No single attribution model has ever told me the truth about what drove revenue. Not last-click, not first-click, not the platform's own reporting, not even a well-built marketing mix model. After running into this problem enough times across enough accounts, I stopped looking for the one model that would finally answer the question, and started thinking about measurement differently.

Every model and every method is a lens. Each one shows you something real and distorts something else. The skill is not finding the perfect lens, because it does not exist. The skill is knowing which lens to pick up for which decision, and holding several of them at once so that the distortions of one are checked by the others.

This matters more as a media mix gets more complex. A brand running one channel can lean on that channel's reporting and be roughly fine. A brand running ten channels across the full funnel cannot, because the channels overlap, claim the same conversions, and interact in ways no single platform can see. What follows is how I actually approach measurement for that kind of brand: the buckets I sort the work into, and the tools that belong in each.

Throughout, keep one thing in mind. None of this closes every gap. The holes never fully disappear. The goal is not a perfect number, it is a composite picture honest enough to make good decisions from.

The one distinction that governs everything: in-platform versus cross-platform

‍

Before any specific model, there is a distinction that determines which tools are even valid: are you making a decision inside a single platform, or across several?

In-platform metrics are good, genuinely good, for in-platform decisions. Google's data-driven attribution, Meta's reported conversions, each platform's optimisation signals, these are built to help the platform's algorithm learn and to help you make choices within that platform. Which campaign to scale inside Google. Which creative to cut inside Meta. For that purpose, they work well, and you should use them.

The mistake, and it is one of the most common and most expensive in our field, is taking those in-platform numbers and applying them across a multi-source media mix. You cannot. Every platform is incentivised to claim credit, and every platform measures in its own walled garden using its own attribution window and its own view of the user. Add up what each platform says it drove and you will routinely generate more revenue on paper than the business actually made, because the same sale gets claimed two or three times over.

So the rule is simple to state and easy to forget. Use in-platform metrics to steer within a platform. The moment you are comparing across platforms or asking what the whole machine produced, you need a different set of tools: multi-touch attribution (MTA), incrementality testing, marketing mix modelling (MMM), and blended frameworks like marketing efficiency ratio (MER, total revenue divided by total ad spend). Those cross-platform methods are the back half of this article.

A worked example: the media mix we are measuring

To make this concrete, take a mid-sized ecommerce brand spending somewhere between five and ten million dollars a year across a mix like this: Meta Ads. Google Ads, spanning Search, Performance Max, Shopping, and Demand Gen. Microsoft (Bing) Ads across Search, Performance Max, and Shopping. Email. SMS. Organic Search. Organic Social.

That is a realistic modern mix, and it is deliberately awkward to measure, because it runs the full funnel. Some of it is demand capture, where someone already wants the product and is choosing where to buy. Some of it is demand generation, where you are creating interest that did not exist yet. No single attribution model handles both halves well, which is exactly the point.

Bucket one: click-based, lower-funnel traffic

‍

Start with the easiest traffic to attribute, because there is a clean, trackable click tied closely to a conversion. This is mostly your demand-capture activity: branded and non-branded Search, Shopping, much of your click-driven Google and Microsoft traffic.

Because the click is trackable and the path is short, click-based attribution models work well here. The nuance is that you do not use one of them, you choose among them based on where the campaign sits in the funnel and what question you are asking.

Last-click is appropriate when you are confident the traffic is genuinely low-funnel, the clearest example being branded search. When someone searches your brand name and clicks, that click is very close to the purchase decision, and giving it the credit is reasonable. Last-click is a fair lens for judging the efficiency of demand capture that sits right at the bottom.

First-click answers a different question: looking only at the click data, where did this journey start? It is useful when you want to understand which activity is opening relationships rather than closing them, and it is a better lens than last-click for judging prospecting-oriented click traffic. It still only sees clicks, so it is partial, but within that limit it tells you something last-click hides.

Google's data-driven attribution has its place too, specifically for in-platform, click-based Google Ads traffic. It distributes credit across the clicks in the Google-observed path using the account's own conversion patterns. Used inside Google, to compare click-driven campaigns against each other, it is more sophisticated than a single-touch rule. The caveat is the one from the section above: it is an in-platform lens, so it is for decisions inside Google, not for judging Google against Meta or against the blended business.

The theme of this bucket: for short, click-driven, lower-funnel paths, click-based models are legitimately useful, and the craft is matching the model to the funnel position and the question rather than picking a house model and applying it everywhere.

Bucket two: rich media and higher-funnel campaign types

‍

Now the traffic that click-based attribution measures badly: Video, Demand Gen, and other rich-media, upper-funnel campaign types whose job is to create demand rather than capture it.

The problem is structural. These campaigns work by influencing people who are not ready to click yet. Judge them by last-click, or even by clicks at all, and you will conclude they do almost nothing, because their effect shows up later and elsewhere, in branded searches that happen afterwards, in direct visits, in a general lift across the account. Attribution that only counts clicks systematically under-credits demand generation.

Meta is the clearest example of why this bucket matters, and it deserves a specific mention. Meta can operate across the whole funnel, and it does run lower-funnel formats like retargeting and catalog sales that convert on a trackable click. But its core competency, the thing it does better than almost anything else, is upper-funnel awareness and discovery. When that upper-funnel work is executed well, it creates a halo across your other channels: more people search your brand afterwards, more arrive direct, and performance improves everywhere in ways that Meta's own reporting will never fully show and that click-based attribution will miss entirely. This is exactly why Meta is so often under-credited by last-click. The lift is real, it is just not visible from inside any single platform, which is the whole reason the holistic methods in the next section exist.

So for these campaign types you widen what you are willing to look at. Views and view-through signals become directional evidence, not gospel, but useful as a read on whether the campaign is reaching and affecting people. When the goal of a Video or Demand Gen campaign is upper-funnel influence, assessing it partly on view-based engagement is more honest than forcing it into a click-based frame that was never going to capture its value.

For Demand Gen specifically, Google's platform comparable conversions metric is worth using when you are trying to assess that campaign type on a more consistent basis. It is still an in-platform view, with all the limits that carries, but for the narrow job of evaluating Demand Gen performance inside Google it is a more appropriate yardstick than standard last-click conversions.

The broader principle in this bucket: the campaign's goal decides which signal matters. A campaign built to generate demand should not be judged primarily by a metric built to measure demand capture. Match the measurement to the intent.

Bucket three: cross-platform and holistic methods, matched to data depth

‍

Everything so far helps you make decisions within platforms and within the click-trackable world. But the original question, what actually drove revenue across the whole mix, lives outside any single platform. This is where the heavier methods come in, and the right one depends heavily on how much data and confidence you have.

Multi-touch attribution attempts to stitch touchpoints across channels into a single view of the path to purchase. It is more holistic than any single-platform report, and useful for seeing cross-channel patterns. It also has real limits in a post-signal-loss world, since so much of the journey is now unobservable, so treat its output as one input rather than the answer.

Incrementality testing is, to me, one of the most valuable things you can do, because it targets the question the models only approximate: if I turned this off, or spent less here, what would actually change? Well-run incrementality tests, including through platforms like Northbeam, give you evidence about incremental value rather than attributed value, and the two can differ enormously. A channel can look highly efficient on last-click and contribute far less incrementally than its reporting suggests. Where you can run them and have the volume to trust them, incrementality tests are worth more than another attribution model.

Lift studies belong here too, used when you have enough data for the findings to mean something. A lift study on thin data produces a confident-looking result you should not rely on, so the precondition is volume, not just intent.

Correlation analysis is a legitimate part of the toolkit, looking at how movements in spend or activity in one area track with outcomes elsewhere. It is not causal on its own, and it is easy to over-read, but as a directional input alongside the others it earns its place.

More sophisticated causal analysis sits above that, and I want to be honest about it: this is often not something I would run myself. Proper causal work depends on both the data being strong enough and the person doing the analysis, frequently a data scientist rather than a media practitioner, being genuinely confident in the data and in the method. When those conditions are met, it is powerful. When they are not, it produces authority without accuracy, which is worse than no analysis at all. Knowing when to bring in that expertise, and when the data cannot support it, is part of the discipline.

Marketing mix modelling is the heaviest tool, and it comes closest to a true cross-channel answer when the conditions are right. But those conditions are demanding: MMM needs high spend per channel, sustained over enough time, to model contribution reliably. Below the very high budget thresholds it really requires, an MMM will still output clean-looking channel contributions, and they will not be trustworthy, because the model does not have enough signal to stand on. Used at genuine scale, MMM is extremely valuable and can get close. Used too early, it is false confidence in a nicer wrapper.

And underneath all of it sits MER, total revenue over total ad spend. It attributes nothing and cares about nothing except whether the whole machine is profitable. It is blunt, and it is very hard to fool, which makes it the essential sanity check against every model above. When your attribution models and your blended MER disagree, the MER is usually the one telling you the uncomfortable truth.

Putting it together on the example mix

‍

Back to the five-to-ten-million-dollar brand. Measuring it well does not mean choosing one of these methods. It means running several in parallel, each where it is appropriate.

You judge branded Search efficiency with last-click, because that traffic is genuinely bottom-funnel. You evaluate your click-driven Google campaigns against each other using data-driven attribution, inside Google. You assess Demand Gen and Video partly on view-based and comparable-conversion signals, because judging them on clicks alone would tell you to switch off the thing generating tomorrow's demand. You use incrementality testing to find out which channels are actually adding sales rather than claiming them. Where the brand's spend and history support it, an MMM gives you a top-down read on contribution. And MER runs across everything the whole time, the blunt instrument that keeps every other model honest.

No single one of those is right. Together, they give you a composite picture that is far closer to reality than any of them alone.

The honest conclusion

‍

There is no perfect attribution model, and there is no perfect combination of models either. Every method here has gaps and holes. In-platform metrics double-count across sources. Click-based models cannot see demand they created upstream. MTA cannot observe the parts of the journey that signal loss has hidden. MMM needs scale most brands cannot feed it. Even incrementality, the sharpest of them, is periodic and partial.

You are not assembling these tools to reach one true number, because that number does not exist. You are assembling them to paint a picture that is not forced or stuck inside any single model, and to reduce the uncertainty as far as it can honestly be reduced. The holes never fully close. Good measurement is not about pretending they do, it is about using every appropriate tool, in the situation it actually fits, so that what you cannot see is as small as you can honestly make it.

That is the real answer to what drove revenue. It is always a mix, measured with a mix, read by someone who knows the limits of each tool.

If you want a second set of eyes on how to measure your own media mix honestly, that is the kind of thing we help with. You can book a discovery call with me here.