This is what we’ve figured out, so you don’t have to.
Meridian uses a Bayesian model that incorporates knowledge of prior media performance to provide insights into ROI and KPIs, and optimize budget allocation. It also takes non-marketing factors into account. It looks a lot like the models we use.
Bayesian modeling for MMM isn’t new…in fact, it’s already open source:
PyMC has had an open-source programming library for making Bayesian models available for a while, and Meta did the same thing with Prophet. Google has used another open-source implementation, TensorFlow Probability, instead of PyMC, but it’s based on similar concepts. That’s not a knock against Meridian – this code is great! But it’s worth noting that Google has taken open-source materials that already existed and reframed them as a shiny new thing that Google created.
Meridian isn’t an out-of-the-box solution.
It’s not a fully developed product. What it is, is open-source code that can be adapted and implemented by a data scientist (or a team of data scientists). This is important to recognize – a lot of people may start talking about Meridian as a holy grail replacement to third-party cookies, but putting it into action will be labor intensive and require significant internal expertise and ability in both data engineering and data science.
There are a few key capabilities that Meridian doesn’t include (that we think are important):
For instance, we don’t yet know if it has (or will have) any sort of predictive calculator where a lay person can estimate the impact of next week’s budget. The documentation also doesn’t say if it can incorporate Milestones – big events that happen for your company, like price changes or show launches. And there are some customization things that it seems like it won’t be able to do.
There are some really cool things about Meridian:
Meridian can incorporate Google Search volume as a control variable when evaluating paid search:
Plus, it lets you track search trends. This is something really cool that only Google could develop, so we’re very excited to now be able to dig into it.
Meridian seems to be designed mostly for use with Google products:
Which makes sense! This is the case any time a platform does attribution – it will always be biased toward the platform that developed it. While it means that insights might be limited for non-Google channels, this also means Meridian might provide some deeper Google insights.
Google’s documentation introduces a lot of questions:
While Meridian is documented more thoroughly than some other Google products, and the explanations are packaged well, the information they’ve released doesn’t give many clues to what will eventually be available downstream. For example: how customizable will Meridian be? Will the geo-population feature actually be useful?
All of that said, we’re excited to see Google lean into an MMM, and to dig deeper into Meridian.
According to Google, 60% of advertisers are already using some form of MMM.
Are you in the 40% that aren’t using MMM? Feeling like you don’t know what’s going on? We wrote a primer on MMMs last month that might answer some of your questions.
