If you are spending $50 million across paid search, social, TV, email, and retail media and your CFO asks you which channels are actually driving revenue, the wrong answer is "last click said Google." AttributionAttributionA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.Voir la définition complète → is not a measurement problem. It is a resource allocation problem. Get it wrong and you are systematically defunding your best channels while overpaying for channels that just happen to be nearby when someone converts. This lesson gives you the frameworks to stop guessing and start making defensible, data-backed budget decisions.
AttributionAttributionA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.Voir la définition complète → is the process of assigning credit to the marketing touchpoints that influenced a conversion. A touchpoint is any interaction a customer has with your brand before buying: a YouTube pre-roll, a branded search ad, an email, a podcast mention, an influencer post, a store visit. The question attributionattributionA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.Voir la définition complète → tries to answer is: which of these touchpoints caused the purchase, and how much credit should each one receive?
The reason this is hard is that customers do not convert in a straight line. A customer might see a Meta video ad on Tuesday, Google your brand on Thursday, open a promotional email on Friday, and then buy in-store on Saturday. Four touchpoints, one conversion, and your current system probably gave 100% of the credit to the email or to Google, whichever was last before the online portion of the journey.
Rule-Based Attribution Models
These are statistical shortcuts. Last-touch gives 100% credit to the final touchpoint before conversion. First-touch gives 100% to the first interaction. Linear splits credit equally across all touchpoints. Time-decay gives more credit to touchpoints closer to the conversion. Position-based (U-shaped) gives 40% to first touch, 40% to last touch, and splits the remaining 20% across middle touchpoints. These models are fast and cheap to run, but they are built on arbitrary assumptions, not actual data about how your customers behave. Use them only for directional gut checks, not for budget decisions above $5 million.
Data-Driven Attribution (DDA)
DDA uses machine learning to analyze your actual conversion paths and calculate the incremental contribution of each touchpoint. Google's DDA model, available inside Google Ads and GA4, uses a counterfactual approach: it asks "what was the probability of conversion with this touchpoint present versus absent?" This is meaningfully better than rule-based models for channels where you have high data volume. The catch is that DDA still only sees what it can track, which means it under-credits channels like TV, out-of-home, and in-store.
Media Mix Modeling (MMM)
MMM is a statistical technique (usually regression-based) that correlates aggregate marketing spend data with aggregate sales outcomes over time. It does not require cookies or individual-level tracking. It can incorporate TV, radio, weather, pricing, and macroeconomic variables. Procter and Gamble has used MMM as a primary measurement tool for decades because their media mix is complex and their customer base is enormous. The downside is that MMM requires 2 to 3 years of clean historical data to produce reliable outputs, and the models need to be rebuilt every 6 to 12 months as the media environment shifts.
Incrementality Testing (Lift Testing)
This is the gold standard. You run a controlled experiment: one group of customers sees your ad, one group does not (the holdout group), and you measure the difference in conversion rateconversion rateThe percentage of visitors or prospects who complete a desired action (purchase, sign-up, contact form), calculated as conversions divided by total opportunities.Voir la définition complète → between them. That difference is the true incremental lift your ad generated. Meta calls its version Conversion Lift. Google calls its version Geo Lift. Airbnb ran geo-based incrementality tests in 2019 and discovered that a significant portion of their paid search spend on branded terms was capturing customers who would have converted organically anyway. They cut that spend and reinvested it into brand-building channels.
Airbnb and Incrementality (2019 to 2022)
Brian Chesky's team ran systematic geo holdout tests across paid search. They found that branded keyword spend was largely non-incremental: customers who saw those ads were going to book anyway. Airbnb cut roughly $800 million in performance marketing in 2020 (partly forced by COVID, but the incrementality data informed the decision to not restore it fully). By 2022, they reported that 90% of traffic was direct or organic. Revenue per dollar of marketing spend improved significantly, and they communicated this explicitly in their Q4 2021 shareholder letter.
Procter and Gamble and MMM-Driven Reallocation
CMO Marc Pritchard has been public about P&G's use of MMM to audit their digital media supply chain. After running MMM analysis in 2017 and 2018, P&G cut $200 million in digital advertising that they determined was either fraudulent, reaching bots, or generating zero measurable lift. They shifted budget to TV, premium digital, and agency consolidation. Sales did not drop. Their MMM told them what was actually working, and they acted on it.
Redbull and Multi-Touch Attribution Stack
Red Bull operates in over 170 countries with a media mix that includes owned mediaowned mediaMedia channels a company owns and controls directly, such as its website, blog, newsletter, social accounts and mobile app. No per-use payment to a publisher is required.Voir la définition complète → (Red Bull TV, Red Bull Media House), experiential, social, and performance channels. Their analytics team uses a layered approach: DDA for owned digital channels, MMM for cross-channel budget planning, and brand lift studies for awareness media. They do not rely on any single model. The layered approach allows them to triangulate: if DDA, MMM, and lift studies all point in the same direction, they move budget with confidence.
Using last-touch attribution for budget planning above the campaign level. Last-touch systematically over-credits bottom-funnelfunnelThe customer journey from awareness to purchase, typically Awareness, Interest, Consideration, Decision, Action, with prospects narrowing at each stage.Voir la définition complète → search and email while making brand and awareness spend look worthless. Companies that run their entire budget off last-touch end up cutting upper-funnelfunnelThe customer journey from awareness to purchase, typically Awareness, Interest, Consideration, Decision, Action, with prospects narrowing at each stage.Voir la définition complète → spend, their pipelinepipelineAll active sales opportunities across the stages of the sales process, together with their combined potential value and probability of closing.Voir la définition complète → dries up 6 to 12 months later, and they cannot explain why.
Treating MMM output as precise rather than directional. MMM models carry wide confidence intervals. A model might tell you that TV drives 23% of revenue, but the real number could be anywhere from 15% to 35%. CMOs who act on MMM outputs as if they are exact numbers make over-confident budget swings. Use MMM to identify the direction and magnitude of change, not the precise percentage.
Running incrementality tests during abnormal periods. Launching a geo holdout test during a major promotion, a holiday, or a supply chain disruption contaminates your results. The control and treatment groups experience different external conditions, and your lift measurement becomes unreliable. Always test during your baseline, steady-state periods.
Google's official documentation on how data-driven attribution works inside Google Ads, including the counterfactual methodology and minimum data requirements.
Meta's step-by-step guide to setting up conversion lift studies, including holdout group sizing and how to interpret lift results for budget decisions.