A customer sees an ad on Instagram and ignores it. Days later they search on Google, click an ad, but do not buy. A week after that they get an email with a discount and, finally, they buy. Simple question: which of these channels deserves the credit for the sale?
The answer most tools give by default — the last click — is also the most misleading. By crediting everything to the email, we ignore that Instagram planted the seed and Google brought the intent. Deciding the budget from that distorted view leads you to cut precisely the channels that start sales.
Marketing attribution is the set of rules that decides how to split the credit for a conversion across the various touchpoints that preceded it. Choosing those rules well changes how you read each channel's performance — and, as a result, where the money goes. This guide walks through the main models and shows how to pick one without getting lost in theory.
The problem: many hands touch a sale
Almost no purchase happens on the first contact. The typical journey crosses several channels over several days: social media, search, email, price comparison, perhaps a second direct visit. Each of these touches does something — create awareness, remind, convince, push over the line.
The trouble is that the revenue all arrives at once, at the moment of purchase, while the merit is spread across the whole journey. Without an attribution model, the tendency is to credit whatever was closest to the end, simply because it is the easiest to see. It is like giving all the credit for a goal to whoever made the last pass, ignoring everyone who built the move.
The most common attribution models
There is no "true" model; there are different lenses for looking at the same journey. It is worth knowing the main ones before choosing:
- Last-click: all the credit goes to the last channel before the purchase.
- First-click: all the credit goes to the channel that started the journey.
- Linear: the credit is split equally across every touchpoint.
- Time decay: touches closer to the purchase receive more credit.
- Position-based (U-shaped): the first and last touch take the biggest share, the middle splits the rest.
- Data-driven: a statistical model estimates each channel's real contribution from the data.
The choice between them is not cosmetic. The same channel can look like a hero or a waste depending on the model — and the budget follows that narrative.
Last-click and first-click: simple but misleading
Last-click is the default in almost every tool, and the reason is honest: it is simple to calculate and does not require stitching together data from the whole journey. The problem is that it systematically overvalues bottom-of-funnel channels — email, brand search, remarketing — and makes the channels that create demand at the top look useless.
First-click has the symmetric flaw: it gives all the credit to whoever opened the door and ignores everything it took to close the sale. A discovery channel looks brilliant, while the channels that actually converted stay invisible.
Both share the same underlying fault: they assign 100% to a single touch when the reality is shared. They work as a starting point, but running an entire strategy off one of them is almost a guarantee of biased decisions.
Multi-touch models: sharing the credit
Multi-touch models try to correct that bias by spreading the credit across several touchpoints. Linear is the simplest — everyone gets an equal share — and it is already a huge improvement over the single click, because it at least acknowledges that the journey exists.
Time decay assumes that more recent touches weighed more on the decision and gives them more credit, without zeroing out the earlier ones. The U-shaped model bets on two phases: it values whoever brought the customer in and whoever made them convert, giving less weight to the middle. None is perfect, but all capture reality better than a single click.
The choice between them depends on the business. A long buying cycle with a lot of research benefits from models that value the middle of the journey; an impulse purchase may be well explained by time decay. What matters is that the rule is explicit and known to whoever reads the reports.
Data-driven attribution: letting the data decide
Instead of imposing a fixed rule, data-driven attribution uses the data itself to estimate how much each channel actually contributes. It compares journeys that converted with journeys that did not and infers the weight of each touchpoint from the difference it makes to the probability of purchase.
It is the most rigorous model, but it has demands. It needs volume: with few conversions, the estimates become unstable. It needs well-collected journey data, which means tagging campaigns consistently. And it is less transparent — it gives you a number, not a simple rule you can explain in a meeting. For many companies, starting with an explicit multi-touch model and only later evolving to data-driven is the more sensible path.
How to choose and implement a model
In practice, choosing a model can follow a few concrete steps:
- Start from the business question: do you want to know which channels start sales, which channels close them, or each one's overall contribution?
- Get the foundation right: tag every campaign consistently so you can reconstruct the full journey.
- Pick a model that matches your data: without volume, prefer an explicit multi-touch model to an unstable data-driven one.
- Compare models side by side before deciding: see how the credit shifts from channel to channel.
- Document the chosen rule and use the same one in every report, so you are not comparing different things.
- Reassess periodically: the channel mix changes and the model should keep up.
The goal is not to find the perfect model, but to swap a view you know is wrong — the blind last click — for one that better reflects how sales actually happen.
Common mistakes to avoid
Some traps repeat themselves. The first is comparing campaigns measured with different models, which makes the numbers incomparable. The second is ignoring channels that do not generate clicks but do influence — like video or display impressions — that no click-based model captures. The third is treating the attribution number as an absolute truth, when it is always an estimate dependent on choices.
Finally, there is the mistake of changing the model without telling the decision-makers. If the budget moves because the model changed, and not because performance changed, that needs to be clear. Transparency about the method is worth as much as the method itself.
Mini case study: an online shop redistributing budget
An online homeware shop, with spend split across several platforms, ran everything on last-click. Under that reading, social media looked weak and was about to lose budget to brand search, which "converted" much better.
Before cutting, the team compared last-click with a U-shaped model over the same three months of data. The picture changed: social media appeared at the start of a good share of the journeys that ended in a purchase — it was social that brought in new customers who then converted via brand search. Cutting social would have dried up the top of the funnel.
Instead of reducing, the team rebalanced: it kept the investment in discovery and adjusted only the excess in remarketing, which several models showed was overvalued. After a quarter, with a virtually identical total budget, the number of new customers had risen by around 15%. The change was not spending more, it was reading better.
In practice
Attribution is not an academic exercise; it is the lens through which you decide where to invest. Last-click, being the default, misleads more people than it helps: it rewards the end of the journey and punishes whoever starts sales. Any explicit multi-touch model is already a leap forward.
Choose a model aligned with your business question and the volume of data you have, apply it consistently, and treat the result as a good estimate, not a sacred truth. The goal is not perfect attribution, it is to stop deciding budgets with a view you already know is skewed.
