Social Rocket
21.000 AI-generated marketing assets – and an in-house system to check their quality and permissibility: The health brand group Opella published new figures from its AI production on 20 October 2026. Together with Microsoft, the company is developing a review process under the name Lumen with 14 specialised AI agents. The approach shows how work in marketing departments is changing: it is no longer the first image idea that becomes the bottleneck, but its reliable approval.
A new production structure lies behind 21.000 assets
Opella organises production into six internal content units called Wonka. According to the company, more than 85 people work there. In the official announcement of 20 October, the company reports more than 21.000 AI-supported marketing materials, roughly 30 times higher production output and approximately 50 percent more engagement. These figures are the company's own information from its internal processes.
The mechanism matters more than sheer volume: brands can produce formats, variants and local-language versions much faster. At the same time, the number of versions that someone must check for medical claims, sources, mandatory information, design and approvals grows.
Why Opella is building a second AI layer
This is exactly where Lumen comes in. According to Opella's description, 14 specialised review agents are meant to examine content for, among other things, scientific claims, evidence and legal requirements. The platform is being built with Microsoft; the build-out has been announced, not presented as a fully rolled-out system. The final expert decision is to remain with humans.
This is especially clear in health advertising: an imprecise sentence in an advertising motif can cause more problems than a weak image. The same principle applies on a smaller scale to other industries – for example to price promises, product comparisons or discount information.
The practical consequence is a new content workflow: approved facts and brand assets at the start, AI variants in the middle, automated checks and responsible human approval before publication. We have already looked at why human review of AI content remains decisive.
Wyndham shows the social media side of the model
That this is not only about pharma or health marketing is shown by a second example. According to Digiday reporting of 2 October, the hotel group Wyndham uses AI tools for its own and paid Instagram content. The starting point is real images of its properties. The company reports 15 times as many creative variants and a 75 percent shorter time to campaign launch.
For social teams this creates a concrete advantage: instead of using a single ad for many audiences, existing, verifiable brand motifs can be adapted to language, format and placement. Which ads work afterwards must be shown by testing.
Meta too is advancing AI automation in Ads Manager. The platform side optimises delivery; the brand must decide which claims and images it approves at all.
What German marketing teams can take from this
A sensible start doesn't need 14 agents. More useful first steps are a central library of approved statements and image rights, clear release responsibilities, and automated checks for outdated prices, impermissible claims, or missing labeling. After that, it's worth linking variant production to real campaign data.
For Meta ads, the company's own responsibilities for AI creatives also belong in quality control. Our article on Runway Agent 2 describes how production tools can respond more strongly to performance data.
The new metric in AI marketing is therefore not simply "assets per hour." It becomes more interesting to ask how many brand-compliant, verified, and actually deployed variants a team can produce economically.
Sources: Opella and Microsoft, 20. October 2026 · Digiday, 2. October 2026