5,866 hotel ad creatives, and what actually went wrong in them
At volume the failures that survive are positional, not aesthetic: type on type, a logo with nowhere quiet to sit, a mirrored currency symbol.
3 min read
Production volume to date.
| Figure | Value |
|---|---|
| Creative images rendered | 5,866 |
| Ad variants | 1,572 |
| Campaigns | 340 |
What the defects were
At this volume the failures that survived to the final week were almost entirely positional rather than aesthetic. None of these are matters of taste and all of them are detectable mechanically.
- Text landing on other text.
- A logo placed over a busy region of photograph, with no quiet area to sit in.
- A currency symbol moving to the wrong side of its own number when a layout mirrored for Hebrew.
- A badge printing across legal fine print.
How it was measured
Counts are of images actually rendered and shipped into campaigns across the production system, not of drafts or previews. A variant is one creative in one aspect ratio bound to one ad; a campaign may hold many variants.
Defect classes are drawn from the failures logged in the final week of the period rather than from the whole history, so they describe the state after the obvious problems had already been fixed.
Why the volume matters more than the defect list
Fixing any single misplacement is mechanical. What volume demonstrates is the compounding effect of fixing them the same way each time.
A guest sees an advert for roughly a second and a half, most of it before reading, and what registers is where the type sits, how much space surrounds it, which corner holds the logo and whether the palette matches what they saw last week. Consistency across that surface is the only component of advertising spend that carries from one month into the next.
Limits
These are our own production counts across many properties, not an industry measurement, and the defect classes reflect the specific pipeline that produced them. A hotel using templates would see a different distribution of the same underlying problem. No customer data is involved: these are counts of images we rendered.