The attribution illusion

Online advertising faces a critical measurement challenge: when customers see ads across multiple websites before converting, which ads deserve credit? A groundbreaking study published in Marketing Science offers surprising insights.

Seven years ago, in July 2017, Procter & Gamble made headlines when they slashed over $100 million in digital advertising spending, citing concerns over effectiveness. According to Ron Berman, Associate Professor of Marketing at the Wharton School of the University of Pennsylvania, this spotlights a persistent industry problem: when advertisers can't accurately measure which ads contribute to conversions, campaign effectiveness suffers.

Berman's study, "Beyond the Last Touch: Attribution in Online Advertising," published in Marketing Science, explores how attribution methods can help advertisers optimize multi-publisher campaigns. His research reveals that common attribution practices like last-touch attribution may significantly underperform compared to alternative methods.

The multi-publisher attribution challenge

The study addresses a fundamental challenge in online advertising: when consumers visit multiple websites during an advertising campaign, the ads they see create cumulative effects that are difficult to measure individually.

"When consumers multi-home and visit multiple publishers during these campaigns, the ads they see often have a cumulative effect that creates an externality between the effectiveness of ads across publishers," explains Berman.

This uncertainty about which ad exposures drive conversions leads advertisers to make suboptimal bidding decisions, resulting in inefficient ad allocations and diminished campaign effectiveness.

"These inefficiencies often result in severe degradation of effectiveness of advertising campaigns, leading to frustration by advertisers and questions about the effectiveness of online advertising," Berman notes.

In today's digital landscape, this problem is particularly acute. According to data cited in the study, Google and Facebook alone account for more than 50% of the U.S. online advertising market as of 2015. When consumers visit both platforms, attributing conversions becomes complex.

Why attribution matters

The study analyzes what happens when advertisers employ different attribution models to measure the contribution of each ad exposure or "touchpoint" to consumer conversions.

Multi-touch attribution uses campaign data to estimate each publisher's effectiveness and contribution to conversions. These estimates help advertisers adjust bids on different publishers to optimize campaigns.

According to a survey cited in the study, 54% of advertisers use last-touch attribution, which assigns full credit to the final ad impression before conversion. However, 42% of advertisers reported being "unsure of how to choose the appropriate method/model of attribution."

Berman's research reveals several surprising findings about attribution methods:

1. Last-touch attribution often hurts profitability: The popular last-touch method can reduce advertiser profits compared to not using attribution at all, unless there is high asymmetry between publishers.
2. The Shapley value outperforms: An alternative method based on the Shapley value (which assigns the average marginal contribution to each touchpoint) achieves higher profits compared to last-touch attribution when there's sufficient asymmetry between publishers.
3. Some advertisers should skip attribution entirely: When advertisers value conversions highly, implementing attribution may provide negligible benefits compared to no attribution.
4. Last-touch benefits early funnel publishers: Contrary to industry assumptions, publishers that appear earlier in the conversion funnel benefit more from last-touch attribution than those appearing later.
5. Attribution affects market pricing: The study found that advertisers will pay higher ad prices when using last-touch attribution and lower prices when using the Shapley value method.

When should advertisers invest in attribution technology? The study suggests several practical considerations:

For advertisers working with highly asymmetric publishers (where one publisher generates significantly more traffic than others), the Shapley value attribution method may substantially improve profits compared to last-touch or no attribution.

"When there is enough asymmetry between publishers, or when the externality is strong, using attribution, and particularly the Shapley value, may yield advertiser profits which are close to the optimal possible," the paper states.

However, the research also indicates that when publishers are relatively similar in their contribution to conversions, or when conversions are highly valued, attribution may offer minimal benefits.

In terms of ad pricing, the research predicts: "When firms moved from not using attribution at all to using last-touch attribution, the price of advertising was expected to increase for the large and global advertisers because of overbidding and increased competition."

The research comes at a critical time when major advertisers are increasingly questioning digital advertising effectiveness. As Berman observes, "Attribution creates a virtual competition between publishers, resulting in a team compensation problem."

This insight explains why different attribution methods may produce dramatically different outcomes. The study shows that the Shapley value method, which splits externality effects equally between publishers, generally outperforms last-touch attribution in increasing advertiser profits.

The research also suggests that as attribution technology evolves, advertisers face a trade-off between market efficiency and individual advertiser profit. While some attribution methods may improve overall market efficiency, they don't necessarily maximize individual advertiser profits.

In an interesting parallel to game theory, the study indicates that in markets with multiple global advertisers, having less sophisticated attribution methods may sometimes benefit advertisers by softening competition.

As measurement technology improves, the study predicts that "as more advertisers become more efficient, the competition becomes stronger which results in lowered profits (and higher revenue going to the publishers)."

The Evolution of Online Attribution

• 2011-2015: Research by Lambrecht and Tucker, Blake et al., and Lewis and Rao demonstrates that online advertising effects are moderate at best and require large sample sizes to identify
• 2016: IAB dubbed this "the year of attribution" in annual predictions
• 2017 (July): P&G cuts over $100 million in "largely ineffective" digital advertising
• 2017: Berman's research reveals fundamental challenges in attribution methodologies
• 2018: Berman publishes "Beyond the Last Touch: Attribution in Online Advertising" in Marketing Science
• 2024: Attribution continues to evolve as advertisers seek more accurate measurement methods

For advertisers navigating the complex world of multi-touch attribution, Berman's research offers clear guidance: evaluate the asymmetry between publishers, consider the value of conversions, and carefully select attribution methods based on these factors rather than defaulting to industry standards like last-touch attribution.

The study concludes with a sobering observation that may explain why attribution remains challenging despite technological advances: "Attribution will never achieve the optimal allocation an advertiser could achieve with complete information."