Experimentation, MMM & Multitouch Attribution

I would agree completely with Dentsu Media’s UK & Ireland CEO Jenny Bullis’ comments in her recent keynote speech at the Media Research Group’s annual conference that increased experimentation is essential to drive marketing performance.

Her concerns about the ability of analytics tools to inform marketers with improved multi-touch measurement however demonstrates that cutting edge tooling which uses the latest advances in measurement technology and, in particular the use of deep learning AI, do not take into account the advances in AI-based attribution analysis in recent years.

Today, by moving away from decades old technologies – specifically cookies – and instead embracing the transformative power of AI path stitching, we can radically improve the quality of the underlying data used in attribution modelling to deliver high quality and accurate customer journeys that move across devices and ultimately through our marketing mix towards conversion with complete confidence.

And we don’t have to trust that the AI “just works” either; we can test the accuracy of its outcomes in the most tangible way possible – new customer acquisition and increased revenue generation.

For example, Corvidae AI – the leading AI path stitching technology brought to market last year in the UK – allows you to include stitching AdTech identifiers, such as Google’s GCLIDs, into its journeys to feed back into Google’s own bid algorithm, delivering savings of 45% on the cost of new customer acquisition.

Building on Solid Foundations

The underlying value of analytics data is at question when we look at attribution. Innovative approaches like Corvidae’s improve on the foundational data that is used for insight and optimisation across all digital activity and give radically different outcomes.

In short, looking at sophisticated attribution models reveals the poor quality in the underlying cookie-based measurement data provided by popular analytics platforms such as Google’s GA4 and Adobe Analytics. Cookie-based data essentially only represents the journeys of individual devices, and doesn’t show the relationship across devices that are caused by the actual person behind the device screen – the very person that advertisers are attempting to understand.

MMMs Have Value But are Not a Panacea

Jenny Bullis suggests that as marketers we embrace Media Mix Modelling (MMMs) and forget about attempting to introduce accuracy and rigour into our measurement.

MMMs are a valuable part of the mix for larger advertisers, and indeed Meta’s release of sophisticated AI MMM modelling – their ‘Robyn’ model that anyone can deploy, is an asset that was previously unthinkable for marketers.

However MMMS still use the broken underlying cookie based data that Jenny is decrying. And indeed her message that we should simply A/B or Multivariate test our way to campaign level insight misses the point of attribution: what data are we testing? If it is cookie-based analytics, we are not crediting the early touchpoints that potential new customers have arrived by; and if we use the AdTech data itself we also know that they are manipulating the data to preference their own performance – Google has openly advised that is weights Paid Search traffic more highly in its platforms because “cookie-based measurement fails to track across devices”.

Media Mix Models have their place, but are no substitute for accuracy in measurement, the best advice today is to embrace new technologies that solve decades old issues and leave the cookie behind.