In the AI World, everything is an Attribution problem

For over a decade, I've dedicated my expertise to assisting the largest Latin American companies in honing their marketing strategies. A persistent challenge these companies encounter is reconciling the data from their reporting systems with the actual value their marketing activities generate. This reconciliation is crucial not just for reporting but for truly enhancing their strategic approaches.

Consider the reality we inhabit: a marketplace saturated with brands clamoring for attention through every conceivable medium. From the endorsements of friends to the pervasive nature of mobile ads and digital billboards, each interaction subtly or overtly shapes our purchasing decisions. However, quantifying the influence of each interaction remains a formidable challenge. Reflect on your last significant purchase; can you pinpoint every factor that swayed your decision and its specific contribution? I bet you can't.

In my fieldwork, applying an econometric lens, I've observed that this attribution conundrum is not unique to marketing but parallels many challenges across various business functions. Whether it's evaluating employee performance or analyzing the elasticity of pricing and product offerings, the need for a robust understanding of causal relationships is ubiquitous. Excelling in forecasting without grasping the underlying drivers might offer short-term clarity but ultimately compromises long-term growth potential.

Let’s draw on a real-world scenario to illustrate this complexity. Imagine overseeing Mercado Libre’s marketing operations in Brazil as of January 2023. Amidst setting fresh annual goals, a major competitor, Americanas, reels from an accounting scandal, virtually halting its operations. Hopefully, you still had some champagne left from New Year's Eve. This disruption likely resulted in a surge in your sales volumes and a drop in average cost per acquisition across various categories. Was this uptick in performance a testament to your branding team’s efforts in establishing top-of-mind awareness, or was it due to strong performance campaigns? Or was it your logistics team’s scalable network innovations? And why didn't competitors like MagaLu capitalize on this market gap? Attribution models will surely claim more marketing driven conversions and may offer some insights, but the precision required to perfectly allocate credit among these dynamics remains elusive.

Behind closed doors, attribution is as much about crafting a compelling narrative as it is about number-crunching. The best leaders I've seen in corporations excel not just in business management but particularly in shaping and selling an attribution story during hard times, skillfully decoupling their team’s responsibilities from adverse outcomes.

Historically, the benchmark for understanding causality has been the randomized controlled trial, a standard established by Ronald Fisher in the 1930s that has sparked transformative insights across fields ranging from medicine to the social sciences. This methodology has proven pivotal in challenging established paradigms across various industries, with online advertising being no exception. An example of this is Meta’s strategic initiatives aimed at countering Google's search dominance, seeking to demonstrate that Facebook and Instagram value aren’t fully captured by Last Click attribution, while Search ads are over-attributed.

Notably, experiments were also conducted internally at Facebook. Recalling an initiative from a decade ago, the mid-market sales team conducted a randomized trial by adding account manager support to brands, which revealed a significant lift in sales attributed to this intervention. However, such excercises raise numerous questions about their broader applicability and ethical considerations. More importantly, they are difficult to scale and extrapolate results across different times and geographies. Leaders may receive some ground truth data points occasionally, but generally, they have to make decisions based on proxies and intuition.

Moreover, the granularity of experiments often fails to capture the full spectrum of variables at play in Marketing. Issues such as peer effects, long-term ramifications, product cannibalization, and the pre-emptive pulling forward of sales are just some of the dynamics that traditional lift tests gloss over. These elements introduce significant complexity, suggesting that the neat conclusions drawn from such experiments might be overly simplistic. Delving into the technical minutiae of these limitations would take us too far afield today, but it’s a topic that demands further exploration.

As we delve deeper into an AI-driven era, the complexity of attribution only intensifies. AI algorithms are now at the forefront, driving everything from SEO strategies to creative content generation. The competitive landscape is evolving too, with AI systems not just supplementing but competing against each other across different brands. Each brand leverages its own set of sophisticated algorithms, striving to outmaneuver its rivals in a high-stakes game where digital supremacy is the prize. This evolution prompts a critical question: do we relinquish control and let algorithms dictate our strategies, or do we strive to maintain an understanding—and thereby control—over these powerful levers? The balance between automated efficiency and strategic oversight is becoming the pivotal battleground in marketing warfare, as AI continues to reshape the rules and tools of competition.

The path forward involves acknowledging the omnipresence of attribution challenges and the pervasive influence of endogeneity in our analyses. In this new era, the ability to craft, test, and refine robust business hypotheses will become paramount. While traditional experiments will remain valuable, they are insufficient on their own to validate complex business strategies fully. Instead, a triangulation of methodologies, as suggested by Google & Facebook in the marketing sphere, informed by a deep business understanding, will be necessary.

In conclusion, recognizing that we are enmeshed in attribution challenges clearly delineates our next steps: effectively frame these issues, move beyond the simplistic last-click mindset, and embrace the nuanced insights provided by advanced methodologies. Last but not least, don’t hesitate to strategically position yourself in front of the last click to reap the benefits of legacy measurement systems within organizations.

Final Remarks:

At Bunker DB , we are dedicated to developing software that enables companies to harness the extensive benefits of artificial intelligence. Our tools are designed to robustly identify causal relationships in marketing, ensuring that our clients remain firmly in the driver's seat.

The title of this discussion is unmistakably inspired by the famous thesis of Eric Seufert , one of the brightest minds in Martech.?

If you appreciated the style of this note, most of the credit should rightfully go to more than just me. I refined my initial draft using ChatGPT, specifically channeling the insight and expertise of Ben Thompson, a leading thinker in the strategy field.

pd. Another great example: ?Norton won the lottery ticket.