Google opens up media mix modelling with Meridian Launch

This week, Google launched its open-source media mix model tool, Meridian, to everyone.

Meridian uses advanced modelling to help optimise your ad budget insights and make spending more efficient. The ability to plug in Google Ads data and other data sources using Bayesian causal inference methods will help marketers better understand how their campaigns impact the performance of all channels.

Historically, media mix models (MMMs) have often focused primarily on offline advertising and missed the curve on more advanced online strategies. Meridian has been designed to fix that issue and offer a more comprehensive way to measure marketing success in the long term. It combines online and offline data, such as how a TV ad has or could impact your online campaigns.

MMMs also help marketers move beyond traditional conversion metrics in reporting to measuring the impact campaigns can have on an overall business.

One of the key features of Meridian is being open source, and that means marketers can take the code as it is and, if needed, customise it to better meet their individual needs. For example, if you’re working on a client which you know has seasonal trends which differ from the average market ups and downs, you can bring that data into the mix to increase the accuracy of the model. Being built by Google also means it has native connections to all of the data from its platforms that you would expect.

MMMs are not new, but this is a significant step towards the future of campaign measurement and also a vital one when you consider the impact privacy laws and cookie restrictions have had on traditional analytics tools over the past few years.

We spoke to the Founder and CEO of the marketing analytics agency ELIYA, Saeed Omidi about his thoughts on Google’s new update:

Expert opinion

Saeed Omidi, Founder and CEO of ELIYA

"Google introduced Meridian as an enhancement to its open-source MMM model, LightweightMMM. Both Meridian and LightweightMMM are based on a similar Bayesian regression model that integrates media channels, seasonal trends, and various external factors. Being a Bayesian model, Meridian allows for calibration through prior distributions, which is particularly beneficial for integrating experimental outcomes like incrementality testing. Moreover, Meridian can leverage reach and frequency data rather than just impressions, enabling it to capture the media impact accurately. This capability is especially important when individuals may be exposed multiple times, with differing effects based on frequency. Meridian is developed using Google's TensorFlow Probability, which provides a way to accelerate the model's execution on GPU hardware. While this is advantageous for processing numerous models simultaneously, it may be excessive for a business with a single MMM and could incur additional costs. Additionally, Google has access to highly accurate query and keyword data. When applied to Meridian, this data transforms GQV into dependable control variables, thereby improving the measurement quality of the Paid Search channel."

We also spoke to Himanshu Sharma, Founder of Optimize Smart:

Expert opinion

Himanshu Sharma, Founder of Optimize Smart

"Meridian is cool because it takes a lot of the manual work out of marketing mix modeling, making it faster and easier to use. But here is the thing, you can’t just trust it blindly. AI models aren’t perfect, and if your data isn’t solid, you will get junk results. Plus, since it is based on historical trends, it might not handle sudden market changes well. It is a great tool, but it is not a magic bullet. You still need real strategy and common sense to get the best out of it."