Hedonic Intelligence · Asset Intelligence brief

Asset Intelligence
from public guest reviews

What 564 reviews of one hotel — read by a structured-extraction LLM — surface for an asset manager.
By
Rui Andrade
Co-founder · MSc Informatics Engineering, FEUP
José Andrade
Co-founder · PhD candidate, U. Vigo
Methodology
Hedonic Intelligence NLP v15
Single-pass LLM extraction · 40+ dimensions per review
Worked case
Upper-upscale Rome city-centre asset (anonymised)
564 reviews · 7-hotel CompSet · 13 months
Hedonic Intelligence · hedonicintelligence.com01 / 12
§ 02 · Framing

The owner-side intelligence gap

Three asset-management questions that fall outside what operator-installed tools were designed to answer.
Asset-side question Conventional data path
Is this operator delivering the guest experience our brief specifies? Operator's own surveys; their installed reputation tool; periodic on-site visits.
On a target we are underwriting, what does the guest experience look like — before NDA-level access? Manual reading of TripAdvisor and Booking; a starred summary; a GM call.
Across our advisory mandates, which assets show early signs of operator-side drift? Quarterly cycle; lagging indicators; reactive rather than predictive.
Operator-side tools — ReviewPro, TrustYou, Customer Alliance — are designed for the GM and front-office team responding to reviews. The owner, the asset manager, and the transaction advisor were not the design buyer.
Hedonic — Capabilities deck02 / 12
§ 03 · Data layer

What we extract from one review

Single-pass LLM extraction with schema-enforced JSON output. Multilingual native, runs from public reviews only — no PMS, no operator cooperation, no NDA.
01 · Aspect-based sentiment
Per-aspect verdict (room, staff, food, location, cleanliness, value, …) with verbatim quote, journey stage, expectation-anchor.
02 · Friction events
Root cause (staffing / system / physical / policy / communication), severity 1–5, guest effort, resolution status.
03 · Loyalty signals
Promoter / detractor classification by semantic signal (advocacy, return intent, will-not-return) — not by rating threshold.
04 · Named entities
Staff names with action verbatim, room types, restaurants, landmarks, named competitors, with interaction-quality grading.
05 · Guest context
Occasion, group size, stay purpose, traveller profile, return-barrier (fixable vs. structural), wellness guest type.
06 · Temporal dimensions
Per-period sentiment, recent-shift detector, aspect-level Improvement Velocity Index over rolling windows.
Reference: NLP Methodology v15 — extraction §2 (per-review fields), composite metrics §3, multilingual handling §5. Underlying model: Gemma 4 26B-A4B with XGrammar-guided JSON; cache signature-keyed and version-controlled; prompt drift triggers automatic re-extraction.
Scale to date — pipeline runs against 250,000+ guest reviews across 118 hospitality assets, processed natively in any source-market language without a translation step.
Hedonic — Capabilities deck03 / 12
§ 04 · Metric stack

Designed for capital decisions, not turn-of-shift decisions

Each metric below is computed from per-review structured fields (slide 3); none uses fixed rating thresholds.
Metric What it measures Range / reference
GSI % of guests showing semantic advocacy behaviour — not a 9–10 rating threshold. 0–100% · NLP §3.1
NLS Net loyalty score — promoter % minus detractor %, both extracted by semantic signal. −100 to +100 · NLP §3.2
CRI Churn Risk Index — friction × effort × severity, with a service-recovery discount. 0–86 (log-compressed) · NLP §3.4
EGI · CTS Expectation Gap Index (% complaints framed against expectation) · Competitive Threat Score (per-competitor verdict roll-up). NLP §3.7, §3.11
CPI · RevPAM · IVI Competitive Position Index (ADR percentile) · Revenue per Available Square Metre · Improvement Velocity Index (aspect-level temporal delta). NLP §3.15, §3.16, §3.10
All metrics are derived from the same NLP cache; reliability bands and minimum-sample floors are documented per metric in the methodology document.
Hedonic — Capabilities deck04 / 12
§ 05 · Reader-type fit

Where this lens fits

The same engine produces a different deliverable depending on the question. Four reader types, four shapes of output.
Reader type What the same lens delivers
Owner / asset manager Recurring operator-monitoring read on a property or portfolio — friction root-cause split, IVI per aspect, NLS drift detector. No PMS access required, no operator cooperation. Refresh on a defined cadence.
Transaction advisor / DD analyst Pre-NDA guest-experience risk score on an underwriting target, anchored on its actual peer set. Surfaces concentration risks the operator's own dashboards smooth over.
Brand & operator-selection advisor Pre-RFP diagnostic on operator candidates and post-handover diagnostic on the incumbent. Aspect-level IVI separates pre-transition from post-transition perception — answers which aspects the new operator improved and which regressed.
Hotel director / GM One-page CompSet intelligence read with semantic loyalty, friction root cause, aspect-level temporal velocity, and pricing perception — calibrated to the property's actual peer set.
All four output shapes use the same data layer (slide 3) and the same metric stack (slide 4); only the question and the cadence change.
Hedonic — Capabilities deck05 / 12
§ 06 · Section divider

A worked example

The next five slides apply the data layer and metric stack to one anonymised hotel and its 7-property CompSet — the Owner / asset manager shape of output (slide 5 row 1).
564
guest reviews analysed across Booking, Google, Expedia, TripAdvisor
7
competitor properties in the same Rome upper-upscale CompSet
13
months of operating window covered (Q1 2025 – Q1 2026)
Property: upper-upscale Rome city-centre asset. Operator and property name redacted.
Hedonic — Capabilities deck06 / 12
§ 07 · Finding type 1

The revenue-loyalty gap

Financial extraction can be excellent while active advocacy remains thin. Two metrics, computed from the same dataset, separate them.
FINANCIAL EFFICIENCY
€24.50 RevPAM
Revenue per Available Square Metre — €662 ADR over ~27m² average room, placing the asset in the ULTRA band of space monetisation. (NLP §3.16)
LOYALTY GAP
+81
+35.8
DELTA
~45 pts
NPS
passive satisfaction
(rating-derived)
NLS
active loyalty
(semantic-advocacy basis)
Only 3% of reviews are pure detractors. The shortfall is not dissatisfaction — it is a systemic inability to convert silent satisfaction into advocacy in the review text itself.
Rating-based NPS uses the 9–10 score as the loyalty proxy; NLS classifies advocacy behaviour in the review text — a 10-rated review with no advocacy language counts as neutral, not as a promoter (NLP §3.1).
NLS = Promoter % minus Detractor %, both classified by semantic signal in review text (NLP §3.2). NPS shown as the conventional rating-derived anchor for contrast (% rating ≥9 minus % rating ≤6 on the 1–10 platform scale) — not the deck's primary loyalty metric. Sample: 564 reviews, 13-month window.
Hedonic — Capabilities deck07 / 12
§ 08 · Finding type 2

Friction with root cause and resolution status

Each friction event is one structured record — root cause, severity, guest effort, resolution status — not a complaint count.
16% FRICTION 84% NO FRICTION REVIEWS · n=564 ZOOM 90 FRICTION EVENTS RESOLUTION 10% RESOLVED · 9 90% UNRESOLVED · 81 ROOT CAUSE · WITHIN UNRESOLVED 33% POLICY 29% PHYSICAL 23% STAFFING 14% OTHER
GOVERNANCE RISK
14.4% of all reviews carry an unresolved friction event (16% × 90%). The number-one root cause is operational policy (33%). In a normal third-party-operator scenario this would be the immediate focus of the next governance meeting.
EFFORT SIGNATURE — LOW
Most friction events carry a low guest-effort signal (NLP §3.8). Guests are not escalating — they are absorbing friction silently and leaving without writing the complaint that would let the operator fix it. Low effort with a 90% unresolved rate is the asset-management red flag, not a comfort signal.
Friction events extracted per review (NLP §2.8a, fields: root_cause, severity 1–5, guest_effort, was_resolved). Aggregations per NLP §3.12 (Friction Resolution Scorecard); guest-effort bands per NLP §3.8 (Low ≤ 1.5 / Moderate ≤ 2.2 / High > 2.2 on the 1.0–3.0 scale).
Hedonic — Capabilities deck08 / 12
§ 09 · Finding type 3

Aspect-level temporal analysis

Improvement Velocity Index per aspect — newest 25% of reviews vs. oldest 25%, threshold ±0.15 — surfaces which aspects are improving, stable, or actively degrading over the window.
Aspect IVI What it means
Check-in ▲ +1.33 Operational improvement; front-office layer is the strongest temporal signal in the dataset.
Value for money ▲ +0.23 Pricing or yield management improving in guest perception over the window.
Room comfort / cleanliness — Stable Within ±0.15. Operational consistency.
Food & dining ▼ −0.29 Active degradation. Threshold-breaching decline of the breakfast layer detected as a primary-driver of detractor conversions.
USE CASE — OPERATOR TRANSITION
Apply the same engine across a window that brackets a management-company handover, and the IVI delta per aspect separates pre-transition from post-transition perception.
Which aspects did the new operator improve? Which regressed?
The answer is in the data layer, not in the operator's narrative.
IVI methodology: NLP §3.10 (newest 25% vs. oldest 25%, ≥2 mentions per aspect per period, minimum 8 dated reviews). Per-period mode and recent-shift detector (last-3-periods divergence ≥ 10pp): NLP §4.5.
Hedonic — Capabilities deck09 / 12
§ 10 · CompSet matrix

The CompSet intelligence matrix

One row per hotel, columns for each metric in the stack. The same dataset, the same definitions, the same window. Hotel names anonymised.
Hotel n NLS GSI Friction ADR CPI RevPAM
Hotel A 3,306 +71.0 74.5% 22% €529 9 €21.95
Hotel B 1,780 +67.5 70.4% 25% €2,101 100 €39.93
Hotel C 1,120 +47.9 52.1% 14% €1,314 54 €35.79
Hotel D 1,379 +45.0 49.9% 22% €1,170 46 €32.98
Hotel E 3,306 +37.5 41.7% 22% €376 0 €17.19
Subject Property 564 +35.8 38.8% 16% €662 17 €24.50
Hotel F 521 +32.8 43.6% 20% €1,372 58 €40.68
→ The extraction paradox
Highest RevPAM (€40.68) but aggressive extraction penalises NLS (+32.8) — the most detractor-heavy advocacy gap in the set.
→ The loyalty benchmark
Loyalty leader at median ADR. Proof that high NLS does not require ultra-luxury pricing.
→ The subject window
Lowest pure-detractor share in the set (3.0%). The asset-management opportunity is converting passive satisfaction into active advocacy — slide 7 quantified the conversion gap.
All metrics computed from the same NLP cache (NLP §3.1–§3.16). n = analysed reviews. Window: ~13 months ending Q1 2026 (or since opening for newer assets). Hotel names redacted to allow reuse across prospects.
Hedonic — Capabilities deck10 / 12
§ 11 · Decision queue

From signal to decision queue

Each finding maps to one decision the asset committee or operator must make. Cost ranges below are owner-engagement estimates calibrated to the property profile — they are not a price quote for our work.
# Intervention Cost (owner engagement) Timeline Accountable owner Target KPI
01 Advocacy solicitation at check-out €0 7 days GM (operator) NLS > +50
02 Operational policy audit €0 60 days Owner / Asset committee Eliminate 30 active friction foci
03 Front-office service-recovery systematisation €5K OpEx 30 days GM (operator) Unresolved-friction rate < 50%
04 F&B (breakfast) restructuring +€3–5K / mo 45 days F&B Director (operator) Halt IVI of −0.29
05 Physical CapEx (HVAC + sleep upgrades) €55–90K CapEx Immediate Owner / Engineering Resolve noise & sleep-quality complaints
What you receive from the engagement is this deck — no markdown, spreadsheet, or raw-extract files are shared with the client. The interventions above are downstream owner-side or operator-side decisions the analysis surfaces — they are not Hedonic services. Costs and timelines are operator-engagement estimates calibrated to the property profile (NLP §3 methodology cross-reference).
Hedonic — Capabilities deck11 / 12
Next step

This deck is one read.

11,976
guest reviews analysed across all 7 hotels in this case study
40+
structured dimensions extracted per review
The same methodology runs across a portfolio, with peer set anchored per asset and refresh on a defined cadence.
To explore what this lens shows on a property of yours, or across a portfolio — get in touch through hedonicintelligence.com.
Hedonic Intelligence
Asset-side guest-experience intelligence · Porto, Portugal
hedonicintelligence.com
No demos · Project-priced
Hedonic — Capabilities deck12 / 12