Part 1: What Your Attribution Model Isn’t Telling You

Attribution is a touchy subject for a lot of people - agencies and clients alike. Client teams under pressure from key stakeholders to prove ROI ask their marketing managers and agencies how to demonstrate the effectiveness of their spend and are fairly often met with a concerned look and an "...it depends." 

The hard truth is that it is impossible to prove the efficacy of advertising in a black and white way. The closest thing to it is setting measurable goals, milestones, and where possible, revenue values at each stage of the user journey - then combining this with the most accurate tracking possible. Even then, you're still working with the best possible estimation of return rather than a definitive answer. 

Arriving at that estimation has gotten considerably harder over the last five years, for a few reasons. Data privacy regulation has tightened - GDPR reshaped how consent is collected, and iOS 14's App Tracking Transparency update in 2021 meaningfully reduced the signal available to platforms like Meta. Third-party cookies, the long-standing solution to cross-site tracking, have been on a slow and unglamorous farewell tour. Firefox and Safari blocked them years ago, Brave blocks them by default, and most other browsers have followed suit or have no plans to reintroduce them. Chrome remains one of the few holdouts still supporting them - Google reversed its deprecation plans in 2024 - which means third-party cookies are still technically available, but only to a shrinking share of the browsing population. Depending on how that changes, the actual measurement value of cookie-based tracking is continuing to erode regardless of what Google decides to do with them. 

Add to this the issue of platforms marking their own homework and a steady rise in black-box AI campaign types, and the picture of what's actually driving results gets muddier day-by-day. The problem is further complicated by the number of different tracking implementations available and a fairly widespread lack of understanding of how each of them actually works. 

The goal of this series isn't to suggest that attribution is broken beyond use - it's to explain what the different tools actually measure, and what they don't. 

Part 1 – Tracking Implementation 

Pixel-based tracking 

Event snippets placed on a website that track user interactions - page views, add-to-basket events, purchases, and anything in between. Data is passed back to the advertising platform for attribution and reporting. The catch is that it's directly affected by the cookie consent status a user provides on entering the site. If they decline advertising cookies - which is increasingly common - or if they're running an ad blocker, that data is lost. 

Server-to-server tracking 

Transmits data directly from a website's backend server to third-party ad platforms - Meta's Conversions API (CAPI) and Google Enhanced Conversions being the most commonly used implementations. Because nothing needs to load on the user's device, it isn't affected by cookie consent status or ad blockers, making it a more complete and privacy-compliant method. Implementation is more resource-intensive and requires technical knowledge to get right, but most platforms provide a walkthrough and the resulting improvement in data quality is worth the effort. 

Statistical modelling and probability attribution 

This method works best with large datasets. Analyses patterns in marketing data to estimate the impact each channel had on a given outcome. For example, if a campaign launches on Instagram and sales increase 25% over the following week, statistical modelling can attribute a proportion of those sales to that activity - using comparable periods and historical performance data to isolate the difference. Particularly useful for upper-funnel or brand campaigns, where measurable direct outcomes are harder to come by. 

First-party data with consent management 

Approaches things from the other direction entirely. Rather than tracking anonymous users through campaigns, it captures data directly from users who have actively provided it - form fills, account registrations, purchases - combined with informed consent on how that data is used. The downside here is that it doesn't give a picture of all campaign-driven traffic, but it does build a reliable list of converted users that can be fed back into targeting and audience strategy to maintain relevance over time.  

Part 2 of this series will cover the different attribution models available within advertising platforms - what they measure, how they differ, and where each of them tends to mislead.