Your CX and BI tools tell you what happened. They cannot tell you why. We answered that from the reviews Marriott’s guests post for free.
Descriptive intelligence tells you what happened, in aggregate, at a lag. Nearly every CX and BI tool, Medallia and Qualtrics included, prioritizes by volume and correlation: what is mentioned most, and what moves with the score. None of that can separate a cause from a coincidence, rank fixes by payoff, or tell you what a fix is worth. This brief shows the gap on a single public company, so that every company living inside it can see their own.
Here is Marriott’s Orlando region, rebuilt the way every CX tool works: score it, rank it, trend it. The program looks healthy. The trend is flat. The topic list ranks issues by how often guests mention them, and the built-in driver analysis flags front desk and check-in as the number-two thing to fix.
Everything here is computed from real reviews, in the descriptive form a survey platform produces. It is accurate. It is also a map of symptoms with no diagnosis.
Nothing here is wrong. It is useful for knowing that a problem exists. But every question a leader asks next lives outside this frame: which driver is real, which fix pays off, what it is worth. Nearly every CX and BI dashboard ranks your complaints by volume; the premium ones, Medallia and Qualtrics included, add a driver analysis that guesses at importance from correlation. None of them can tell you which ones change the outcome. Hold onto that front-desk flag.
We took the same reviews and asked a different question. Not what guests mention, but what actually moves the outcome. A formal causal model, holding the hotel, the tier, the language, the review length, and the reviewer fixed, isolates the effect of each signal on the chance a stay ends in one or two stars. The dashboard’s priority list rearranges itself.
Left: the dashboard’s five flagged drivers, in its own priority order. Right: the same five, ranked by their true causal effect on a one-star stay. Follow the front-desk line.
Front desk and check-in, the dashboard’s number-two priority, has a causal effect of two points, and it is not significant. It looked important because guests mention it constantly, and it rides along with long, unhappy reviews. A team acting on the dashboard would have spent a budget on the front desk and moved nothing. The real driver, cleanliness, is a 34-point cause with airtight confidence. This is the difference between counting and knowing, and no amount of survey volume closes it.
Each estimate is the effect of a signal on the chance of a one- or two-star review, separated from everything that travels with it and stress-tested three ways.
A descriptive tool told Marriott to fix the front desk. A causal read says loyalty is won by the people the survey never counts. Same data. Opposite instructions.
Run the model on the other outcome, a five-star review, and the deepest gap appears. The levers a hotel manages hardest, price, location, and fees, are the weakest ways to earn a loyal guest. The strongest by far is a single human being, the one thing no dashboard isolates.
The causal effect of each signal on a one-star review (left) and a five-star review (right). A person is the strongest lever on both sides; price and location are the weakest.
A good price lifts the odds of a five-star review by six points, against forty-two for an employee who goes beyond. Your tools’ driver analysis reports correlation, so it would never surface this: on the dashboard, everything a happy guest mentions correlates with a high score. Only a causal model can rank a human moment above a discount and put a number on it, because it asks what would change if you changed the cause.
This is not a knock on your tools. It is a knock on stopping at the dashboard. Descriptive intelligence is the first half of the job. The causal layer is the half that tells you what to do.
| The question a CX leader actually asks | What the dashboard answers | What a Causal Brief answers |
|---|---|---|
| Are we doing well? | NPS +32, CSAT 68%, flat for 18 months. Status: healthy. | The same, and the reason it is flat: strong stays and broken ones are cancelling out. The average hides the business. |
| What should we fix first? | The top-mentioned drivers. Front desk and check-in are flagged priority #2. | Cleanliness, which causally raises one-star risk 34 points. Front desk is a statistical mirage: near-zero effect once you control for context. |
| Is that driver real, or a coincidence? | No way to tell. The dashboard shows correlation, and calls it a driver. | Every effect is stress-tested (placebo, random-cause, subset). What survives is reported with a confidence rating. |
| What happens if we fix it? | No answer. A dashboard describes the past. | The what-if, quantified. Closing the review-score gap is worth roughly 5% on the average room rate (Cornell), in the highest-demand market in the country. |
| What actually earns loyalty? | Whatever happy guests mention: everything correlates with a high score. | A person who goes beyond (+42) beats a good price (+6) seven to one. Location and fees barely move it. Fees are pure downside. |
| Where is the money? | Not in the dashboard. CX and finance live in separate systems. | Fused with public financials: the resort fee books revenue today and trades tomorrow's rate, borrowed against the 271M-member loyalty base. |
The same causal read runs at every hotel. The top of this table is where a market leader acts first; the bottom is the internal proof set of what good looks like. Your CX stack has this data too. Its dashboard reports the column. It cannot produce the cause.
| Hotel | Tier | Reviews | Avg star | Poor-stay % | Flag | |
|---|---|---|---|---|---|---|
| Sheraton Orlando Lake Buena Vista Resort | Full-Service | 340 | 2.917 | 41.18% | watch | |
| Courtyard by Marriott Orlando LBV in the Marriott Village | Select-Service | 199 | 2.892 | 40.70% | watch | |
| Delta Hotels Orlando Celebration | Select-Service | 531 | 3.296 | 35.78% | watch | |
| Sheraton Suites Orlando Airport | Full-Service | 166 | 3.125 | 35.54% | watch | |
| SpringHill Suites Orlando Airport | Select-Service | 135 | 3.281 | 30.37% | watch | |
| Residence Inn Orlando at SeaWorld | Extended-Stay | 222 | 3.331 | 29.73% | ||
| Renaissance Orlando Airport Hotel | Full-Service | 149 | 3.545 | 26.17% | ||
| Residence Inn Orlando Airport | Extended-Stay | 94 | 3.598 | 25.53% | ||
| Castle Hotel, Autograph Collection | Luxury | 256 | 3.647 | 25.00% | ||
| Sheraton Orlando North Hotel | Full-Service | 288 | 3.478 | 24.65% | ||
| Fairfield Inn Orlando at SeaWorld | Budget | 165 | 3.551 | 24.24% | ||
| Fairfield Inn Orlando Kissimmee/Celebration | Budget | 104 | 3.567 | 24.04% | ||
| Fairfield Inn Orlando Airport | Budget | 197 | 3.552 | 23.86% | ||
| Renaissance Orlando at SeaWorld | Full-Service | 398 | 3.695 | 23.62% | ||
| Fairfield Inn Orlando Lake Buena Vista in the Marriott Village | Budget | 315 | 3.269 | 23.49% | ||
| Four Points by Sheraton Orlando International Drive | Select-Service | 323 | 3.579 | 22.60% | ||
| SpringHill Suites Orlando at Millenia | Select-Service | 101 | 3.750 | 21.78% | ||
| Marriott Orlando Downtown | Full-Service | 190 | 3.873 | 21.05% | ||
| Residence Inn Orlando Convention Center | Extended-Stay | 77 | 3.767 | 20.78% | ||
| Residence Inn Orlando Lake Buena Vista | Extended-Stay | 249 | 3.852 | 20.48% | ||
| Courtyard by Marriott Orlando International Drive/Convention Center | Select-Service | 91 | 3.658 | 19.78% | ||
| Orlando World Center Marriott | Full-Service | 653 | 3.944 | 19.60% | ||
| SpringHill Suites Orlando at Flamingo Crossings | Select-Service | 219 | 3.863 | 19.18% | ||
| Fairfield Inn Orlando International Drive/Convention Center | Budget | 164 | 3.647 | 18.90% | ||
| Grand Bohemian Orlando, Autograph Collection | Luxury | 272 | 4.043 | 18.38% | ||
| SpringHill Suites Orlando Convention Center/International Drive Area | Select-Service | 135 | 3.903 | 17.78% | ||
| Element Orlando Universal Blvd | Extended-Stay | 177 | 4.111 | 15.82% | ||
| Marriott Orlando Airport Lakeside | Full-Service | 300 | 4.153 | 15.33% | ||
| TownePlace Suites Orlando at Flamingo Crossings | Extended-Stay | 190 | 3.980 | 15.26% | ||
| Courtyard by Marriott Orlando South/Grande Lakes Area | Select-Service | 86 | 3.921 | 15.12% | ||
| AC Hotel Orlando Downtown | Select-Service | 166 | 4.081 | 14.46% | ||
| The Westin Lake Mary, Orlando North | Full-Service | 236 | 4.307 | 14.41% | ||
| Fairfield Inn Orlando at Millenia | Budget | 107 | 4.147 | 14.02% | ||
| The Ritz-Carlton Orlando, Grande Lakes | Luxury | 325 | 4.316 | 13.85% | ||
| JW Marriott Orlando, Grande Lakes | Luxury | 369 | 4.210 | 13.82% | ||
| Courtyard by Marriott Orlando Airport | Select-Service | 101 | 4.126 | 12.87% | ||
| Aloft Orlando Downtown | Select-Service | 167 | 4.325 | 12.57% | ||
| TownePlace Suites Orlando Airport | Extended-Stay | 125 | 4.374 | 12.00% | ||
| Courtyard by Marriott Orlando Downtown | Select-Service | 138 | 4.294 | 10.87% | ||
| Residence Inn Orlando at Flamingo Crossings Town Center | Extended-Stay | 172 | 4.435 | 10.47% | ||
| Residence Inn Orlando Downtown | Extended-Stay | 87 | 4.278 | 10.34% | ||
| Courtyard by Marriott Orlando Altamonte Springs/Maitland | Select-Service | 68 | 4.154 | 10.29% | low n | |
| Fairfield Inn Orlando at Flamingo Crossings Town Center | Budget | 203 | 4.400 | 9.85% | ||
| JW Marriott Orlando Bonnet Creek Resort & Spa | Luxury | 320 | 4.513 | 9.06% | ||
| TownePlace Suites Orlando Downtown | Extended-Stay | 76 | 4.460 | 7.89% | best |
A dashboard is built on the 10% of your data that is already structured: scores, counts, timestamps. The other 90%, what your customers actually said, sits unread. Dimension Labs turns that language into analytics in three steps, and every number in this brief came out of them.
Through Text Dimensionality, we read each conversation, a review, a ticket, a survey verbatim, a call, and enrich it into a structured record. Here that meant extracting 23 dimensions from every one of 9,446 reviews. Unstructured language becomes data you can query. The dictionary below is that Meaning Layer.
The dimensions are categorized in the operator’s own language, so a signal like a surprise resort fee routes to the lever that owns it, and a named employee becomes a measurable driver rather than a comment.
With the Meaning Layer in place, the Causal Correlation Engine applies formal causal inference, the DoWhy models in this brief, to separate a driver from a coincidence, and returns a confidence-rated effect. Not what happened. Why.
All 23 dimensions are extracted in a single enrichment pass over each review, then queried like any other column. That table is what the causal engine runs on.
Marriott is one company. There are a thousand more paying for the descriptive half and living with the gap: a program that says you are healthy, points you at the wrong fix, and cannot tell you what any of it is worth. If that sounds like your dashboard, it is because the gap is structural, not a Marriott problem.
Dimension Labs reads the open-ended voice your customers already give you, survey verbatims, support tickets, reviews, and turns it into causal answers: the driver, the proof, the payoff, the what-if. On top of the platform you already own.
Dimension LabsCausal IntelligenceEvery figure in this brief is derived from 9,446 public Google reviews of 45 Marriott-family hotels in the Orlando metro, January 2025 to July 2026, deduplicated to one row per review. No proprietary or survey data was used. Marriott’s guest-feedback program, guestVoice, is publicly documented as built on Medallia; that is the basis for the example, and the critique is of the descriptive survey approach generally, not of any single vendor’s software.
The descriptive dashboard is reconstructed from the reviews using the standard methods a survey platform applies: an NPS proxy (share of five-star minus share of one-and-two-star), a CSAT proxy (share of four-and-five-star), and a driver ranking by mention volume and correlation with low scores. The causal estimates come from a DoWhy model on the 8,504 reviews carrying a star rating, with treatment defined from the review text, outcome the star, and controls for hotel, tier, language, review length, and reviewer type, cross-checked with propensity-score weighting and stress-tested with placebo, random-common-cause, and subset refutation. The room-rate figure applies a published reputation-to-price elasticity (Cornell Center for Hospitality Research) to the measured review-score gap; it is illustrative, not a Marriott figure. Public company facts (Marriott FY2025 results, the FTC fee rule) are cited to their sources and never merged with the review estimates.