Marketing Revenue Attribution: Stop Measuring the Wrong Thing

Marketing revenue attribution is the process of assigning credit for revenue to the marketing activities that contributed to it. Done well, it tells you which channels, campaigns, and touchpoints are actually driving commercial outcomes. Done badly, which is how most businesses do it, it tells you a compelling story that happens to be wrong.

The problem is not the data. The problem is the model sitting on top of it, and the assumptions baked into that model that nobody ever questions.

Why Attribution Is a Commercial Problem, Not a Technical One

I have sat in a lot of boardrooms where attribution came up, and the conversation almost always went the same way. Someone from finance wanted to know exactly which channel drove which pound of revenue. Someone from the agency wanted to show a chart that made their channel look essential. And someone from the marketing team was trying to reconcile the two without getting fired.

The framing of attribution as a technical problem, something to be solved with the right tool or the right model, misses the point. Attribution is a commercial problem. It is about how you allocate budget, how you justify investment, and how you make decisions under uncertainty. The technical implementation matters, but it serves the commercial question, not the other way around.

When I was running iProspect and we were managing hundreds of millions in paid media spend across 30-plus industries, the attribution question was never really about which platform’s reporting was most accurate. It was always about whether the business was making sensible budget decisions based on the information available. Those are different questions, and conflating them leads to expensive mistakes.

If you want to understand how attribution fits into a broader measurement framework, the Marketing Analytics hub at The Marketing Juice covers the full landscape, from GA4 implementation through to commercial reporting and measurement strategy.

What the Common Attribution Models Actually Do

There are five attribution models most marketers encounter in practice. Each one makes a different assumption about how credit should be distributed across the customer experience, and each one produces a different answer from the same data.

Last-Click Attribution

Last-click gives 100% of the credit to the final touchpoint before conversion. It is the default in many analytics platforms and the model most businesses are still using. It is also the model most likely to mislead you, because it systematically rewards the channel that happens to be present at the moment of conversion, regardless of what drove the customer to that point.

Paid search captures a disproportionate share of credit under last-click because people often search for a brand name immediately before buying, after they have already been influenced by email, social, content, or word of mouth. The search click gets the credit. Everything else gets nothing. This is why paid search teams tend to look brilliant under last-click attribution and why brand and content teams tend to look expendable. It is not because they are expendable. It is because the model does not see what they do.

First-Click Attribution

First-click gives all credit to the first touchpoint. It is the mirror image of last-click and has the opposite problem. It overvalues awareness channels and undervalues the activity that actually closes the sale. Neither extreme is useful as a primary model, though first-click can be valuable as a secondary view when you are specifically trying to understand what is initiating customer journeys.

Linear Attribution

Linear attribution distributes credit equally across all touchpoints in the experience. It is more balanced than single-touch models, but it treats every interaction as equally important, which is rarely true. A display impression someone scrolled past in three seconds gets the same credit as the email that prompted them to click through and read a product page for ten minutes.

Time-Decay Attribution

Time-decay gives more credit to touchpoints closer to conversion, on the assumption that recent interactions are more influential. This logic is defensible for short sales cycles but becomes misleading for considered purchases where a piece of content consumed six weeks ago may have been the most influential moment in the entire experience.

Position-Based Attribution

Position-based, sometimes called U-shaped, splits the majority of credit between the first and last touchpoints, with the remainder distributed across the middle. It acknowledges both the initiation and the close, which makes it more commercially intuitive than most models. It is still a heuristic, but it is a more honest one.

The Case for Data-Driven Attribution (and Its Limits)

Google’s data-driven attribution model uses machine learning to assign credit based on the actual conversion paths in your account, rather than a fixed rule. In theory, this is closer to reality than any of the heuristic models. In practice, it requires significant conversion volume to produce reliable outputs, it is a black box that you cannot interrogate, and it only sees the touchpoints inside Google’s ecosystem.

That last point matters more than most people acknowledge. If a customer saw a LinkedIn ad, read a blog post, opened an email, and then clicked a Google search ad before converting, Google’s data-driven model sees one of those four touchpoints. It builds its model from incomplete information and presents the output with the confidence of complete information. That is not a criticism of the model. It is a structural limitation of any platform-level attribution.

Forrester has written about this tension directly, noting that much of what gets sold as marketing measurement is closer to snake oil than science. The platforms have an incentive to show you data that justifies continued spend on their platform. That does not make the data useless, but it does mean you should read it critically.

Multi-Touch Attribution: More Accurate, More Complex, More Expensive

Multi-touch attribution attempts to solve the platform-silo problem by stitching together data from across channels and applying a model to the full customer experience. Done properly, it gives you a much more complete picture of how marketing activity contributes to revenue. It also requires clean data infrastructure, consistent tagging, a reliable identity resolution approach, and someone with the analytical capability to build and maintain the model.

Most businesses are not set up to do this well. The tagging is inconsistent. The data sits in different systems. The customer identifiers do not match across channels. And the person responsible for analytics is already managing five other priorities. This is not a reason to abandon multi-touch attribution. It is a reason to be honest about where you are on the capability curve before you invest in a solution that requires capabilities you do not yet have.

There is also a measurement gap that no attribution model can close: offline influence. Word of mouth, a conversation at an event, a podcast someone listened to on the commute, a physical piece of collateral. These things influence purchasing decisions and they leave no trackable footprint. Any attribution model that claims to account for 100% of the factors driving revenue is either lying or measuring a very narrow slice of the world.

Understanding how to track the digital touchpoints you can see is still worth doing carefully. The Moz guide to GA4 custom event tracking is a useful reference if you are trying to build a more complete picture of on-site behaviour as part of your attribution setup.

Marketing Mix Modelling as an Alternative Frame

Marketing mix modelling, sometimes called MMM, takes a different approach entirely. Rather than tracking individual customer journeys, it uses statistical regression to estimate the contribution of different marketing inputs to overall revenue outcomes. It works at an aggregate level, does not require individual-level tracking, and is therefore less affected by privacy changes, cookie deprecation, and cross-device fragmentation.

MMM has been around for decades. It fell out of fashion when digital tracking made individual-experience attribution seem possible, and it has come back into fashion as it has become clear that individual-experience attribution is less reliable than it looked. The resurgence is not nostalgia. It is a rational response to the limits of digital-only measurement.

The limitations of MMM are different from those of multi-touch attribution. It requires a reasonable volume of historical data to produce reliable outputs. It is less useful for granular channel optimisation. And it is a lagging indicator, telling you what worked in the past rather than what is working right now. Used alongside digital attribution rather than instead of it, it provides a more complete picture than either approach alone.

Forrester’s perspective on how measurement approaches can undermine rather than support the buyer experience is worth reading in this context. The measurement framework you choose shapes the decisions you make, and some frameworks actively discourage investment in the activities that create demand.

The Incrementality Question Nobody Wants to Ask

Attribution, in all its forms, answers the question: which channel was present when a conversion happened? The more useful question, and the harder one, is: which channel caused a conversion that would not have happened otherwise?

This is the incrementality question, and most attribution models do not answer it. They measure association, not causation. A customer who was going to buy anyway, who did a branded search immediately before converting, will show up in your attribution data as a paid search conversion. The paid search click did not cause the purchase. It just happened to be in the path. But the model credits it as a conversion, and you keep spending on it, and the numbers look fine, and nobody asks whether the revenue would have come in anyway through organic search or direct traffic.

I saw this clearly during a paid search campaign I ran at lastminute.com for a music festival. The campaign generated six figures of revenue within roughly 24 hours, which looked extraordinary on the surface. But when we looked at the branded search volume, a significant proportion of those clicks were from people who already knew about the festival and were going to book regardless. The incremental contribution of the paid campaign was real, but it was smaller than the raw attribution numbers suggested. Understanding that distinction changed how we thought about bid strategy and budget allocation going forward.

Incrementality testing, through geo-based experiments, holdout groups, or conversion lift studies, is the most reliable way to answer the causation question. It is also more resource-intensive than running a report. Most businesses do not do it consistently. The ones that do tend to make materially better budget decisions over time.

How to Build a More Honest Attribution Practice

The goal of attribution is not perfect measurement. It is better decisions. These are not the same thing, and conflating them leads to paralysis or, worse, false confidence in numbers that do not deserve it.

A more honest attribution practice starts with acknowledging what your current model does and does not capture. If you are on last-click, you know it undervalues upper funnel. If you are on data-driven at the platform level, you know it cannot see off-platform touchpoints. That acknowledgment should be explicit in how you present and interpret the data, not buried in a footnote.

From there, a few practical principles hold across most contexts.

Use multiple models as lenses rather than picking one and treating it as truth. Running last-click and linear attribution side by side on the same data will show you which channels look different under different assumptions. That gap is informative. It tells you where the model choice is doing the most work.

Segment your attribution analysis by customer type and purchase stage. A new customer converting for the first time has a different experience profile than a repeat customer doing a routine repurchase. Mixing them together in the same attribution report produces averages that are not representative of either group.

Bring commercial judgment to the output. Attribution data should inform decisions, not make them. If your model is telling you to cut brand spend because it does not show up in last-click conversions, but you have seen what happens to paid search efficiency when brand investment drops, the model is missing something. Trust the pattern you have observed over time, not just the current report.

Invest in tagging hygiene before investing in a more sophisticated model. The most advanced attribution methodology in the world will produce unreliable outputs if the underlying event tracking is inconsistent. Getting your analytics setup to a reliable baseline is a prerequisite, not an afterthought.

For email specifically, the attribution picture is worth reviewing separately. Email attribution has its own set of measurement quirks, particularly around open tracking and click-to-conversion windows, that can distort how email performance appears in a multi-channel view.

Content performance adds another layer of complexity. Content marketing metrics rarely map cleanly onto revenue attribution because content often influences decisions long before any trackable conversion event occurs. That does not make content unmeasurable. It means the measurement approach needs to be appropriate to how content actually works in the funnel.

What Good Attribution Reporting Actually Looks Like

Good attribution reporting is not a single number. It is a structured view of the data that makes the assumptions visible and presents multiple perspectives alongside each other.

A reporting structure that works in practice includes a primary attribution view for operational decisions, a secondary view that challenges the primary (typically a different model applied to the same data), a separate incrementality view where you have test data available, and a commentary layer that explains what the numbers do not capture.

That last element is the one most often missing. The commentary layer is where you note that branded search is inflating paid search numbers, or that a channel looks weak in attribution because it operates primarily at the top of the funnel, or that last month’s spike in direct traffic coincides with a PR campaign that does not show up anywhere in the digital attribution model. Without that context, the numbers will be misread.

When I was building out the analytics function at iProspect, one of the things we worked hard on was making sure the reporting told a story that a commercial director could act on, not just a data scientist could admire. That meant being explicit about uncertainty, flagging where the model was likely to be wrong, and making recommendations that acknowledged the limits of what we could measure. It was a harder conversation to have than presenting a clean dashboard with confident numbers. But it was a more honest one, and over time it built more trust.

There is more on building measurement frameworks that hold up commercially in the Marketing Analytics hub, including how GA4 fits into a broader attribution and reporting setup.

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