Ravina Square — The Audit Found a Site Advertising a Product That Did Not Exist

The brief was to connect a Meta ads account and help with targeting. The first API call answered the actual question: the business had zero ad accounts, zero pixels, and a Facebook page with three followers. There was nothing to analyse. Everything downstream of that assumption had to change before any work was worth doing.

So the first deliverable was a competitive picture rather than code. The Meta Ad Library is public and needs no permissions, so I swept every advertiser running in the catchment and read what they actually say. Ten direct competitors, and the pattern was immediate: every single one leads with the down payment and the instalment term before mentioning the building. Two to thirty-five percent down, three to twelve years. Nobody leads with the property. That one observation reframed the entire brief.

The second finding was the one that mattered commercially. The live homepage advertised units from 15 to 45 square metres. The brochure lists 17 to 52, and the largest unit in the building is a single 52 square metre ground-floor space. It promised delivery in a year that had already passed, on a project still at excavation. It implied prices increase at construction milestones, for a developer who had deliberately decided not to publish a price at all. And the static head still carried the previous project name in five places, so every share card and every pre-JavaScript crawl showed the wrong brand. Paid traffic was about to land on a page describing a different building.

The tracking work is where the engineering went. Browser pixel and Conversions API both fire, sharing one event ID so Meta collapses the pair rather than counting a conversion twice. Server events run inside waitUntil, after the response has already been returned, so a tracking failure cannot add latency to the lead form or turn a saved lead into an error message. Phone numbers are normalised before hashing, which sounds trivial until you notice the site defaults to Arabic and JavaScript treats Arabic-Indic digits as non-numeric: without folding them first, every Arabic-entered phone number silently hashes to nothing and matches nobody. Gulf numbers now return null instead of being coerced into Egyptian ones, because a fabricated number still hashes, still transmits a real person's data, and still matches zero people.

The audit also found a conversion event firing from thirty-five different URLs. Every unknown path and all thirty-four project slugs in the sitemap rendered the homepage, and the homepage fired a "viewed this project" event. Any audience built on that signal would have been mostly people who scrolled past a banner. It is now scoped to the two URLs where it means something.

The product page replaced a route that silently fell through to the homepage. It carries all seventy units across four floors as real tables with captions, row headers and sticky column headers, correct in right-to-left because Arabic is the default locale. The unit schedule does the qualifying work a price normally does: publishing a 17 to 52 square metre range means anyone hunting two hundred square metres self-selects out before they ever message the sales team.

One measurement is worth recording because it would never have shown up in review. The reveal animation uses an IntersectionObserver at a fifteen percent threshold, and a tall element can never reach that threshold if the viewport is shorter than fifteen percent of its height. All four unit tables in one wrapper is 3,610 pixels, which needs a 542 pixel viewport. A 320 by 568 phone with the browser toolbars gives about 480. The entire seventy-unit schedule would have rendered at zero opacity on small phones, with no error and nothing in CI to catch it. Measured in a headless browser, then split into one reveal per floor table.

Finally, the outbound list. Seventy-two prospects merged from the client's own market research and the live ad sweep: thirty brokers already selling in the catchment, and forty-two businesses currently paying rent inside it and running their own Meta ads, which is a reliable signal of both budget and a working phone. Split into brokers, who are being sold a way to earn, and occupiers, who are being sold their own next branch. Those are different conversations and the same script fails both.

Worth stating plainly: there is no campaign performance data in this case study. At the time of writing no money has been spent. What shipped is the stack and the intelligence to aim it, plus a set of live copy corrections the client would otherwise have paid for in wasted clicks.