its bespoke peer set, the friction trajectory, the levers, each closing in a decision your committee can act on.
Confidence intervals + floor
Every proportion carries a Wilson 95% confidence interval. Floor: 39 reviews on the relevant dimension before a finding appears. Every finding traceable to the specific review sentence behind it.
Guests already told you where the leak is. It's the kitchen, not the postcode.
Of guests who named a competitor as better, the rival won on food 81.8% and value 79.1%, but location was closer to a coin-flip at 51.6%, the smallest-n category among the four. This is where guest sentiment is leaking to a named rival, not an inspected capex line: the read that tells you where to point your own operator's or QS's line-by-line review, not a substitute for it.
Denominator: guests who named a competitor as better (food n=132, value n=196; 2024+). Full analysis on the rival-leak chapter.
A 9-plus rating is not a clean PIP. It just means nobody's read the maintenance line yet.
Roughly 1 in 9 of an asset's own 9 to 10 reviews narrates unresolved maintenance in the same glowing line, and the per-property share runs from under 1% to over 20%, invisible to a scan of the published average. This is the line a QS finds later at cost; here it's free, before the bid.
Denominator: reviews rated 9–10 with maintenance narrative present (severity ≥ 0.4), n=45,937, 2024 and later.
If the finding nobody else brings to the room turned out to be about your asset, would you want to know before the bid or after?
The investigation runs every angle. What survives is the line you say out loud in the room.
A dashboard returns what it was programmed to find. This investigation forms the question from the data, attacks it, and ships only what survives. Every read is done by the person who built the method.
Illustrative. Constructed to show extraction depth, not from a real guest record.
From one sentence, the read pulls apart what actually happened. A guest writes one paragraph. The system doesn't just see "unhappy guest": it sees what broke (the check-in system), whether staff apologised (no), how many times the guest brought it up (three), and what it means for whether they come back (they won't). That's the difference between reading a review and reading past it.
See the full extraction
40+ top-level fields per review; objects expand with sub-fields. A review with 3 aspects, 2 friction events, 2 named staff members resolves to 88+ discrete data points in a single pass. The extraction schema is identical across every property and every language.
The 40+ extraction fields are the infrastructure that makes the investigation possible. They are not the investigation. The investigation is what decided to ask whether the check-in failure pattern persisted specifically in guests who named the star rating as their anchor, sized how many guests it would lose, confirmed peers show no equivalent cluster, and closed: systems problem, training problem, or a property-class mismatch the flag change cannot fix.
See a real one, killed before it shipped, on the Portugal 500 →
A pattern your own team can eyeball is not what you're paying for. What you need before you sign off on capex is the finding that survived an adversarial kill-test and closed in a systems, training, or property-class verdict your committee can act on.
What's the question about your own asset you'd want this investigation to form and attack first?
A read that hides its own limits is one your committee will eventually catch you on. This one tells you exactly where it stops.
You need a number your committee can defend. That means knowing what it counts, what it excludes, and where it cannot reach. Every finding in the read carries that boundary explicitly, which is why each one is actionable rather than decorative.
The perimeter is how you know the rest is true.
I hunt findings that were never on the menu.
A fixed pipeline returns only what it was programmed to label. The investigation decides what to ask.
| The question | Fixed pipeline / dashboard |
Investigation |
|---|---|---|
| Can it generate a question nobody pre-specified? | No Returns what its schema names |
Yes Hypothesis formed from the data |
| Does it kill its own findings adversarially before shipping? | No | Yes Findings killed adversarially before shipping; kills disclosed |
| Can it read a hotel you do not operate or own? | No Requires operator login / data |
Yes Public reviews only, any hotel |
| Delivers a finding defensible to an investment committee? | No Dashboard metric, not a closed finding |
Yes Wilson CI + peer comparison + decision close |
Where does your own dashboard stop, and have you ever seen it say so in writing?
From a hotel name to a decision your committee can act on, in two weeks.
Name the asset
Your named hotel, plus the question that aims the read: an asset-committee lens, an operator evaluation, a capex decision. Same day.
Bespoke peer set, one question
I build the comp set for that property specifically, not a static category bucket. The question shapes which diagnostic lenses run on the full record. Two business days.
The finished read
One PDF, every metric recomputed for your named hotel against its specific peer set, each finding closing in a decision. 14 days.
Re-read on cadence
Optional: re-run quarterly to track before/after on a repositioning or operator transition.
Which hotel would you name first: the one you're bidding on, the one you're exiting, or the one whose operator is up for renewal?
One person reads every finding that reaches you. That is not a gap.
I built this because no one on the buy-side had a read of the asset that was not coming from inside the operator's own system. MSc in Informatics Engineering, University of Porto. I designed the investigative method: the question architecture, the adversarial kill-test, the peer-set construction, the decision framing. I also built the extraction infrastructure that makes the investigation reproducible and measurable across any language and any property.
There is one person running the reads. That means every read is done by the person who built the method, not handed to a junior analyst, not summarised by a tool.
The numbers are reproducible from the same public base. The method, the perimeter, and the cohort are documented: see the public 117-hotel study.
This is an independent practice in its first engagements: commission a read and you are among the first, and the work is priced and scoped accordingly.
Is being among the first a reason to wait, or the reason to go first yourself?
The read, in plain terms. No brochure language.
What does Hedonic Intelligence do?
I read the full public review record of a hotel and return why its guests come back, why some feel the stay was overpriced, and which specific frictions are driving them away, sitting underneath its published rating. It arrives as one finished PDF, with every number traceable to the guest sentence behind it. The read works on any property from public reviews alone, with no operator cooperation.
How is it different from ReviewPro, TrustYou, or STR?
The short version: STR won't tell you about a target's own guest experience, and reputation platforms need the seller's cooperation to log into, so neither works before a bid. This reads any property from public reviews alone, no seller cooperation needed, and returns a finished brief instead of a dashboard.
Do you need the hotel's cooperation, PMS data, or an NDA?
No. The read is built from public guest reviews only: no property-management data, no first-party guest information, no operator sign-off. That means any hotel you name: a target before a bid, a rival you will never get inside, an asset under an operator you are deciding whether to keep.
Who is it for?
Hotel transaction advisors and brokers, asset managers at funds that hold hotels, and acquirers running due diligence. Also operators who want a true outside read of the rivals their own tools cannot see at depth. The common thread is a decision on a specific property, not a dashboard to monitor.
What do you deliver, and how long does it take?
One finished PDF on the hotel you name, in 14 days. It places the property against a real peer set, names the concrete problems to fix on the property, and closes each finding in a decision your committee can act on, with confidence intervals on every rate and an explicit statement of what the read cannot see.
What does the first read cost?
One-off per engagement: no subscription, no onboarding, no recurring commitment. I keep the first read small: one hotel, 14 days. The terms, scope and price are agreed on a short scoping call once you have named the property, so you know exactly what you are committing to before anything starts. Ongoing or portfolio work is priced from there.
Can you analyse a hotel I don't own or operate?
Yes, and that is the point. Because the read uses only public reviews, I can read any hotel from the outside: an acquisition target, a competitor, or an asset under an operator you are deciding whether to keep. No cooperation from the property is needed.
What data do you use, and is it GDPR-compliant?
Public reviews only, processed under the GDPR legitimate-interest basis (Article 6(1)(f)). No hotel-systems data and no first-party guest information. The extraction method and the synthesis are mine; the raw material is the same public reviews anyone can read. A sub-processor list is available for institutional procurement on request.
What decision is this actually for?
Not day-to-day reputation management, that's what your existing stack already does well. This is built for the moment a decision is irreversible and expensive: before you bid, before you sign a PIP, before you renew or replace an operator, before you position an asset for exit.
What is the Portugal 500?
The first national index of Portugal's celebrated hotels' guest experience: 500 hotels, four platforms, read the same way as the 117-hotel cohort. Published 2026-07-10, findings verified adversarially before shipping. See the findings →
Name a hotel. I'll read it against its real peer set.
A target before the bid, a rival you want read from the outside, or an asset under an operator you are assessing. Tell me the property and the question to aim the read. That is the whole first step.
A quicker first look at one named hotel, before you commit time to it: fine, worth a closer look, or a genuine problem.
First-read terms, scope and price are agreed on a short scoping call. One hotel you name, delivered in 14 days. I won't tell you what your hotel is worth; I tell you what its guests are deciding.