Moving beyond multi-touch attribution - DigiMarCon CanWest 2024
0 likes19 views
The document discusses the limitations of traditional attribution models in marketing, highlighting issues such as privacy, cookie tracking, and session challenges. It advocates for an incrementality-first approach, which focuses on the actual impact of marketing activities on key performance indicators rather than individual consumer interactions. Key takeaways emphasize the distinction between attribution and measurement, asserting that incrementality is crucial for effective marketing strategy and decision-making.
![…and DDA sounds great, right?
[Data-driven] Attribution uses machine learning
algorithms to evaluate both converting and
non-converting paths. The resulting Data-driven model
learns how different touchpoints impact key event
outcomes. The model incorporates factors such as time
from key event, device type, number of ad interactions,
the order of ad exposure, and the type of creative assets.
Using a counterfactual approach, the model contrasts
what happened with what could have occurred to
determine which touchpoints are most likely to drive key
events. The model attributes credit to these touchpoints
based on this likelihood.”
“
r”](https://image.slidesharecdn.com/digimarconcanwest-movingbeyondmulti-touch-240428183232-60a7f887/85/Moving-beyond-multi-touch-attribution-DigiMarCon-CanWest-2024-6-320.jpg)
Moving beyond multi-touch attribution - DigiMarCon CanWest 2024
- 1. April 25, 2024 Headlight x DigiMarCon Moving Beyond Multi-Touch With Rich Ingilby
- 2. Let’s start with some definitions
- 3. Attribution Directly connecting an individual or cohort to the marketing activities they’ve engaged with. Incrementality The actual impact that a marketing activity has on a KPI, and which would not have happened without it Measurement Modeling Using statistics and machine learning to find correlations and signals that can describe the incremental impact of marketing activities.
- 4. The first responder: The nervous system of our marketing campaigns Attribution:
- 5. GA4 is a good example of an attribution channel* *and there are lots of others
- 6. …and DDA sounds great, right? [Data-driven] Attribution uses machine learning algorithms to evaluate both converting and non-converting paths. The resulting Data-driven model learns how different touchpoints impact key event outcomes. The model incorporates factors such as time from key event, device type, number of ad interactions, the order of ad exposure, and the type of creative assets. Using a counterfactual approach, the model contrasts what happened with what could have occurred to determine which touchpoints are most likely to drive key events. The model attributes credit to these touchpoints based on this likelihood.” “ r”
- 7. But, there’s a few pretty fundamental issues…
- 8. Where attribution models fall down They only know when people have clicked an ad 1
- 9. Where attribution models fall down Privacy (e.g. SKAN) 2
- 10. Where attribution models fall down Cookie banners 3
- 11. Where attribution models fall down 4 Tracking across sessions
- 12. But wait… THERE’S MORE Google going cookieless sometime1 Apple blocking all UTMs on Safari 2 And shared links3 1. https://developers.google.com/privacy-sandbox/blog/cookie-countdown-2023oct 2. Apple WWDC 2023 — June 5 2023 3. https://www.pcmag.com/news/to-improve-privacy-apple-to-strip-tracking-parameters-from-shared-urls 4. https://backlinko.com/duckduckgo-stats 5. https://headerbidding.co/walled-gardens-in-ad-tech/ 6. https://hbr.org/2023/08/how-the-pandemic-changed-marketing-channels#:~:text=Nearly%20two%2Dthirds%20(61%25), The rise of privacy-centric browsers4 The rise and rise of digital walled gardens 5 Rise in the number of marketing channels 6 MORE PRIVACY MORE SKAN
- 13. What impact does this have? Take a look at your first and last click attribution report And… how many people are coming directly to your site? Well…
- 14. Attribution is a first responder. It’s your campaign’s nervous system. It’s great for instrumentation Attribution is actually great for some stuff
- 15. Where attribution models fall down ∞ …But, would they have converted anyway? Reason #
- 16. The solution? Incrementality “The actual impact that a marketing activity has on a KPI”1 … 1. https://www.adjust.com/glossary/incrementality/ … and which wouldn’t just have happened anyway 2 2. Rich Ingilby (2024)
- 17. It doesn’t care about individuals It looks at cohorts and aggregates. No tracking? No problem. What does an incrementality-first approach do? It doesn’t care if someone clicked or viewed It looks at outcomes, not the journey It tracks actual impact It compares what actually happened against a counterfactual: What was your impact?
- 18. So how do you measure incrementality? Source: Towards Data Science - Power Analysis for Data Scientists https://towardsdatascience.com/power-analysis-17636d3f059b You run a test We need two things: 1) We need results with our marketing present, and results without our marketing present 2) We have to be confident of effect size
- 20. You could also look at alternative data sources For example: “How did you hear about us?” surveys
- 21. Or take this straight to MMM* (*Media Mix Modelling)
- 22. MMM aims to reattribute based on incrementality
- 23. Response Decomposition Given current spend levels, and current results, what proportion of the results is a product of the constituent channels? Budget Allocation Optimization Given a spend X, what is the optimal allocation of that spend? Response Optimization Given prior performance scaling curves, at what point will increased spend start resulting in diminishing returns? What do we use MMM for?
- 24. Day to Day performance reporting The model by definition is built on prior performance, it won’t tell you anything about current performance. You cannot optimize to maximize MMM outputs. Compensation for Lack of/Noisy Signal The models depend heavily on long term (2y+) trends, stable performance, strong data points and knowledge of external factors that impact performance. Discovering Unexpected External Performance Impacts Models by their nature are for projecting forward fixed assumptions. If those assumptions are wrong, so is the model. What don’t we use MMM for?
- 25. MMM requires a large number of inputs You need: Historic data + Building* + Know-how* *We can provide those parts dw
- 26. Each of these has a part to play Speed Comparability Accuracy Attribution Surveys Testing MMM High Med Low
- 27. Sweet, so how do I use Incrementality?
- 28. Thinking in terms of incrementality opens a lot of doors
- 29. Channel X says it’s doing a lot. Is it? Or is it just getting in front of people before they buy? It shows where you need to fix bad channels
- 30. You can’t measure the impact of brand awareness with attribution You can with incrementality Incrementality enables full funnel marketing https://www.thinkingunstuck.com/think-pieces/the-two-most-important-charts-in-marke ting-and-why-they-matter-now-more-than-ever
- 31. All of your spend is going to statics, you say? Those things that are great at driving clicks, you say? It lets you measure creative impact
- 33. Measuring the actual impact of your marketing… + Operating across the funnel + Focusing on incremental revenue Incrementality lets you GROW
- 34. Key takeaways 1. Attribution and measurement are different things 2. Attribution will only get worse 3. Good attribution enables good instrumentation, good measurement enables good strategy 4. Incrementality, is the most powerful method at our disposal to measure impact 5. When you know what is delivering impact, you know how to grow