If you’ve read our previous articles on the pitfalls of Consent Mode (where cookie blocking blinds your analytics) and on the dark funnel (where the most important buying decisions happen without you knowing), you’ve probably asked yourself a fundamental question: How are we supposed to measure what actually works in marketing today, and what doesn’t?
Picture this typical customer journey. A potential client listens to your CEO’s interview on an industry podcast. A week later, they read your LinkedIn post. Then they meet you over coffee at an offline conference. A month later, once their budget is approved, they open Google, type in your company name, click a sponsored link and sign a €20,000 contract.
So which channel actually won the deal? If you look at standard analytics tools, all the glory and 100% of the credit goes to that last paid search ad. And that’s exactly the moment your company starts losing money.
The Google Analytics Trap: Why “Last Click” Is a Road to Nowhere
Traditional digital analytics (led by Google Analytics) has long been built on so-called deterministic models. Put simply, it tries to track a specific user with cookies from the first ad impression all the way to the purchase. Many companies still rely on the last-click model (the last click wins) and never even try other attribution models.
Today, that approach falls hopelessly short. Why?
- Technological blindness: People use ad blockers, reject cookie banners and switch from phone to laptop. Your analytics only sees fragments of the journey.
- Ignoring the brand: If the last click gets all the glory, the algorithm (and often your CFO too) gets the impression that you only need to invest in performance Google Ads campaigns, and everything else – podcasts, PR, brand building – is just money down the drain.
- Strangled growth: If you cut investment in brand and educational channels (the ones that generate real demand), your “cheap” performance marketing at the bottom of the funnel dries up completely within a few months.
What Is Marketing Mix Modeling (MMM)? Back to the Roots, Thanks to AI
Marketing Mix Modeling (MMM) isn’t exactly new. Big FMCG brands (makers of fast-moving consumer goods like Coca-Cola or Procter & Gamble) were using it decades ago to measure whether their TV ads really lifted supermarket sales. They had no cookies or clicks, so they had to rely on pure math and statistics.
Before AI came along, though, the model was too expensive and slow for ordinary companies. It took data analysts months to crunch the numbers.
Today, thanks to AI, MMM is having a renaissance and becoming accessible to mid-sized B2B companies and online stores too. MMM doesn’t track a specific John Smith. It takes aggregated (summary) data on your spend across all channels (X euros on Facebook, Y euros on a podcast, Z euros on billboards) and compares it with your historical revenue. Using machine learning and statistical models, it then calculates how much each individual investment really contributed to your bottom line.
Why Management Loves It: The End of Budget Fights for Good
For CEOs and CFOs, MMM is a genuine relief. They finally get a tool that speaks the language of business, not the language of likes and impressions.
- The truth about your brand: Management finally sees in black and white that even though the podcast didn’t generate a single direct click to the cart, its presence in the marketing mix noticeably lifted overall sales.
- A holistic view: MMM can factor in external variables that Google Analytics completely ignores. The software will account for the fact that your sales didn’t jump because of a brilliant Facebook ad, but simply because your competitors raised prices across the board or the weather changed.
- Predictive modeling: MMM doesn’t just look at the past. It lets you play with scenarios. You can ask the system: “What happens to our revenue if we move €5,000 from Meta Ads to running offline events next month?” The model calculates a forecast for you.
In Practice: How to Launch MMM Without a Data Scientist
You used to need a whole department of data scientists for this. Today, thanks to open-source tools and AI assistants, it’s much simpler.
Tech giants have released their own models to the public for free. The best known are Robyn (from Meta) and Meridian (from Google, the successor to the older Lightweight MMM). You don’t have to code them from scratch. Marketing agencies or your in-house analysts load your data from spreadsheets (monthly ad spend, offline activities and total revenue) into these free libraries, and the software calculates charts showing the incremental return (incremental ROI) of each channel. You don’t need to track user privacy, just solid internal accounting data.
Stop Measuring Clicks, Start Measuring Impact
Chasing the perfect “last conversion” in an increasingly restrictive digital environment is a losing battle from the start. Privacy protection and dark social have changed the rules for good. Marketing Mix Modeling (MMM) brings common sense back into companies. It shows you the forest, while you’ve been desperately trying to count every single tree. If you want to allocate your marketing budget with surgical precision and real profit, you have to move from tracking clicks to deep statistical modeling.
Frequently Asked Questions (FAQ)
Does this mean we can delete Google Analytics 4 (GA4) from our website completely?
Definitely not. MMM and GA4 complement each other, they don’t replace each other. GA4 is still an absolutely critical tool for measuring micro-metrics: you need it to see whether your website loads fast, which page people leave from, or whether your shopping cart works properly. GA4 is for optimizing your digital product. MMM is for optimizing your marketing budget.
How much data do we need for MMM to be accurate?
Since MMM works by finding correlations and patterns over time, the rule is simple: the more historical data you have, the more accurate the model. For truly relevant results, it’s recommended to load detailed data (weekly spend and revenue) into the model, ideally for the last two to three years. If a company has only existed for three months, MMM isn’t a good fit yet, and it will have to rely on other models for now (e.g. self-reported attribution).
How often should we update and evaluate the model?
Historically, MMM reports were done twice a year because they were extremely time-consuming. Thanks to AI and automation, leading companies now update their models monthly, and in some highly dynamic industries (like fast-fashion e-commerce) even weekly. The model keeps learning from new data and becomes your ongoing strategic advisor.
Want to know which campaigns really bring you customers and where to move your budget? See how we work with Google Ads and measurement – so every euro counts.
