Marketing Attribution Models: Which One Is Worth Using

Marketing attribution models assign credit to the channels and touchpoints that contribute to a conversion. The model you choose shapes how you read performance, where you invest budget, and which teams get rewarded. Get it wrong and you end up optimising for the wrong things.

There is no attribution model that tells you the complete truth. Every model is a simplification. The question is which simplification is least likely to mislead you given your business, your customer experience, and your data quality.

Why Attribution Matters More Than Most Teams Admit

Early in my career, I watched a business cut its display budget three quarters because the channel showed almost no last-click conversions. Paid search got the credit. Display got cut. Within two quarters, paid search volumes dropped noticeably and nobody could explain why. The display work had been priming audiences and the business had no way of seeing it.

That story is not unusual. Attribution decisions have real consequences for budget allocation, team structure, and in the end what kind of marketing a business ends up doing. A team rewarded on last-click metrics will optimise for last-click performance. Over time, that means investing heavily in channels that capture demand and underinvesting in channels that create it.

Attribution is not a technical footnote. It is one of the most commercially significant decisions in performance marketing, and it deserves more deliberate attention than most businesses give it. If you want broader context on how attribution fits into a wider analytics practice, the Marketing Analytics hub covers the full landscape.

What Are the Main Attribution Models?

There are six models you will encounter regularly. Each one makes a different assumption about where credit should sit.

Last-Click Attribution

All credit goes to the final touchpoint before conversion. Paid search and direct traffic tend to dominate under this model because they sit closest to the point of purchase. It is simple, easy to implement, and deeply misleading for any business with a long or complex customer experience.

Last-click is still the default in a surprising number of GA4 setups and ad platforms. It persists because it is easy to explain and easy to act on. That does not make it accurate.

First-Click Attribution

All credit goes to the first touchpoint. This model is useful if you are specifically trying to understand acquisition, meaning which channels are introducing new customers to your brand. It has the inverse problem of last-click: it ignores everything that happened after the first interaction.

Linear Attribution

Credit is distributed equally across every touchpoint in the experience. This sounds fair, and it is a reasonable starting point for businesses that have no strong reason to weight any particular stage. The problem is that it treats a brand awareness touchpoint from six months ago with the same weight as the retargeting ad someone clicked two hours before converting. Equal credit is not the same as accurate credit.

Time-Decay Attribution

More credit goes to touchpoints closer to the conversion. The logic is intuitive: the interaction that happened yesterday probably had more influence than the one that happened three months ago. This model works reasonably well for short sales cycles. For longer B2B journeys, it can still undervalue the early awareness work that started the relationship.

Position-Based Attribution

Also called U-shaped attribution. The first and last touchpoints each receive 40% of the credit, with the remaining 20% distributed across the middle touchpoints. This model acknowledges that acquisition and conversion are both important, which makes it more defensible than single-touch models. It is a reasonable default for businesses that want to value both ends of the funnel without building a full data-driven model.

Data-Driven Attribution

Credit is assigned based on the actual contribution of each touchpoint, calculated using machine learning across your conversion data. Google Ads and GA4 both offer versions of this. In theory it is the most accurate model. In practice, it requires substantial conversion volumes to produce reliable outputs, and the underlying logic is opaque. You cannot inspect the weighting the way you can with a rules-based model.

I have seen data-driven attribution produce genuinely useful insights on accounts with high conversion volumes. I have also seen it produce nonsense on accounts with thin data, and nobody noticed because the model looked authoritative. The Forrester piece on improving marketing measurement makes a useful point here: the sophistication of your model should match the sophistication of your data, not exceed it.

How Does Attribution Work in GA4?

GA4 defaults to data-driven attribution for conversion reporting, which sounds like progress. For many accounts, it probably is. But it is worth understanding what you are looking at before you trust it.

GA4 uses data-driven attribution for conversion events in the Advertising section of the platform. In other reports, it uses last-click by default. This means you can be looking at two different pictures of the same performance depending on which report you are in, and if nobody has flagged that discrepancy, it will cause confusion.

The Moz overview of GA4 features worth knowing covers some of the attribution settings that are easy to miss during initial setup. It is worth checking your configuration rather than assuming the defaults are right for your business.

One thing GA4 does well is the model comparison tool, which lets you run multiple attribution models against the same conversion data and compare how credit shifts across channels. This is genuinely useful. Running last-click alongside data-driven and position-based will quickly show you which channels are being systematically undervalued or overvalued under your current default. That comparison is often more instructive than any single model on its own.

Where Standard Attribution Models Break Down

All of the models above share a fundamental limitation: they only credit touchpoints they can see. If a customer saw a billboard, listened to a podcast ad, or had a conversation with a colleague before converting, none of that shows up in your attribution data. You are measuring the visible fraction of the experience and calling it the whole thing.

When I was running iProspect and managing significant paid media budgets across a wide range of clients, one of the consistent challenges was that attribution models would show paid search as the dominant conversion driver across almost every account. That was partly true. Paid search is often the last click before purchase. But it was not the complete picture, and making budget decisions based on that data alone would have been a mistake.

The channels that were actually building demand, brand search volume, organic traffic, display, were invisible in last-click reporting. The businesses that understood this distinction made better long-term decisions. The ones that chased last-click efficiency often found themselves in a position where short-term ROAS looked strong but overall growth had stalled.

There is also the cross-device problem. A customer might research on mobile, compare on desktop, and convert on a work laptop. Standard attribution models handle this poorly. GA4’s cross-device reporting has improved with Google Signals, but it still depends on users being logged into Google accounts, which is not universal.

And then there is the offline conversion problem. For businesses where a significant portion of revenue comes through phone calls, in-store visits, or sales team follow-up, any purely digital attribution model is measuring a partial picture by definition. Unbounce’s piece on simplifying marketing analytics makes the point that the goal is honest approximation, not false precision, which is exactly the right framing.

What Is Marketing Mix Modelling and When Does It Matter?

Marketing mix modelling (MMM) is a statistical approach that uses historical data to estimate the contribution of different marketing channels to business outcomes, typically revenue or sales. Unlike digital attribution models, MMM can incorporate offline channels, seasonality, pricing, and external factors like economic conditions.

MMM has been around for decades. It fell out of fashion when digital tracking made channel-level attribution feel more precise. It is coming back into focus now, partly because of privacy changes that have degraded digital tracking, and partly because more businesses are recognising that digital attribution models alone do not give them the full picture.

The honest trade-off is this: MMM is better at capturing the full picture across channels, including offline and brand activity. Digital attribution models are better at granular, near-real-time optimisation. Most mature businesses benefit from both, used for different purposes. MMM for strategic budget allocation, digital attribution for in-platform optimisation.

The barrier to MMM has traditionally been cost and complexity. That is changing. Lightweight MMM tools have become more accessible, and Google’s Meridian open-source framework has lowered the entry point for businesses that want to explore it without a large consultancy engagement.

The Internal Politics of Attribution

This is the part that attribution articles usually skip, but it matters as much as the technical choices.

When you change attribution models, you change which channels look good. That affects team budgets, agency contracts, and in some organisations, bonus structures. I have been in rooms where a shift from last-click to data-driven attribution was resisted not because anyone had a technical objection, but because the paid search team knew their numbers would look worse and the display team knew theirs would look better.

This is not cynicism. It is a real dynamic in most marketing organisations, and ignoring it means your attribution project will stall at the point of implementation regardless of how technically sound the model is.

The way to handle it is to involve the relevant stakeholders early, run the model comparison openly rather than presenting a conclusion, and frame the change as improving collective understanding rather than reassigning blame. Attribution is a better measurement tool, not a verdict on past decisions. That framing matters.

Forrester’s thinking on what to consider when building automated marketing dashboards touches on the governance side of this: who owns the data, who can change the model settings, and how changes get communicated. These are not exciting questions but they are the ones that determine whether your attribution setup actually gets used.

How to Choose the Right Attribution Model for Your Business

There is no universal answer, but there are useful filters.

If your sales cycle is short and your customer experience is simple, time-decay or last-click with a clear-eyed understanding of its limitations will probably serve you adequately. If you are running a paid search campaign for a straightforward product with a same-session conversion rate, last-click is not going to mislead you much.

If your sales cycle is long, your customer experience involves multiple channels over weeks or months, or you have significant offline conversion activity, you need a more sophisticated approach. Position-based attribution is a reasonable starting point. Data-driven attribution is worth testing if your conversion volumes support it.

If you are spending meaningfully across brand and performance channels, and you want to understand the interaction between them, MMM is worth the investment. Not as a replacement for digital attribution, but as a complement to it.

Whatever model you choose, run it in parallel with at least one other model for a period before making budget decisions based on it. The comparison is where the insight lives. The Semrush breakdown of content marketing metrics illustrates this well in the context of content: the same activity looks very different depending on which metrics and attribution windows you apply to it.

One practical recommendation I give consistently: define your attribution model before you run a campaign, not after. I have seen too many post-campaign reviews where the attribution model was chosen to support a conclusion rather than test a hypothesis. That is not measurement, it is storytelling.

What About View-Through Attribution?

View-through attribution credits a channel for a conversion if the user saw an ad but did not click it. Display and video channels use this heavily because their click-through rates are low but their influence on downstream behaviour can be real.

The problem is that view-through attribution is extremely easy to game and extremely difficult to validate. A user who was served an impression and later converted may have converted anyway. Without a proper holdout test, you cannot distinguish genuine influence from coincidence.

I am not saying view-through attribution is worthless. I am saying it should be treated with more scepticism than most ad platforms encourage. If a display partner is claiming significant view-through conversions, the right question is: what would conversion rates look like in a matched audience that was not served the impression? That test is inconvenient for the platform to run. That is exactly why you should ask for it.

For video specifically, Wistia’s thinking on video and webinar metrics is useful context for thinking about what engagement signals actually mean versus what they are often assumed to mean.

Incrementality Testing: The Honest Alternative

If you want to know whether a channel is actually driving conversions, the most honest method is an incrementality test. You split your audience into a group that sees your marketing and a matched control group that does not, then measure the difference in conversion rates between the two groups. The difference is the incremental contribution of that channel.

This approach bypasses the attribution model entirely. It does not try to assign credit across touchpoints. It asks a simpler and more useful question: would these conversions have happened without this activity?

Incrementality testing is harder to set up than running an attribution report, and it requires holding back some budget from a group that might have converted. That cost is real. But for high-spend channels where the attribution picture is genuinely unclear, it is often worth it. The output is a defensible number rather than a model-dependent estimate.

Google, Meta, and several independent measurement vendors offer incrementality testing frameworks. The methodology matters: the control group needs to be properly matched, the test needs to run long enough to be statistically meaningful, and the results need to be interpreted with the right context around seasonality and external factors.

Attribution models and incrementality testing are not competing approaches. They answer different questions. Attribution tells you how credit is distributed across the experience you can observe. Incrementality tells you whether the marketing activity itself is generating outcomes that would not have occurred otherwise. Both are useful. Neither is sufficient on its own.

If you are building out a measurement practice that goes beyond individual models, the broader marketing analytics resources at The Marketing Juice cover how attribution fits alongside forecasting, budget allocation, and reporting frameworks.

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