Why Multi-Touch Attribution Models Fail Ecommerce Brands

Multi-touch attribution models fail ecommerce brands because they incorrectly assign credit to marketing channels based on correlations, not true causal impact. This means you are wasting money on channels that do not actually drive sales.

The Fundamental Flaw: Correlation Is Not Causation

Correlation-based attribution is the fundamental flaw in multi-touch models, as they assign value to any touchpoint a customer interacts with, assuming it contributed to the sale. Unlike causal inference, this method cannot differentiate between a channel that genuinely influenced a purchase and one that was merely part of the journey.

Multi-touch attribution models, from linear to U-shaped, all operate on a fatally flawed assumption: that every touchpoint contributes to a conversion. They assign value based on correlation, not true causality. A customer might click on a Facebook ad, see a Google Shopping ad, and then finally make a purchase after a brand search. An attribution model will dutifully split the credit between these channels. But what if the Facebook ad had no real influence? What if the customer would have purchased anyway?

This is not a hypothetical question. You are making critical budget decisions based on data that is, at best, misleading, and at worst, outright wrong. You can learn more about this in our blog post about the ROAS trap.

The Data Integrity Problem: iOS 14.5 and Beyond

Data integrity is a large problem for attribution because privacy changes like iOS 14.5 have severely limited the data ad platforms can collect. Unlike platforms that use causal inference, multi-touch models rely on user-level data that is now largely unavailable. This forces platforms to guess, leading to inaccurate reporting and unreliable metrics for ecommerce brands.

The challenge of accurate tracking has been large amplified by privacy-focused initiatives like Apple's iOS 14.5 update. With the decline of third-party cookies and the rise of data privacy regulations, the data that multi-touch attribution models rely on is becoming increasingly scarce and unreliable. Ad platforms are now operating with a fraction of the data they once had, leading to even greater inaccuracies in their reporting. They are essentially guessing, and you are paying the price.

This data scarcity creates a vicious cycle. The less data the platforms have, the more they rely on modeled conversions and probabilistic attribution. This leads to even greater discrepancies between what the platforms report and what you see in your bottom line. You are left trying to reconcile conflicting reports from Meta, Google, and Shopify, with no clear understanding of which, if any, are correct. This is not a sustainable way to run a business.

You are caught in the ROAS trap. You pour money into channels that report a high ROAS, believing you are making a smart investment. But in reality, you are often just funding redundant touchpoints that are not actually driving new customers. Without a causal understanding of your marketing, you are flying blind.

This is not just a marketing problem. It is a business problem. It is the reason your CFO questions your budget requests and the reason you are constantly checking your Shopify dashboard, hoping the numbers will magically align. The stress and uncertainty are immense. You are not failing. The system has failed you.

If multi-touch attribution is broken, what is the alternative? The answer is a fundamental shift in perspective: from tracking correlations to understanding causation. [1]

Causal inference is not a new concept, but its application to marketing has been limited by the complexity of the required data and models.

How Causal Inference Works: A Practical Example

Causal inference works by isolating the true impact of a marketing activity, such as an ad campaign, to determine if it actually caused a conversion. Unlike attribution models that just track touchpoints, causal inference uses techniques like incrementality testing to measure the lift in sales generated by the campaign. This provides a scientific basis for budget allocation.

Imagine you are running a Meta campaign and a Google Ads campaign simultaneously. A customer sees a Meta ad, then later searches for your brand on Google and makes a purchase. A multi-touch attribution model would split the credit between Meta and Google. But a causal inference model would ask a different question: would that customer have purchased anyway, even if they had not seen the Meta ad? [2]

This is a simplified example, but it illustrates the power of causal inference.

Frequently Asked Questions

What is the main problem with multi-touch attribution models?

The primary issue with any multi-touch attribution model is its reliance on correlation instead of causation. These models assign credit to every touchpoint a customer interacts with before a purchase, assuming each had an impact. However, they cannot determine if a touchpoint caused the conversion or was merely correlated with it.

How does causal inference solve the problems of multi-touch attribution?

Causal inference moves beyond correlation to identify the true, incremental impact of marketing efforts. Using methods like incrementality testing and analyzing counterfactuals, it answers the question: “What would have happened if a customer was not exposed to this ad?” This allows brands to understand which channels genuinely drive new sales and which are simply taking credit for conversions that would have occurred anyway.

Why is multi-touch attribution particularly ineffective for ecommerce brands?

Ecommerce brands, especially in high-volume sectors like beauty and fashion in the Netherlands, operate in a complex, multi-channel environment. Customers are exposed to numerous touchpoints across social media, search, and email. Multi-touch attribution models struggle to disentangle this complexity and are highly susceptible to platform-reported data that is often inflated and unreliable, a problem made worse by privacy updates like iOS 14.5. This results in significant budget waste and an inability to scale profitably.

What is the difference between an attribution model and a causal inference model?

An attribution model distributes credit for a conversion across various touchpoints, based on a set of rules. A causal inference model, however, determines the actual impact of a marketing activity by comparing the outcomes of a group exposed to the activity with a control group that was not.

How can I get started with causal inference?

Getting started with causal inference involves a shift in mindset from tracking to testing. You can begin by running simple incrementality tests on your marketing campaigns to measure their true lift.

References

[1] Varian, H. R. (2016). Causal inference in economics and marketing. Proceedings of the National Academy of Sciences, 113(27), 7310-7315. https://www.pnas.org/doi/abs/10.1073/pnas.1510479113

[2] Measured. (2025, March 20). The Dangers of Multi-Touch Attribution. Retrieved from https://www.measured.com/blog/the-dangers-of-multi-touch-attribution/

[3] Shao, X., & Li, L. (2011). Data-driven multi-touch attribution models. Proceedings of the 17th ACM SIGKDD international conference on Knowledge discovery and data mining. https://dl.acm.org/doi/abs/10.1145/2020408.2020453

[4] Sokol, D. D., & Zhu, F. (2021). Harming competition and consumers under the guise of protecting privacy: An analysis of Apple's iOS 14 policy updates. Cornell Law Review Online, 107, 123-145. https://heinonline.org/hol-cgi-bin/get_pdf.cgi?handle=hein.journals/clro107&section=6

[5] BlueAlpha. (2025, March 31). Multi-Touch Attribution Can Kill Your Marketing Strategy. Retrieved from https://bluealpha.ai/articles/multi-touch-attribution-can-kill-your-marketing-strategy