25 June 2025

Media mix modelling vs observable attribution

In summary

• Media mix modelling (MMM) is gaining relevance due to privacy regulations and browser restrictions, offering a privacy-safe way to measure marketing impact using aggregate data instead of individual tracking.
• MMM and traditional analytics tools are complementary. MMM helps understand high-level effectiveness and external factors, while analytics tools provide granular, real-time insights into user behaviour and conversion paths.
• AI is accelerating MMM adoption, making it easier to clean data, run frequent models, and uncover early trends—bringing advanced measurement capabilities to more businesses.

What is media mix modelling (MMM)?

Media mix modelling is a statistical methodology that uses aggregated data to measure the impact of various marketing activities such as advertising campaigns.

The interest around media mix modelling (MMM) has been increasing over the last few months, so it’s the ideal time to explain why it is being discussed, how it fits into existing analytics toolsets and to answer some of the most commonly asked questions.

How does MMM work?

MMM uses statistical techniques to estimate how different marketing activities contribute to a desired outcome, such as increased sales. It works by analysing historical campaign data and simulating incremental changes, like adding a new channel or adjusting spend levels, to observe the resulting impact. Any uplift or shift in outcomes helps indicate the effectiveness of each campaign element.

Increased interest led by regulation and privacy changes

In recent years, a perfect storm of changes has reshaped the marketing and measurement landscape. Increased regulation, such as GDPR, CCPA, and Australia’s Privacy Act review, combined with browser-level changes like Apple’s Intelligent Tracking Prevention (ITP) and growing consumer demand for privacy, has challenged traditional digital measurement methods. Historically, web analytics tools have relied on tracking individual user behaviour through cookies, device IDs, or even fingerprinting.

These methods now face significant legal, technical, and ethical hurdles. MMM offers a privacy-resilient alternative by analysing campaign effectiveness at an aggregate level. Rather than tracking an individual, it evaluates overall trends and performance across media channels. This ensures it is well suited to a world with limited user level observability.

Automation has made MMMs more relevant

While MMMs have been used in marketing for decades, their adoption by most businesses has historically been limited. This is largely due to the need for specialist in-house expertise to collect, process, and analyse large sets of data, as well as to run complex statistical models.

Traditionally, MMMs were primarily designed for offline channels like TV, radio, billboards, and print, making them less suitable for digital campaigns involving search engines or social media.

However, newer open-source solutions like Google’s Meridian and Meta’s Robyn are addressing these limitations. These models incorporate digital channel data and, when combined with the scalability of cloud computing, can be run much more frequently. What once may have happened perhaps once or twice a year using traditional MMMs is now able to be executed monthly, weekly, or even daily through automation, bringing MMMs into the modern data driven marketing toolkit.

Can MMMs replace attribution tools in analytics?

Whilst MMMs sound like a great solution in a privacy centric world, there are a few reasons why they should not yet replace existing analytics tools.

1. Aggregate level insights: MMMs work at the aggregate level, which also limits their ability to provide granular level breakdowns and further segmentation. This means you can answer questions such as what is the impact on adjusting spend on total conversions, but you cannot ask which path a user took to complete a conversion.
2. Feedback loops are slower: MMMs are great at reviewing historical performance based on larger datasets, but are inferior at real time insights and decision making. It also takes time to build up the necessary data, which may not exist in a new business such as a start up environment.
3. Minor misconfigurations can lead to wild inaccuracies: Minor misconfigurations in MMMs can lead to some relatively wild inaccuracies. They require careful configuration and expertise to ensure accurate data is returned out of the models. Analytics tools on the other hand, are potentially less sensitive to small misconfiguration issues making them a little more fool proof to work with. Also, minor configuration issues can be addressed relatively quickly and the data can be fixed in a shorter space of time.
4. Funnel Abandonment: Since MMMs work at the aggregate level, they are limited when it comes to observing at which point in a conversion journey users are likely to abandon the conversion process. This is where traditional analytics tools come into play. They are very good at measuring friction points in a conversion journey and showing where users are going instead.

Complement rather than compete

For the reasons listed above, MMMs and analytics attribution tools should be considered as complementary, rather than directly competing with one another as a source of truth. Each relies on a slightly different data input method:

• MMMs: Aggregate level data such as spend, and total conversions.
• Analytics tools: Event or user level information like clicks, events and purchases.

In other words, analytics tools are very good at explaining what happened, but not necessarily why, especially if outside factors (off-site) are involved. MMMs (if the data is available) can factor in external factors, such as competitor pricing changes, and help articulate what impact these industry level changes may have had on your companies bottom line.

On the other hand, traditional analytics tools can help build audiences and target them with specific messaging to see which message resonates and performs best for that cohort.

It is clear that each attribution measurement method has its own strengths and weaknesses.

The future of MMMs, Analytics and AI

So, how will AI impact the use of MMMs and traditional attribution tools in the future?

• Tools such as Robyn and Meredian are early examples of automated MMMs. AI is likely to make it easier to push potentially sub-optimal / messy datasets into these models and run multiple models and iterations for comparison and analysis without the requirement of a team of expert statisticians.
• AI tools will be better at spotting early patterns as they emerge, rather than needing to wait weeks, months or years to spot opportunities and emerging insights.
• Older MMMs struggled with digital channels like paid search and social, AI will help facilitate understanding the casual impact of auction based and user level media platforms (e.g. impression caps).
• As observability erodes, AI will step in to fill gaps in datasets using look-alike type analysis to predict specific outcomes.

Over time it’s likely that AI will lower the barrier to entry for many businesses. It will be more feasible to run sophisticated models on a frequent basis and help inform and predict where marketing dollars should be spent for a specific outcome.

Final thoughts

MMM is regaining popularity not because it’s new — but because it’s newly viable in a world that demands privacy-conscious, high-level measurement. Combined with traditional analytics and powered by AI, MMM is becoming a critical part of the modern marketing stack — not a replacement, but a strategic partner.