What a Guest Sentiment Benchmark Can’t Tell You (and What to Pair It With)

Key findings at a glance

• Review-based benchmarks describe what guests say. They don’t describe what happens inside a hotel, its revenue or the guests who never write a review.
• Coverage differs by aspect. In Interlaken, Front Desk & Staff has 8,605 Review Mentions across 42 hotels, while Maintenance & Upkeep has 1,040 across 19.
• Zermatt’s Maintenance & Upkeep figure (56.3%) rests on 32 mentions across 2 of 45 hotels. It’s too thin to read in either direction.
• Lucerne and Interlaken’s headline figures moved between dataset versions, mainly because the hotel set changed. Drift needs a same-hotel comparison.
• For rate, occupancy and revenue questions, pair guest scores with STR and PMS data.

What a sentiment benchmark is

A guest sentiment benchmark starts with public reviews. Each review is split into aspects, such as Front Desk & Staff, Room Quality or Maintenance & Upkeep. Each mention of an aspect is classified as positive or negative, and the benchmark reports the share that’s positive.

Two terms matter here. Review Mentions count how many times an aspect is mentioned. One review can mention several aspects, so a mention isn’t a unique review. Hotels Reporting shows how many hotels have the aspect among their top eight. That measures how prominent a topic is, not whether it’s absent elsewhere.

The SENTEEZ dataset uses a 12-month window, refreshed quarterly. The overall SENTEEZ rating is calculated from review content, not copied from a booking platform, and the data comes only from publicly available reviews.

Five questions a benchmark can answer

Used within its limits, a review benchmark can answer these five questions.

1. How do guests rate each aspect of your hotel? Coverage varies by aspect, so read mention counts next to every score. In Interlaken, Front Desk & Staff has 8,605 Review Mentions across all 42 hotels, and Maintenance & Upkeep has 1,040 across 19. Both are usable, but they carry different weight.
2. How do you compare with one rival? A one-to-one comparison works when both hotels report the same aspect with enough mentions on each side. The free teaser report is built on that approach.
3. Which complaints repeat? A complaint that recurs across many hotels in a town is a pattern, not an accident. In Interlaken, Maintenance & Upkeep is positive in 24.1% of mentions across 19 hotels. That points to a shared pattern, not one building’s problem.
4. Is a weak score yours or the market’s? Compare your score with the town figure, the top third of hotels in town and your own tier. Article 2 walks through that method: Is It Your Hotel or Your Market?
5. Is sentiment moving? A benchmark can show change between versions. Whether that change is real depends on whether the same hotels are being compared. See the drift section below.

Five questions it doesn’t answer

Public reviews can’t answer five questions. Each has a better source, and a good benchmark says so.

1. Why a guest felt that way. Themes are signals, not proof of cause. A recurring theme about air conditioning suggests where to look. It doesn’t show that air conditioning caused the score. Pair it with conversations with staff and guests who can explain their own experience.
2. What happened inside the hotel. A review can say the room was hot. It can’t say whether a unit failed, a policy was broken or a repair was slow. Pair it with operational audits and maintenance logs.
3. How the hotel performed financially. Rates, occupancy and revenue aren’t in reviews. Pair guest scores with STR and PMS data and read the two side by side. A guest score is a signal to investigate, not a revenue figure.
4. What silent guests think. Most guests never write a review, and those who do aren’t a random sample. Pair the benchmark with direct post-stay surveys to hear from the rest.
5. What happened beyond the 12-month window. The window is fixed by design, so older events fall outside it. Pair the benchmark with your own historical records and earlier reports.

The drift trap: why refreshes can mislead

Refreshed datasets are useful, but they can mislead. In the most recent refresh, Lucerne moved from 77.0% positive across 15 hotels to 70.1% across 39 hotels. Interlaken moved from 73.4% across 5 hotels to 72.6% across 42.

Those movements mainly reflect which hotels were measured. We haven’t yet published a same-hotel comparison over time, so we can’t say how much, if any, of the change is drift. Before you read a trend from two benchmark versions, check whether the hotel set is the same. If it isn’t, the first explanation to test is the sample.

Guardrails built into the data

Several rules in the SENTEEZ method exist to stop a benchmark from claiming more than it knows.

• Thin-data rule. Aspects with fewer than 50 Review Mentions or fewer than three hotels are excluded from comparisons.
• Ranking threshold. Hotels are ranked only when they have at least 100 reviews.
• Coverage beside every score. Each aspect figure appears with its Review Mentions and hotels reporting.
• Themes are samples. The “why” lines illustrate recurring patterns. Their counts aren’t summed, ranked or published.

Zermatt’s Maintenance & Upkeep shows why the rules matter. The figure is 56.3% from only 32 Review Mentions across 2 of 45 hotels. It’s too thin to read in either direction. We document it as insufficient coverage, not as evidence that Zermatt’s hotels maintain their properties well or badly.

How to spot a weak benchmark

Before you trust any review benchmark, including ours, check these seven points.

1. Sample size: how many hotels and reviews sit behind the headline figure?
   Coverage: how many hotels report each aspect you care about?
2. Counting: are the figures Review Mentions or unique reviews? Calling mentions “reviews” overstates the evidence.
3. Window: which months are covered, and how often is the data refreshed?
4. Source of the rating: is it calculated from review text, or copied from a booking platform?
5. Changes: did the hotel set change between versions, and is that stated?
   Thin aspects: are they excluded or clearly flagged?

Pairing map

Use the benchmark for what it does well, and bring in the right source for each question it can’t answer.