Marketing Mix Modeling — measuring impact with Google Meridian
A multi-tenant platform for agencies and brands to measure which channel actually caused revenue — not who clicked last. It builds on Google Meridian, the openly published Bayesian MMM framework. Connectors consolidate media data from Google Ads, Meta, LinkedIn, DV360 and GA4 at weekly granularity; modelling and analysis run in the Google Cloud Zurich region.
Why attribution does not answer the question
Platform attribution follows individual user paths and assigns the conversion to a touchpoint. Add up the conversions reported by every platform and the total routinely exceeds what was actually sold — each platform claims the same sale, and each is an interested party. Cookie loss and consent rates have further hollowed out the data without making the reports look any less precise. Marketing mix modelling takes the other route: it works on aggregated time series and estimates each channel's revenue contribution without needing a click path. That captures channels which produce no click at all — outdoor, radio, TV — and works independently of cookies.
Meridian instead of our own black box
We could have built our own model. We build on Google Meridian instead because its assumptions are published and therefore checkable — the decisive difference from the platforms' closed attribution models. Bayesian here means every statement comes with the uncertainty band the data supports. Point-precise ROI figures without a range are always an invention where media effects are concerned, and a tool that emits them manufactures false confidence. Saturation curves and adstock beyond the campaign week are part of the model, not after-the-fact corrections.
Tenant isolation and data residency
Agencies serve several clients, and their media data must never mix. The platform is therefore tenant-isolated by design, with row-level security at the database layer rather than filter logic in the application. Processing and modelling run in Google Cloud region europe-west6 in Zurich, the application is served from Frankfurt — media data does not leave the European area. That matters for Swiss nDSG compliance and is a hard requirement in many tenders anyway. For our advisory work the platform is the proof that we can discuss model assumptions rather than feature lists — and that we say where a model reaches its limit.