Revenue-Prioritized Actions: Why We Built an Algorithm That Knows Its Limits

In short

We'd rather have a tool that says "this isn't clear from the data I have" than one that invents a certainty it doesn't have. That difference is what lets you trust a recommendation instead of just taking it on faith.

The temptation to promise more than the data allows

When we began designing Mosana's review analysis, the natural temptation was to present it with the most appealing promise possible: "artificial intelligence that analyzes your review sentiment and tells you exactly what is happening." It is an approach that is convincing in a sales demonstration. The issue is that such a promise belongs more to marketing than to engineering: predicting a guest's score from the text with total precision, or "automatically" detecting the exact reason behind every instance of dissatisfaction with no margin of error, would mean selling a certainty that the nature of the data itself does not allow — as already explained when analyzing why a review's score and its text do not always tell the same story.

This is the specific reason we designed Mosana differently. There is a structural noise ceiling between a review's score and its text: ignoring it and presenting the prediction as a verified fact, rather than as an inference with a margin of error, would pass that problem on to your business without your knowledge.

What we learned analyzing thousands of reviews

While building Mosana we analyzed large volumes of real reviews from properties in Spain, cross-referencing score and text review by review. The pattern repeats over and over: some guests score high despite complaining in the text, and others score below the maximum without leaving any textual clue as to why. This isn't isolated noise or a data-quality problem — it's a structural feature of how people score subjective experiences.

A model that ignores this and treats the text as a reliable predictor of the score doesn't fail randomly — it fails systematically on the same type of cases: precisely the ones that matter most to catch, because they hide a real complaint behind a generous score, or silent dissatisfaction behind neutral text.

Our product decision: not faking impossible precision

With that evidence in hand, we decided to prioritize reliability over marketing impact: we don't promise to predict your score from the text, and we don't present any automated conclusion as a verified fact when it's really an inference with a margin of error. Where the data is ambiguous, Mosana flags it as ambiguous — we don't pick whichever interpretation sounds better in a demo.

We prefer to state "this points to X, with this level of confidence" rather than "this is X," even though the latter is more persuasive in a sales meeting. In the medium term, it is the only way the tool remains useful precisely in the case that does not fit the usual pattern — which is almost always the case that actually matters.

From isolated signals to revenue-prioritized action

Everything explained so far focuses on reviews, because that's where the contrast between score and text is clearest. But Mosana isn't limited to reviews: it also aggregates the feedback that never becomes a review — your own surveys, front-desk conversation, email, chat — together with your PMS data, and cross-references all of it with the discovery dynamic of the Billboard Effect, that is, with the traffic OTAs are already driving to your own website.

The result of that cross-reference isn't a score or an isolated metric: it's a set of concrete actions, prioritized by their estimated impact on your revenue. The same honesty principle applied to the score-versus-text contrast applies here: every proposed action states the confidence level behind that estimated impact, instead of presenting a revenue figure as if it were a guaranteed fact.

What the property gains from this approach

The practical consequence isn't a less useful tool, but a more reliable one. Trusting a system that states with total confidence why your score dropped, when that confidence is unfounded, leads to acting on the wrong cause, while the real problem remains unresolved until it surfaces again in a public review. A system that acknowledges its limits provides something that can actually be used to decide: what is clear, what constitutes a reasonable hypothesis worth investigating, and what still requires someone on your team to read the original review before acting.

An engine that adapts to your judgment, not the other way around

Acknowledging the limits of the data does not mean giving up decision-making power — quite the opposite. Mosana is a data-analysis and artificial-intelligence engine that you can calibrate across the full set of signals it processes (reviews, qualitative feedback, surveys, PMS data and Billboard Effect-linked traffic): through fully personalized templates and configurations, you decide which signal categories matter most for your business, which channels are the priority for your operation, and at what point a minor complaint becomes a pattern that calls for action.

That configurability exists because nobody understands a property's own guests, seasonality or regional particularities better than the person managing it. An urban, business-focused hotel prioritizes different signals than a family vacation property in high season, and two properties of the same type in different markets may require different alert thresholds. Mosana delivers the analysis; the direction given to that analysis — what to prioritize first and why — is decided by the team that knows the business from the inside.

The result: an auditable algorithm

We designed Mosana so that every conclusion can be traced back to the data that produced it — which reviews, which channel, which date — rather than being a black box that hands over a visually appealing dashboard with no way to verify where each figure comes from. It is not the promise that shines brightest in a sales demonstration. It is the one that can be audited, and the one that does not fail precisely when it matters most: when explaining to management why the score dropped, a number on its own is not enough — it must be possible to show where it came from.

Frequently asked questions

Closing the cluster: from signal to product

This post closes the journey that started with the Billboard Effect: OTAs bring you the traffic, your website converts it (or not) into a direct booking, and the reputation you build — properly measured, properly aggregated, and analyzed without faking a precision that doesn't exist — is what sustains that conversion long term. Every piece of the path matters; none replaces the others.