You are Camille Aubert, creative director of Lumen & Co., the in-house studio of a Paris cosmetics group. The chief marketing officer wants the relaunch of Aurore, an €88 million skincare line, shipped across six channels in five weeks on €1.8 million. Before you split the work between your people and your artificial-intelligence tools, you have to name the task where getting that split wrong would do the most damage.
The situation
Monday 4 May 2026. Florette Beauté turns over €620 million at a 13% operating margin, and Aurore, its flagship line, is losing ground to a direct-to-consumer brand that has taken six share points in eighteen months. The brief: six channels, five weeks (40% less than usual for a campaign this size) and a fixed production budget of €1.8 million. That morning the studio dashboard shows projected delivery at 8.0 weeks, distinctiveness at 70 against a baseline of 100, expected audience engagement at 45 out of 100, and not a euro committed yet.
Lumen has 62 people, about 140 campaigns a year and a three-person data-and-AI cell set up last year. Its tools come in two families: generative ones, which produce copy, images and variants, and non-generative ones, which analyse audiences and predict performance without creating anything a customer sees. You championed them. Last spring the launch that leaned on them most tested 18% below the brand's distinctiveness baseline: polished and forgettable, the board said. And last quarter a generated visual came close enough to a rival's hero image to need a legal review before print.
Aurore's work splits into very different pieces: the core concept and brand voice; the production of hundreds of variants and resizes for each channel; the localisation of copy and assets market by market; and the support work, from predicting which audiences will respond to tagging the asset library. Doing it all by hand would cost about €2.4 million and miss the date, so part of the work will go to the machines.