Marketing Attribution Is Broken. Here’s Why It Stays That Way.

Marketing attribution is the attempt to assign credit for a conversion to the marketing touchpoints that preceded it. The challenge is that this sounds more tractable than it is. Real buying decisions involve multiple channels, long time horizons, and human behaviour that no model fully captures, which means every attribution system is an approximation at best and a politically convenient fiction at worst.

Most marketers know this. The uncomfortable part is that the industry keeps selling attribution as a solved problem, and organisations keep buying dashboards that confirm whatever the buyer already believed.

If you are working through how to build a more honest measurement practice, the Marketing Analytics hub covers the full landscape, from GA4 configuration to measurement frameworks and channel-specific measurement problems.

Why Attribution Has Never Been as Reliable as the Tools Suggest

Early in my career, I was managing a paid search account at lastminute.com. We ran a campaign for a music festival, and within roughly a day we had driven six figures of revenue. It felt clean. Someone clicked an ad, they bought a ticket, the revenue was attributed to paid search. Done.

But even then, I knew that picture was incomplete. Some of those buyers had seen an email. Some had heard about the festival through a friend. Some had been on the site three times before clicking the ad that finally converted them. Paid search got the credit because it was the last touchpoint we could see, not because it was the only one that mattered.

That was the early 2000s. The tools have changed enormously since then. The underlying problem has not.

Attribution is hard for reasons that are structural, not technical. Customers do not move through a tidy funnel. They research products across devices, read reviews on platforms you do not own, talk to colleagues, and then convert weeks later in a way that your analytics platform records as a direct visit. Forrester has written extensively about how measurement frameworks can actively undermine the buyer’s experience by forcing non-linear behaviour into linear models. The data confirms what practitioners have long suspected: the models we use are not neutral. They shape what we see, and therefore what we fund.

The Last-Click Problem Is Not a Technical Glitch. It Is a Political Choice.

Last-click attribution assigns 100% of the credit for a conversion to the final touchpoint before the sale. It has been widely criticised for decades. It is still the default in most organisations.

That tells you something important about how attribution decisions actually get made.

Last-click persists not because anyone genuinely believes it is accurate, but because it is auditable, simple to implement, and easy to defend when someone asks why you spent money on a particular channel. It produces a number. The number has a clear lineage. In a budget meeting, that matters more than epistemological rigour.

The channels that benefit most from last-click are typically brand search, retargeting, and direct. These are channels that capture demand which was often created elsewhere. Channels that build awareness, consideration, and intent, things like display, content, social, and email, tend to be systematically undervalued because they rarely get the last click. Over time, last-click attribution pushes budget toward demand capture and away from demand creation. The short-term numbers look fine. The long-term brand health does not.

Understanding the theoretical foundations behind this matters. Attribution theory in marketing explains why different models exist and what assumptions each one bakes in. Most practitioners skip this and go straight to implementation, which is part of why the same mistakes repeat across organisations.

Multi-Touch Attribution Fixes Some Problems and Creates Others

The logical response to last-click’s distortions is to distribute credit across multiple touchpoints. Linear attribution gives equal weight to every touchpoint. Time-decay models give more credit to touchpoints closer to conversion. Position-based models weight the first and last touches heavily and distribute the remainder across the middle.

These are all improvements over last-click in the sense that they acknowledge the customer experience has more than one moment. But they introduce their own distortions.

Linear attribution assumes every touchpoint contributes equally, which is almost certainly false. A brand awareness display impression and a product comparison page visit are not equivalent contributions to a sale. Time-decay models assume recency equals relevance, which may be true for impulse purchases and false for considered B2B decisions that take months. Position-based models are essentially a compromise between first-touch and last-touch, which means they inherit the weaknesses of both.

Data-driven attribution, which uses machine learning to assign credit based on actual conversion path patterns, is theoretically superior. In practice, it requires significant data volume to produce reliable outputs, it is a black box that is difficult to audit or explain to a CFO, and it still cannot account for touchpoints that happen outside your tracking ecosystem.

I have sat in agency reviews where a client’s data-driven attribution model was confidently presented as the truth about their customer experience. Nobody in the room asked what the model could not see. That is the question that matters most.

What Attribution Models Cannot See

Every attribution model is bounded by its tracking infrastructure. It can only assign credit to touchpoints it can observe. The problem is that a substantial portion of the customer experience is invisible to standard analytics.

Word of mouth is the obvious example. If a colleague recommends your product and that recommendation drives a purchase, your attribution model records a direct visit or a branded search. The word-of-mouth touchpoint does not exist in your data. It gets no credit. You have no way of knowing it happened.

Offline touchpoints present the same problem. A trade show conversation, a print ad, a sales call, a physical store visit: these influence decisions and rarely appear in digital attribution. Connecting them to online conversions requires deliberate effort, unique tracking mechanisms, and often a degree of inference that most organisations are not willing to invest in.

Cross-device journeys are a persistent technical problem. A customer who researches on mobile and converts on desktop looks like two different users to most analytics platforms. The touchpoints on the mobile device do not get connected to the conversion on desktop. GA4’s cross-device reporting is an improvement, but it depends on users being logged in, which limits its coverage.

Privacy changes have compounded all of this. The deprecation of third-party cookies, increased use of ad blockers, iOS privacy changes, and stricter consent requirements have collectively reduced the observable portion of the customer experience. Understanding what Google Analytics goals are unable to track is a useful starting point for mapping the gaps in your own setup, because the gaps are larger than most practitioners assume.

The Budget Misallocation Problem Compounds Quietly

The reason attribution matters is not the dashboard. It is the budget decisions that flow from the dashboard.

When I was running an agency and growing the team from around 20 people to over 100, one of the things I watched closely was how clients made channel investment decisions. The ones who relied on last-click attribution consistently underinvested in upper-funnel activity. They would cut brand awareness budgets because the attribution model showed poor ROI, then wonder six months later why their cost per acquisition was rising and their brand search volume was declining. The two things were connected. The attribution model could not show the connection.

This is the compounding problem with bad attribution. Each budget cycle, money shifts toward channels that look good in the model. Those channels often capture demand that was created by the channels being defunded. Demand creation shrinks. Demand capture gets more expensive. The attribution model continues to show the demand capture channels performing well because they are still getting the last click. The cycle continues until something breaks visibly enough to force a rethink.

Forrester’s research on sales and marketing measurement alignment makes a related point: measurement frameworks that are optimised for marketing efficiency can actively work against the broader commercial goals of the business. Attribution is not just a marketing problem. It shapes how the entire revenue function allocates its resources.

Incrementality Testing Is More Honest Than Attribution Modelling

If attribution models are approximations built on incomplete data, incrementality testing is a different approach to the same question. Instead of asking “which touchpoints preceded this conversion,” it asks “would this conversion have happened without this channel or campaign?”

Incrementality testing works by creating a holdout group: a set of users or markets that are not exposed to a particular channel or campaign. You then compare conversion rates between the exposed group and the holdout group. The difference is the incremental contribution of that channel.

This approach is methodologically stronger than attribution modelling because it measures actual causal contribution rather than correlation with conversion paths. It is also harder to run, requires sufficient scale to produce statistically meaningful results, and cannot be done continuously across all channels simultaneously.

Measuring affiliate marketing incrementality is a good concrete example of this in practice. Affiliate is a channel where the gap between attributed revenue and incremental revenue is often very large, because affiliates frequently intercept customers who were already going to convert. Incrementality testing reveals that gap. Attribution modelling hides it.

The same logic applies across channels. Retargeting, brand search, and email are all channels where attributed revenue tends to significantly overstate incremental contribution. Running holdout tests on these channels is uncomfortable because the results often show that a portion of the attributed revenue would have happened anyway. But knowing that is more useful than not knowing it.

Emerging Channels Make Attribution Harder, Not Easier

Every new channel that enters the marketing mix adds complexity to the attribution problem. The channels that are growing fastest right now are also the ones that are hardest to measure.

Generative AI is beginning to influence how people discover and evaluate products, but the touchpoints happen inside AI interfaces that do not pass referral data in the way a standard web visit does. Measuring the success of generative engine optimisation campaigns requires different thinking because the standard attribution infrastructure does not capture these interactions at all.

AI avatars and synthetic influencers are another example. The engagement metrics are measurable. The downstream contribution to purchase is much harder to attribute. Measuring the effectiveness of AI avatars in marketing requires moving beyond reach and engagement to look at brand lift, consideration shifts, and downstream conversion behaviour across cohorts. That is a more sophisticated measurement challenge than most organisations are currently set up to handle.

The pattern is consistent across emerging channels: the technology moves faster than the measurement infrastructure. Organisations that invest in new channels without a clear measurement approach end up either over-crediting or under-crediting them, and making budget decisions based on whichever error they have made.

Marketing Mix Modelling as a Complement, Not a Replacement

Marketing mix modelling, sometimes called econometric modelling, takes a different approach to the attribution problem. Rather than tracking individual user journeys, it uses statistical analysis of aggregate data to estimate the contribution of different marketing inputs to business outcomes.

MMM has several advantages over user-level attribution. It can incorporate offline channels, it does not depend on cookies or device-level tracking, and it can account for external factors like seasonality, competitor activity, and macroeconomic conditions. It is also more resilient to the privacy changes that are degrading user-level tracking.

The limitations are real. MMM requires substantial historical data, typically two or more years, to produce reliable outputs. It operates at an aggregate level, which means it cannot inform decisions about individual campaigns or creative executions. It is slow, expensive to build properly, and the outputs are directional rather than precise. And like all models, it reflects the assumptions of whoever built it.

The most sophisticated measurement approaches use MMM and user-level attribution together, with incrementality testing as a third check. Each method has blind spots. Used in combination, they produce a more complete picture than any single approach can. HubSpot’s distinction between marketing analytics and web analytics is relevant here: the goal is to understand commercial impact, not just measure digital activity.

The Organisational Dynamics That Keep Bad Attribution in Place

Bad attribution persists partly because it is technically difficult to fix, and partly because it is politically convenient to leave alone.

Channel owners have a vested interest in the attribution model that makes their channel look best. Paid search teams prefer last-click. Social teams prefer first-touch or position-based models that give credit for awareness. Email teams want multi-touch models that count their touchpoints. The attribution debate in most organisations is not primarily a measurement debate. It is a budget debate conducted through the language of measurement.

I have been in those rooms. The conversation about which attribution model to adopt rarely starts from first principles about what is most accurate. It starts from the current budget allocation and works backward to find a model that justifies it. When I was turning around a loss-making agency, one of the first things I looked at was whether the measurement framework was designed to find the truth or to defend existing decisions. In most cases, it was the latter.

MarketingProfs has made the point that failing to prepare in web analytics is preparing to fail, and the preparation they mean is not technical. It is strategic. It is deciding in advance what questions you are trying to answer and what decisions the data will inform, before you build the dashboard. Most organisations do this in reverse.

The way to break this cycle is to separate the measurement question from the budget question. Run the measurement work independently. Define what good attribution would look like before you know what it will show. Bring in external validation where possible. And be willing to act on results that are inconvenient.

What Honest Attribution Practice Actually Looks Like

Honest attribution does not mean finding the perfect model. It means being explicit about what your model can and cannot see, and making decisions that account for the gaps.

That starts with documentation. Every attribution setup should have a clear record of what it tracks, what it cannot track, and what assumptions the model makes. Most organisations do not have this. They have a dashboard. The dashboard has numbers. The numbers are treated as facts rather than estimates.

It also means running regular incrementality tests on your highest-spend channels. If you are spending significantly on retargeting or brand search, you should know what the incremental contribution of those channels actually is. The answer will probably be lower than your attribution model suggests. That is useful information, not a problem to be hidden.

Inbound marketing creates its own measurement challenges in this context. Measuring inbound marketing ROI requires connecting content and organic touchpoints to revenue outcomes across long consideration cycles, which standard attribution models handle poorly. The content that educated a buyer six months ago rarely gets credit for the sale that happens today.

Buffer’s framework for content marketing metrics offers a useful way to think about this: measuring content at multiple stages of impact, from reach and engagement through to downstream conversion influence, rather than expecting a single attribution credit to capture the full contribution.

The goal is not to achieve perfect attribution. That is not achievable. The goal is to make better decisions than you would make with no measurement, while being honest about the uncertainty in your estimates. That is a lower bar than the industry often sets, but it is a more honest one.

Early in my career, when I was refused budget for a new website and built it myself instead, I learned something that has stayed with me ever since: the constraint forces you to be clear about what you actually need versus what you think you need. Attribution is similar. The organisations that get the most value from it are not the ones with the most sophisticated models. They are the ones who are clearest about what question they are trying to answer.

For a broader look at the measurement challenges that sit alongside attribution, including how to approach channel-specific measurement and analytics infrastructure, the Marketing Analytics hub brings together the full range of topics in one place.

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