Why you under-spend TOFU channels and how to fix attribution stack in 2026

5 min.
Gabriel Berger Gabriel Berger

In 2026 we can no longer pretend that simple click based attribution is enough to evaluate real ROI / ROAS and correct budget allocation. Let us compare what possibilities and tools exist and how to use or not use them.

Option 1: click based multi touch attribution (MTA)

Tools like Google Analytics (in 99 % of cases) track only web data (if you are lucky you import cost data) chronologically from session_start with UTM parameters to (eventually) purchase / conversion events.

UTMs in URL are not guaranteed. Browsers strip them, ad platforms replace them with their own 1st party parameters (e.g. gclid for Google Ads, fbclid for Meta). In app browsers skew data: you count as 2 users if you open a link in Instagram / other in app browsers and then open the link in Safari (or you go through direct and the session is lost completely) and your cross device behavior is untrackable. In a better case you integrate 1st party data collection for GA4, but honestly, how many of your users are logged in?

And finally: they do not measure incrementality. How many of your remarketing buyers would convert even without the campaign? (I am not saying most, but you should ask this question and find out the answer yourself). Evaluating marketing performance based on GA4 data is drawing conclusions from a fraction of (skewed!) data.

Option 2: post view / impression based MTA

Tools like Roivenue ($$$) / data warehouse with a custom attribution stack (even more costly :) ) allow you to expand the customer journey even outside your website.

Post view MTA allows you integration with digital (only) ad platforms / channels (Meta, Google, Sklik, Bing, Criteo, …) which is smart because it captures a wider part of the user journey. It is also fairer for „TOFU“ channels (like Meta) because user touchpoints for these channels are more on platform than on web = it means that Meta users are exposed to your content more on Meta, compared to more „BOFU“ users from Search campaigns: they directly click to your website and spend time there. So if you take into account only website touchpoints (option 1), you are missing out and massively under spending on TOFU channels! For this reason do not do budget allocation based on GA4 ROAS / PNO.

Marketing Mix Modeling (MMM)

Platforms (they are not plug and play so I do not call them tools) like Meridian from Google or Robyn (Meta) work on a fundamentally different principle. They do not look at individual customer journeys but build regression models based on aggregated data. The simplest way to explain regression is the „ice cream shop example“. Let us say you sell ice cream and you are affected by temperature. You draw a simple graph: daily temperatures on the X axis and daily sales on the Y axis. You see some relationship and can build a simple „mental model“ to predict and close your shop when there is 10 C the next day.

In marketing mix modeling it is similar but with:

  • much more input variables (marketing spends, impressions, emails, footfall in your offline stores, pricing / promotions, … instead of temperature in the ice cream example) and the same 1 output variable: revenue or other KPI (app installs, sign ups): what you are „modeling“
  • channel transformations: you transform your spends / impressions using adstock function (how the impact of your channel decays across time: TV probably has a more long term effect than Search) and with saturation function (your incremental spend from 1M to 2M probably has more impact than incremental spend from 10M to 11M)

Most MMM platforms also give you a recommended long term budget allocation based on the model and your spend preferences: same / smaller / higher budget.