Marketing Attribution Has a First Principles Problem

Marketing attribution has a first principles problem. Not a tooling problem, not a data problem, and not a budget problem. The models we use to assign credit for conversions are built on assumptions that were never quite true, and most teams have been too busy optimising within those models to question whether the models themselves are sound.

That distinction matters. You can have perfect data flowing into a broken framework and still walk away with the wrong answer. The challenge with attribution is not that we lack information. It is that we have been asking the wrong questions of it for a long time.

Why Attribution Is a Harder Problem Than It Looks

When I was running an agency and managing significant paid media budgets across multiple clients, one of the most common conversations I had with senior clients was about attribution. They wanted to know which channels were working. A reasonable question. The problem was that what they were really asking, and what attribution was actually answering, were two different things.

Attribution tells you which touchpoints appeared in the path before a conversion. It does not tell you which touchpoints caused the conversion. That gap, between observed sequence and actual causation, is where most attribution thinking falls apart. And it falls apart quietly, because the models still produce numbers. They look like answers. They feel like answers. They are not always answers.

The deeper issue is that attribution was designed to solve a budget justification problem, not a measurement problem. Someone in finance wanted to know why the display budget existed. Someone in performance wanted to defend their paid search spend. Attribution became the mechanism for that conversation, and the industry built increasingly sophisticated models around a fundamentally political question rather than a scientific one.

If you want to think more clearly about how measurement fits into your broader analytics practice, the Marketing Analytics hub at The Marketing Juice covers the full landscape, from GA4 implementation to incrementality testing and beyond.

The Causation Problem Nobody Wants to Talk About

Here is the core issue with every attribution model, from last-click to data-driven: they all observe what happened, and then assign credit based on rules or patterns. None of them can observe what would have happened if a touchpoint had not existed. That counterfactual, the world where you did not run that display campaign or send that email, is invisible to attribution systems by design.

This is not a minor technical limitation. It is the central flaw. A customer who saw a retargeting ad and then converted two days later may have converted anyway. They were already in the consideration phase. The ad appeared in their path, but it did not create the path. Attribution credits it. The budget stays. The model reinforces itself. And nobody asks whether the conversion would have happened regardless.

I saw this play out repeatedly when I was managing large-scale performance campaigns. Retargeting consistently showed strong attributed returns. It almost always does, because retargeting reaches people who have already demonstrated intent. When we ran proper holdout tests on some of those campaigns, the incremental lift was materially lower than the attributed numbers suggested. The attribution was not lying exactly. It was just answering a different question than the one we thought we were asking.

The MarketingProfs piece on preparing for web analytics makes a point that has aged well: the failure mode in analytics is usually not a lack of data, it is a lack of clarity about what question you are trying to answer before you start collecting it. Attribution suffers from exactly this problem at scale.

What Multi-Touch Attribution Actually Solves

Multi-touch attribution was a genuine improvement over last-click. Last-click was absurd. Giving 100% of the credit to the final touchpoint before conversion was always a simplification so aggressive it bordered on fiction, particularly for considered purchases with long research cycles.

But multi-touch did not solve the causation problem. It redistributed the credit problem. Instead of one channel claiming all the credit, multiple channels now shared it according to a weighting scheme. Linear, time-decay, position-based: each of these is a different political settlement, not a different insight into what actually drove the sale.

Data-driven attribution, which Google now defaults to in GA4, is more sophisticated. It uses machine learning to weight touchpoints based on their observed correlation with conversion outcomes. That is better than arbitrary position-based rules. But it is still a correlation model. It still cannot observe the counterfactual. And it is trained on your own conversion data, which means it inherits whatever biases and gaps exist in your tracking setup.

The Moz overview of Google Analytics alternatives is worth reading not because you should necessarily switch tools, but because it forces you to think about what different platforms are actually measuring and where their assumptions differ. That exercise alone tends to surface how much of what we treat as measurement is actually modelling.

The Observer Effect in Marketing Measurement

There is a subtler problem that gets less attention: the act of tracking changes the behaviour being tracked. This is not a philosophical abstraction. It has practical consequences for attribution data.

When you use cookies to track a user across a experience, you are measuring a cookied user’s experience. Users who clear cookies, browse in private mode, switch devices, or use ad blockers are either invisible or appear as different users. The experience you are attributing is not the experience your customers take. It is the experience of the subset of customers whose behaviour is legible to your tracking infrastructure.

This matters more than most teams acknowledge. If your highest-value customers are also your most privacy-conscious ones, and in many B2B and financial services categories that correlation is real, then your attribution data is systematically underweighting the touchpoints that influence them. You are optimising for the customers you can see, not necessarily the customers who matter most commercially.

I spent a significant part of my agency career working in financial services and other regulated categories where customers were particularly sensitive about data. The gap between what our tracking showed and what clients knew from their own CRM data was often striking. The attribution model said one thing. The actual revenue source said another. We learned to treat the attribution as directional intelligence, not gospel.

The Cross-Device and Cross-Channel Problem

Modern purchase journeys span multiple devices, multiple sessions, and multiple channels that operate in entirely different measurement environments. A customer might discover your brand through a podcast, research you on their phone, read a review on a third-party site, see a paid social ad on their tablet, and convert on their laptop after clicking a branded paid search ad.

Your attribution model will credit the branded paid search click. The podcast, the organic research, the review site, the social ad: all invisible or partial. This is not a failure of your specific setup. It is a structural limitation of digital attribution. The parts of the customer experience that happen outside your tracked environment, which is often the majority of the experience for considered purchases, simply do not appear in your data.

The practical consequence is that channels which operate earlier in the consideration cycle, brand advertising, content, PR, word of mouth, tend to be systematically undervalued by attribution models. Channels that operate at the point of decision, paid search in particular, tend to be systematically overvalued. This is not a new observation, but it remains one of the most consequential and least acted-upon insights in performance marketing.

When I was growing an agency from around 20 people to over 100, one of the recurring tensions I managed was between the performance team, who lived in attribution data and could point to clear numbers, and the brand team, whose work was harder to quantify. The performance team almost always won the budget argument in the short term. In the long term, the clients who invested in brand alongside performance consistently outperformed those who optimised purely for attributed return. The attribution model was not wrong about what it measured. It was incomplete about what mattered.

Why the Data-Driven Label Creates False Confidence

One of the more damaging trends in marketing analytics is the conflation of “data-driven” with “accurate.” Data-driven attribution is still a model. It makes assumptions. It has blind spots. Calling it data-driven does not make it more truthful than a simpler model. It may just make it harder to interrogate, because the complexity of the machine learning process obscures the assumptions underneath.

This is where critical thinking becomes essential, and it is the skill I find most consistently underdeveloped in marketing teams. The ability to look at a number produced by a sophisticated system and ask: what assumptions is this built on, and are those assumptions valid in my context? That question is harder to ask when the model is a black box, and the industry has an incentive to keep it that way. Complexity is good for vendors. Clarity is good for clients.

The Forrester perspective on automating marketing dashboards touches on a related tension: the more automated your reporting becomes, the more you need human judgement to interpret it correctly. Automation amplifies whatever assumptions are baked into the system. If those assumptions are wrong, automation makes you wrong faster and at greater scale.

I judged the Effie Awards, which evaluate marketing effectiveness based on demonstrated business outcomes. What struck me reviewing entries was how many campaigns had strong attribution data and weak business results, and how many had weak attribution data and strong business results. The correlation between attributed performance and actual commercial impact was not as tight as the industry would like to believe.

What Honest Measurement Actually Looks Like

None of this means attribution is useless. Used carefully, with appropriate scepticism about its limitations, it provides genuine directional value. The problem is not the tool. It is the misplaced confidence in the tool.

Honest measurement starts with acknowledging what attribution can and cannot tell you. It can tell you which touchpoints appeared in the conversion paths of tracked users. It can help you compare channels on a consistent basis within its own framework. It can surface patterns that are worth investigating further. What it cannot do is tell you what caused a conversion, what would have happened without a given touchpoint, or what is happening in the parts of the customer experience that fall outside your tracking.

The teams that use attribution well treat it as one input among several. They combine it with revenue data from their CRM, with brand tracking, with periodic incrementality tests, and with the kind of qualitative customer insight that no analytics platform can replicate. They ask their customers how they heard about them. They look at what happens to their business when they turn channels off, not just what the model says. They maintain healthy scepticism about numbers that look too good.

Forrester’s take on what to do once you have a marketing dashboard is worth reading in this context. The point that resonates most is that a dashboard is a starting point for questions, not a source of answers. The same applies to attribution reports. They tell you where to look. They do not tell you what you will find when you look there.

The MarketingProfs framework for building a marketing dashboard makes a point that applies directly here: the metrics you choose to display are themselves a set of assumptions about what matters. Attribution is no different. The model you choose encodes a theory of how your marketing works. The question worth asking regularly is whether that theory is still sound.

The Practical Implications for How You Run Measurement

If you accept that attribution has a first principles problem, the question is what to do about it practically. A few things follow logically.

First, stop treating attribution as a source of truth and start treating it as a hypothesis generator. When attribution tells you that a channel is underperforming, that is a prompt to investigate, not a verdict. When it tells you a channel is overperforming, the same applies. The model is pointing at something. You still need to figure out what.

Second, invest in measurement methods that can address the counterfactual question, even imperfectly. Incrementality testing, geo-based experiments, and holdout groups are not perfect either, but they are asking a fundamentally better question than attribution models ask. They are trying to measure what changes when you do or do not run a campaign, rather than who appeared in a conversion path.

Third, maintain a healthy relationship between your marketing data and your business data. Revenue, margin, customer lifetime value, retention: these are the outcomes that matter. If your attribution model is telling you one story and your P&L is telling you another, believe the P&L. The attribution model is a model. The P&L is the business.

Fourth, build critical thinking into how your team engages with data. Not cynicism, not paralysis, but the habit of asking what assumptions underlie a number before acting on it. That habit is rarer than it should be. I have hired a lot of analysts and performance marketers over the years, and the ones who asked “how was this calculated?” before “what should we do about it?” were consistently the more valuable ones. The answer to the first question usually changes the answer to the second.

For a broader perspective on building measurement practices that hold up under scrutiny, the full range of articles in the Marketing Analytics section covers everything from GA4 configuration to experimental design. The thread running through all of it is the same: tools are only as useful as the thinking behind them.

About the Author

Keith Lacy is a marketing strategist and former agency CEO with 20+ years of experience across agency leadership, performance marketing, and commercial strategy. He writes The Marketing Juice to cut through the noise and share what works.

Frequently Asked Questions