From the noise of El Dorado to the science of conversion

Before I built stores and software, I was, and still am, an environmental engineer. My undergraduate thesis at Universidad Distrital asked a question with a familiar face in Bogotá: what has been researched this century about the noise from El Dorado airport and the health of the populations in its area of influence. I wrote it with Ana Lucía Ramírez Montes, under the direction of Jaime Alberto Moreno, and it took me close to a year.

Today I design pages so they sell. It looks like a huge leap, but it is the same craft: putting evidence in order before forming an opinion. These are the four lessons I brought back from that research↗, and the ones I use every time someone asks me why their site is not converting.

First, the method, which is what you actually inherit

We did not go out into the street with a sound-level meter. It was a qualitative documentary review: searching, reading, and organizing academic and institutional sources published through 2021. The heart of the work was a matrix of 88 documents, each one described across ten columns: source type, subject area, year, title, authors, key content, contribution to the research, references of interest, and link. Out of those 88, we prioritized 28 studies focused specifically on El Dorado.

Filling in that matrix forced us to answer, for every single document, what it actually contributed to the question. Copying the abstract was not enough. That step, which sounds like paperwork, is exactly what separates a pile of PDFs from an actual investigation.

Lesson 1: not every number is worth the same

Part of the method was looking at which instrument produced each data point: sleep quality with the Pittsburgh index, daytime sleepiness with the Epworth scale, and hearing with audiometries read using the extended ELI and Larsen indices. Knowing that let us separate prevalence studies, which measure how many people are affected, from documents that only describe effects. That is not a small detail: 35 of the 88 documents were institutional, built for purposes different from a scientific study.

The exact same thing happens on a website. What people say (a survey, a kind comment) is one thing, and what they do (how far they scroll, what they tap, what they abandon) is another. That is why, in the URG Technology↗ panel, each of its 31 metrics comes with a button that explains what it is, why it matters, and how it is calculated. If you cannot explain how a number is calculated, do not use it to decide anything.

Lesson 2: the yardstick you measure with changes the conclusion

One of the thesis's most-cited conclusions is a subtraction. There are 10 decibels of difference between the World Health Organization reference we used (65 dB, the threshold its own table associates with a higher cardiovascular risk) and the 75 dB that Resolución 0627 of 2006 sets as the daytime maximum for industrial-use zones, which is how airports get evaluated. In a quiet, moderate-noise zone, the cap is 65 during the day and 50 at night. Depending on which standard an author uses, the same noise can look like a health problem or something tolerable.

And even against the higher bar, the rule gets broken: in Fontibón, in 2019, 17% of daytime measurements and 39% of nighttime ones exceeded it.

Translated into conversion terms: define what counts as success before you measure. A click on the WhatsApp button? A submitted form? A closed sale? At La Roca, its own panel counts clicks on the WhatsApp button and the Google Maps button, because that is where that business actually closes deals. For this site, the bar is stricter, and we chose it on purpose: at least one project sold through the website. Changing the bar changes the grade, so it pays to set it before the exam, not after.

Lesson 3: more exposure, more effect (and still, watch out for causation)

Across several of the studies we reviewed, people more exposed to the noise came out worse off. In studies done in Fontibón, poor sleep quality showed up in ranges from 60% to 92.08% of the people surveyed. Hearing loss reached at least 43% among the exposed, versus 15% among the unexposed (Londoño et al., 2004). And at night, each extra decibel was associated with 18% more sleep disturbances in Engativá and 6% in Fontibón (Rivera, 2008).

Here comes the part that stuck with me the most. The authors of the 2004 study themselves warned that, given its design, they could not prove with certainty that the difference was caused by air traffic, and the thesis recommends running prevalence studies for most of the health conditions identified. In other words: there is a strong, consistent association, and you still have to say out loud what the evidence does not let you claim.

I see the commercial version of that same trap all the time. Sales go up after a redesign and someone concludes it was the redesign. What if it was the season, the new ad campaign, or a price change? That is why we measure before touching anything, change one thing at a time when we can, and measure again to check whether it actually got better.

Lesson 4: naming your limits also sells

An honest piece of research devotes a chapter to what it could not prove. We carried that habit into this site's case studies: we say that URG is built to 95% and has not launched due to budget for photography, marketing, and ads, and that at La Roca the website alone needs marketing to grow. It is uncomfortable to write, and it does two things: the client knows exactly who they are talking to, and we do not sell what we cannot back up.

That is why, when we apply conversion psychology, the rule is the same: real social proof, never invented, and honest urgency. If a team is stretched thin, we say so. If a delivery promise is two business days, we write the actual date. Trust is built the same way evidence is: by showing where every claim comes from.

What would take weeks today

That thesis took close to a year, and not for lack of method, but because of the sheer manual labor: finding each document, reading it in full, filling in ten columns for every one of the 88, classifying them, counting by hand how many mentioned each health condition, and building the analysis. Today, with well-directed artificial intelligence, I estimate a review like that would take weeks.

What AI speeds up is the slow part: searching and filtering sources, pulling data out of each document, counting mentions, proposing a first classification. What stays human is what adds the value: deciding the inclusion criteria, telling a prevalence study apart from a descriptive document, not mixing figures from different samples as if they were one, and checking every number against its own page.

From noise to conversion, the distance is smaller than it looks: both are questions about people that refuse to be answered with opinions. If you have a decision hanging on scattered data, that is how we approach user experience and conversion, and every figure in this article is referenced, chapter by chapter, in the full story of the research↗. If you want us to take a look at your numbers, write to us.