Luigi.

Luigi is one of the largest fashion eshops in Greece. After seven years of operation it employs 45 people, runs an in-house photo studio and design team, and was presented as a success story at the largest Google Retail event ever held in Greece. By spring 2020, every textbook performance lever had already been pulled — landing page tests, audience refinements, creative rotation, bid strategy experiments. The easy wins were gone.

Deep analysis of the eshop data surfaced something subtle but massive: conversion rate could lift by up to 30% if every paid visitor saw only products that were in stock in their specific size. A T-shirt marked "in stock" is, for a size-M customer, effectively out of stock if the remaining inventory is only XL. The product appears available to the system, unavailable to the human — and the gap between those two states was silently killing CR on every fashion ad click.

Standard CRO work could not solve this. Landing page copy, checkout flow, trust badges, discount stacking — none of them address the root issue, which lives upstream in the product feed. The fix required rebuilding how availability is represented per user, then propagating that representation into every Meta audience and creative. That is not a CRO optimization; it is a feed engineering problem.

We rebuilt the Meta product feed with custom labels encoding size-level availability for every SKU. Instead of one flag per product ("in stock" / "out of stock"), the feed now carries per-size stock state, updated from the live inventory API every 15 minutes. The feed became the source of truth for all downstream ad logic — no more stale assumptions about what was actually buyable.

To match users to the right feed slice, we deployed custom cookies that capture each user's size preference from first-party signals: sizes they filtered by, product pages they viewed in detail, and past purchases. No third-party tracking, no iOS 14.5 fragility — just first-party data captured at the user's own interaction pace. After a small number of sessions, the system has a confident size hypothesis for the user.

On Facebook, audiences were restructured around the intersection of browsing size × product size × live stock. Instead of one ad set per product category, we ran targeted ad sets per size cohort, each pointed at the subset of the catalog that was actually purchasable for that cohort in that moment. Dynamic creative adapted to the cohort, and the CRO lever moved from landing-page tweaks into feed architecture — where the real leverage was.