There is a myriad of reasons to give up on digital attribution and move to Marketing Mix Modelling (MMM). If Google and Meta are offering MMM solutions, then MMM must be the right way to go, no? However, it’s not as straightforward as that. The forward-looking question is which is the best option to choose to create, use, update your MMMs. And then bonus marks are given on the ability to integrate MMM into the workflow for use by all the internal and external stakeholders.
There are four broad buckets to choose from:
- The Manual Approach, do it yourself, get a team of data scientists and turn them loose on MMM and other data projects,
- The Provider Approach, find a turnkey solution, send them your data, and then,
typically months later, receive a presentation of the results, - The Open-Source Approach, choose from Meta’s Robyn, Google’s Meridian, Uber’s
Orbit, hire a team to download, customize and operate, - The Platform Approach, choose a platform that’s easy to operate and do it yourself.
One might start with the question what the best Marketing Mix Model? A simple analogy is you want to go sailing but you don’t have a boat. Before asking “what is the best boat?” ask how much time you want to spend gluing cedar strips and painting versus paddling or fishing. The Manual Approach is you’ve decided to build a boat and will hire a bunch of craftsmen to design, build, test and sail the boat. The Provider Approach is to charter a boat and crew. The Open-Source Approach is you’ve downloaded some free plans off the internet and now you and some buddies are going to assemble (with some mods) and sail the boat yourself. The Platform Approach is you’ve bought a boat, and you are your own Captain.
The primary rationale to use MMM’s is for a neutral evaluation across all media and marketing while including the influence of non-media and competitive factors into the equation. Let’s review what an MMM must provide.
Principles of Marketing Mix Models:
1. All media is treated equally, transparently, and customizable
2. Scope realistically reflects the media, market, and competitive situation
3. Provides usable reports and data that are timely and insightful
4. Flexible capabilities from very simple to very complex models
Google’s MMM product, Meridian, is very Googley. Their strategy is to launch new products with integrations with other Google products. It makes a bridge between the media evaluation and the media products, as long as you continue to think the Google way. It’s sort of like buying an odd sized boat that only fits OEM trailers, motors, and other accessories. On top of that, it doesn’t get you to all the good fishing spots except Google Bay.
The point of MMM is to evaluate all media. If the MMM treats one medium in a favourable, non-transparent way that can’t be revised, that compromises its core competency. Harsh news for Meridian that seems to treat all Google media products as one channel with some kind of machine learning optimization applied to Pmax. The more objective option would avoid an integration and customize how the MMM defines each channel as the advertiser wants to see it. Plus, extend the value of MMM through geo-analysis, again supporting media buyer control over data input/reporting output structure instead of an opaque integration.
Central to MMM is the concept of adstock. Each ad impression adds to the overall campaign while also decaying in time. Not all ad impressions are the same in terms of impact, decay length or steepness of decay curve. How all these variables are defined per medium, or media supplier is called adstock. Paid search and Youtube should possess very different ad stock variables and be defined by a neutral operator whose goal is to maintain the neutrality of the MMM, not direct more media spend in certain directions. Most critically, Adstock variables should not be defined by a media seller even if they are adjustable after the fact.
The basis of MMM is a backwards looking analysis of inputs and output over a long period of time. For digital marketers used to 7-day cookie-based lookback windows limited only to digital media tagged with identifiers, MMM is totally different. MMM works best with broad data that includes all media types, non-media factors and competitive pressures across 2 years. That scope gives a view of seasonality (e.g., Christmas x2), trade promotion schedules, and uncontrollable environment factors like covid, interest rates inflation. Google’s Meridian seems to lack these critical capabilities.
Who produces the MMM is often a choke point preventing wide use of the results. For instance, if the media planner can calculate the Return on Ad Spend (ROAS) on a channel that they are responsible for planning and buying, the likelihood is they will make changes to the spend and come up with new tests to drive higher effectiveness vs a person whose sole job is to calculate ROAS. If the hurdle is a requirement to learn Python or R to operate the MMM, that hardly “democratizes” the Market Mix Modelling function.
In summary, Google and Meta offering Mix Modelling is an endorsement of the methodology; boats are good for fishing. Is the Open-Source approach the best way to get an MMM up and running in your company or agency? Pull up a Muskoka chair, survey the lake and start counting the number of boats built from kits that fish only in one spot.