Fifty-four people in a church, in six days

The most expensive mistake was not in the media — it was in the plumbing. Four separate registration pages were live for one event: three Luma pages, including a second English page that no ad ever pointed at, plus a Hipsy ticket page. Traffic, social proof and retargeting pools were split four ways, and there was never a single number anyone could look at and call “registrations”. One canonical page with language variants behind it would have concentrated everything the campaign was trying to build, and would have made every figure on this page an order of magnitude more trustworthy. This is the first thing to fix for the tenth edition, and it costs nothing.

The second was the missing Meta Pixel. With no pixel on the destination, there was no audience of site visitors to retarget — the people who had already shown intent by clicking through were unreachable after the click. It also meant the campaign could not optimise for registration; Meta was asked to buy landing page views instead, which it did, very efficiently, at €0.11 a click and 2,911 landing page views for €375.77. But a landing page view is not a signup, and roughly two per cent of them became one. On a six-day runway with a warm local audience, retargeting the clickers would plausibly have been the cheapest registrations available, and the campaign simply could not run it. Installing the pixel and defining a registration event is the second free fix.

The Hipsy route was a deliberate test and it answered its question, just not in the way we wanted. Sending €49.73 to a Hipsy ticket page bought the cheapest clicks of the whole campaign — €0.07 a click at a 7.4% click-through rate — and seven tickets came out the other end. But Hipsy records no source against an order, so those seven are a count, not an attribution: we know they exist, we cannot prove the ads produced them. The lesson is not that Hipsy underperformed; it is that handing traffic to a platform that will not hand measurement back ends the experiment at the moment it gets interesting. TikTok taught a sharper version of the same thing. It delivered 83,750 impressions for €47.10 — a CPM of fifty-six cents, nine times cheaper than Meta — and not one registration on any Luma page carries a TikTok source. Cheap reach is not cheap attention, and on a six-day window for a physical event in one city, a platform with no local intent signal had no time to earn its place.

Finally, registrations and attendance are different things, and at a free event the gap between them is the whole commercial question. Sixty-six people put their name down across four pages and fifty-four walked in — a show-up rate of 82%, which for a free event with no deposit and no ticket is a good result rather than an average one. But that number exists only because the team counted heads: no check-in was taken in Luma, so nothing in any dashboard would have told us. It also moves the real media cost from €6.41 per registration to €7.83 per person actually in the room, which is the figure worth planning the next budget against. For the tenth edition the fix is operational rather than creative: take check-ins at the door, so the ratio between registration and attendance is measured rather than remembered. What did work, and is worth carrying forward, is the shape of the creative and the speed of the loop. A six-part game series gave an audience a reason to come back to the feed rather than one announcement to ignore, Meta and organic reinforced each other on the same “Come and play” line, and reading the dashboard every morning meant budget moved off the ad sets that were not converting while there was still a campaign left to spend it on.