Why Attribution is More Art Than Science

Attribution modeling is a notoriously confusing marketing topic. It attempts to be a deterministic science, yet it heavily depends on subjective decisions such as model selection, lookback windows, and handling data-tracking limitations. Because of this, attribution is as much an art as it is a science. Unfortunately, this point is typically overlooked by marketers who solely rely on attribution for measuring marketing impact and are unaware of its major pitfalls. Lacking this awareness can make or break a marketing strategy. If you want to scale growth successfully, you need to understand where it is lying to you and why that is. Here are ten of attribution modeling’s biggest flaws.

10 Reasons Your Attribution is Flawed

1. Data will always be incomplete - Attribution requires perfection to be accurate, but it is impossible to capture every individual's data point in a purchasing journey. From missing view-through attribution data to social media in-app browsers stripping tracking mechanisms and creating “dark traffic” disguised as organic, there are an endless number of reasons like these and the remaining list items that prevent your data from being complete enough.
2. Not every exposure leaves a digital trail - Digital advertising might be the largest form of advertising in the world, but it isn’t the only form of advertising. Traditional media like out-of-home (OOH), radio, and linear TV lack clean, traceable data points that models require. Additionally, there are untraceable influences like word-of-mouth that have a sizeable impact on purchasing decisions, but are invisible to this model.
3. Short attribution windows are blind to brand awareness initiatives - Traditional attribution models have a lookback window of 90 days fewer. This setup systematically removes top-of-funnel brand awareness efforts, which have an adstock (delayed) effect that can’t be weighted fairly in these models. 
4. The illusion of cross-device tracking - Attribution only works perfectly if you can piece together every digital and physical touchpoint for a given customer. Something that is not possible with modern privacy regulations and browser/device tracking limitations. This means that identity stitching is an incredibly difficult exercise to complete. Even if you were to solve this problem, the typical user jumps between personal, shared, and work devices throughout a purchase journey, making it unrealistic that the problem can ever be solved.
5. Paid advertising favoritism - Because paid advertising has direct financial implications on the business, marketers naturally build rigorous tracking systems to track these data points. However, because organic efforts like word-of-mouth, referrals, and baseline brand strength are harder to track, they tend to be invisible. As a result, the model's distribution favors paid channels with the best tracking, rather than the touchpoints that might have a greater impact.
6. Inflation of bottom-of-funnel influence - Attribution models organically overindex bottom-of-funnel due to the difficulty of digitally tracking top-of-funnel awareness. This creates a myopic view of growth performance and guides the marketers towards bottom-of-funnel optimization and away from feeding the growth engine with top-of-funnel brand awareness.
7. “Walled garden” attribution - Every digital publisher, like Meta, Google, and TikTok, operates within a self-contained tracking ecosystem. Their internal tracking systems excel at tying their internal data with the traffic on a website through the use of pixels and server-side tracking, but they operate in silos. They do not have visibility into the touchpoints coming from other ad networks and marketing initiatives, which leads each platform to claim 100% of the credit for a shared conversion.
8. Humans aren’t averages - A key aspect of a marketer’s job is to find patterns in consumer behaviors, distill them into averages, and create customer segments. However, these averages oversimplify the messy, unique nature of human decision-making. When you market to a generalized segment identified with an attribution model, you risk optimizing your messaging for the wrong audience entirely.
9. Misguided assumptions lead to wasted spend and poor performance - Proper attribution requires a mix of technical tracking prowess, mathematical rigor, and a deep understanding of customer behavior. Naively built models using arbitrarily selected lookback windows, out-of-the-box reporting, and a lack of recognition of your true sales cycle will lead you to flawed assumptions. Flawed assumptions result in misguided decisions that waste spend and limit growth.
10. Attribution ≠ Incrementality - One of the most dangerous misconceptions with attribution is that it is treated as causality. Finance teams and executives often view attribution reports as absolute, objective truth about what is and isn’t working with their marketing. However, attribution is a relative accounting method. It tells you which touchpoints a converting user interacted with, but it cannot prove that the conversion wouldn’t have happened if it weren’t for that ad exposure. An important and impactful data point that is captured with incrementality and is grounded in truth, rather than chance.

How to Balance the Art and Science

Despite these flaws, attribution isn’t useless. It remains a foundational requirement for real-time reporting and ad optimization. The goal isn’t to abandon it due to its flaws, but to validate it with a layered measurement framework with other techniques like experimentation and modeling. 

When you blend attribution with experimentation and modeling, you turn marketing data into a strategic roadmap. A system I called the “Suite of Truth”.

• Attribution for day-to-day directional optimizations at a tactical level.
• Experimentation (Incrementality) to prove true causality. These insights will adjust and ground the attribution weights.
• Media Mix Modeling (MMM) provides a cross-channel model that captures the missing attribution links, like offline media and external variables, to serve as a true impact distribution of performance.

Three distinct measurement approaches that, when combined, bridge the gap between art and science, and give your team more confidence in making smarter strategic decisions and a high likelihood of growth success.