Meta's Robyn and Google's Meridian are free, sophisticated and built by the companies they are measuring. The structural conflict in platform-built MMM tools is getting a closer look.
Using a platform's free MMM tool to evaluate that platform's performance is like asking your media agency to audit their own work. The incentive structure does not support an honest answer.
At Mumbrella360 last week, the session that generated the most uncomfortable recognition in the room was not about AI taking jobs or agency consolidation. It was about something more quietly damaging: the structural conflict in the free MMM tools that major platforms provide to help marketers model their own media spend.
Meta gives you Robyn. Google gives you Meridian. Both are free, both are sophisticated and both were built by the companies they are measuring.
The coverage gap problem
Marketing mix modelling works by attributing outcomes across all the media that influenced a purchase. The problem with platform-built tools is architectural. When you run Meta's Robyn, it models the channels you have given it data for. If your Google Ads, Trade Desk or programmatic spend is missing, the model fills the gap with the data it has.
Researchers tracking MMM outcomes have consistently found that models built with incomplete data systematically over-credit the channels they can see. The estimated bias range is 15 to 25 percent ROI over-attribution when coverage falls below 80 percent of total media spend.
The estimated ROI over-attribution bias in platform-built MMM tools when total media coverage falls below 80%
For the average mid-sized Australian business running two or three paid channels alongside organic, print or out-of-home activity, hitting 80 percent data coverage in a platform tool is genuinely difficult. The missing channels become invisible to the model and the modelled channels absorb their contribution.
The conflict nobody talks about
The conflict is not malicious. Meta and Google are not designing their tools to deceive. The problem is structural. A tool built by a platform to help marketers understand their media mix will, by its architecture, weight toward the data the platform has access to. It cannot weight what it cannot see.
Independent MMM vendors who charge for the work have no incentive to inflate the performance of any single channel. They earn on accuracy. Platform tools earn on continued adoption, which tends to be higher when the results are favourable.
What this means for Australian budgets
Australian businesses are increasingly sophisticated about channel attribution, particularly post-Apple privacy changes and the decline of third-party cookies. But there is a gap between attribution sophistication and MMM adoption. Many businesses running Robyn or Meridian are doing so because the tools are free and a platform representative helped set them up.
The question to ask is not whether the tool is producing numbers. It is whether the coverage of your total media spend is high enough that the model's output is trustworthy.
Independent MMM practitioners recommend treating any model with below-80 percent coverage as directional rather than decisional. A directional output helps you understand trends. A decisional output drives budget allocation. Using a directional tool for decisional outcomes is how budgets get misspent at scale.
The independent alternative
Investing in independent MMM is not universally justified. For businesses spending below $500,000 per month across channels, the cost of independent modelling may outweigh the improvement in decision quality. But for businesses at that scale and above, the platform conflict argument is strong enough to warrant a critical review of the tool's coverage assumptions before trusting the output.
The conversation is now at the industry level. The more useful version of it is happening inside individual businesses: who built the model, what does it not see and how much of our budget is being allocated based on its output?