Attribution in e-commerce — how to assess which channel really makes money from customers

Attribution in e-commerce only makes sense when it helps you make a better budget decision, not just when it neatly splits sales between channels. In practice, the same revenue can be attributed to different sources depending on the model, the length of the path and the quality of the data. First you need to establish what you are really looking for: the channel that creates demand, closes the sale, attracts new customers or delivers the highest margin. Only then is it worth assessing SEO, paid campaigns, remarketing or direct. Without this, it is easy to declare as the winner a channel that appears at the end, but does not create real growth.

What business questions should you ask before choosing an attribution model?

Before choosing an attribution model, you need to establish what business problem you want to solve. That determines whether you look at the channel that initiates demand, supports the decision, or closes the transaction. Evaluating new customer acquisition requires a different approach from assessing channels that only bring the user back before purchase. If you do not define the goal, the model will start steering the conclusions instead of organising them.

The most useful questions are those that can be translated into budget and action. In e-commerce, it usually makes sense to start with a few specific issues:

• Which channels generate the first visit before purchase?
• Which channels most often appear as an assist before conversion?
• Which sources deliver new customers, and which returning ones?
• Which actions bring revenue with a sensible margin, not just turnover?
• Which channels shorten the time to purchase, and which lengthen it?
• Which sources support the sale of more expensive or harder-to-decide products?

Such questions change the way reports are interpreted. If the goal is to grow the number of new customers, a channel with a high number of assists may be more valuable than the channel winning the last click. If margin matters, attributed revenue alone is not enough. In that case, you need to separate channels building demand from channels that only capture existing intent.

A good attribution model is the result of a business question, not a substitute for it. Last click can be useful for assessing sales closure, but it will not show the full contribution of channels that introduce the brand. First click may better reveal sources of demand, but it usually overvalues the first contact. That is why it makes sense first to name the role of the channel, and only then check which model reveals that role best.

How does the customer journey affect attribution in e-commerce?

The customer journey affects attribution directly, because it shows how many touchpoints precede a purchase and what function each channel performs. When the journey is short, the share of the last source usually grows. When a customer returns several times, compares offers and switches devices, the importance of assisting channels becomes much greater. That is precisely why the same model can work well for one shop and poorly for another.

In practice, you need to look at the elements of the journey, not just the final conversion. The most important are the first visit, the number of sessions before purchase, days to conversion, intermediate channels and the last contact. It also matters whether the user starts on mobile and buys on desktop. If you do not take this into account, part of the influence of upper-funnel channels may disappear from the report.

A long journey usually means the customer needed several triggers before making a decision. In such a setup, informational SEO, brand discovery campaigns or return visits may play a preparatory role in the purchase. The final channel remains important, but it does not have to be the main source of value. The more interactions there are before the order, the less reliable a simple last-click reading of sales becomes.

A short journey also requires interpretation, because it does not always mean full independence of the last channel. The user may have previously discovered the brand outside measurable traffic, returned via direct, or clicked a brand ad just before purchase. That is why the number of steps alone is not enough. You need to assess whether a given channel really creates demand, or whether it only appears when the decision was already almost made.

Customer journey analysis also helps separate the roles of channels. Some sources more often start the relationship, others support comparison, and others still recover an abandoned user. This matters when assessing SEO, paid campaigns and remarketing, because not every channel should win the sales report at the end of the journey. Sensible attribution begins when you look at the sequence of touchpoints, not just the final stage.

Why is data quality key to correct attribution?

Data quality is key, because an attribution model only works on what has been collected and assigned correctly. If the source of traffic is wrong, all revenue will end up in the wrong channel. Then the problem is not the model, but poor measurement. Even the best attribution model will not fix incorrectly tagged traffic or unconnected sessions.

In practice, the most damage is done by inconsistent campaign tagging and messy source/medium. The same channel can be reported under several names, so its share of sales is artificially split. Cross-domain tracking is equally important, because without it, transitions between domains can overwrite the visit source. The effect is simple: the channel that creates demand disappears, while the share of direct or last click grows.

Order deduplication, user identification and backend data are also very important. If one order is counted twice or a user on different devices looks like two people, the customer journey stops being reliable. To assess profitability, you also need cost import and information about returns and margin, because revenue alone is not enough. Sensible measurement usually combines event analytics, orders, CRM and BI reporting; otherwise attribution easily mistakes turnover for real profit.

Which attribution models are used most often and how do they differ?

The most commonly used models are last click, first click, linear, positional and data-driven, and they differ in how they assign a larger share of the conversion. Last click gives everything to the last channel before purchase, so it shows sales closure well. First click rewards the first contact, which is why it reveals brand discovery sources better. None of them shows the full truth on its own, because each shifts the share of revenue between channels.

The linear model spreads value evenly across all interactions, so it can be useful for longer paths. The position-based model gives greater weight to the first and last touchpoints, and less to the middle assists. This helps when you want to see both demand sources and the channels that finalise the purchase at the same time. In practice, the choice only makes sense if it matches the role of the channels in your store.

For e-commerce, it is important not to treat models like a leaderboard of winners, but as different ways of answering different questions. If you want to assess the channels that close the transaction, last click can be useful. If you care about new customers and brand discovery, first click, assists or a data-driven model will show more value. Sound analysis means comparing several models and checking how they change the assessment of SEO, remarketing, brand search and discovery channels.

What roles do different channels play in the attribution process?

Different channels perform different functions in attribution: some build brand discovery, others help compare the offer, and yet others finalise the purchase. This matters in practice, because a channel that supports the decision will rarely win on last click. If you assess all sources with a single metric, you will start underestimating activities that create demand. In e-commerce, not every good channel should look like a sales channel at the end of the path.

This is easiest to see on the example of organic traffic and paid brand campaigns. Non-brand SEO often appears earlier, when the customer is looking for a solution, category or comparing options. By contrast, brand search, direct or remarketing more often capture a user who already knows the store and is close to purchase. Landing page also changes the role of the channel: a guide usually discovers, while a product page more often closes the deal.

A practical assessment should combine the role of the channel with the appropriate KPI. For discovery sources, what matters is the share of new customers, first visits and assists. For closing channels, what matters are final conversions, shorter time to purchase and capturing existing demand. In SEO, it is worth looking separately at brand and non-brand, and splitting informational traffic from transactional traffic, because otherwise one report mixes different functions.

What are the most common mistakes and myths related to attribution?

The most common mistake is assuming that last click shows the true impact of a channel on sales. This model mainly says who appeared at the end of the path. In stores with a longer decision-making process, it therefore usually overstates remarketing, brand search and direct. At the same time, it understates the channels that build interest earlier and bring in new users.

Just as often, the problem is not the model itself, but the wrong interpretation of reports. The most costly mistakes are:

• adding together reports from ad platforms, even though each attributes the same sale to itself,
• ignoring direct and dark traffic, which makes earlier contacts disappear from the path,
• assessing channels solely by revenue, without returns, margin and acquisition cost,
• too short a lookback window, which cuts off assists in a longer purchasing process.

It is also a myth that one model is enough for all decisions. You will assess a channel acquiring new customers differently from a channel closing the basket. That is why it makes sense to compare several models, and to verify the conclusions against order data, costs and behaviour after reducing the channel. If the report improves the channel’s performance, but this is not visible in margin or the number of new customers, the decision may be wrong.

How do you verify the correctness of decisions based on attribution?

The correctness of attribution-based decisions is verified by comparing models with the real business outcome after changing activity. If, after shifting budget, only the channel’s share in the report increases, but not revenue, margin or the number of new customers, the decision was off. Consistency with one dashboard is not enough, because every model distributes credit differently. That is why you first check whether the conclusion holds across several models and different lookback windows.

Cohort analysis and customer behaviour after purchase are very useful. They make it possible to distinguish a channel that genuinely brings valuable customers from one that only closes people who were already decided. In practice, it is worth comparing how CAC, the share of new customers, margin and time to conversion change after increasing or reducing the share of a given source. If a channel looks good only in the short term, and does not improve customer quality, its impact may be overstated.

The strongest verification comes from tests that limit the impact of reporting assignments alone. A holdout, geo test or time-based test shows what happens to sales when you quieten a channel in part of the traffic, regions or days. In such a test, measure not only final orders, but also new customers, brand search, direct traffic and assists. This is especially important with AI search and discovery channels, which more often build brand interest than capture the last click.

Four control questions work well in day-to-day work:

• Does the channel’s result hold up in more than one attribution model?
• Do margin, CAC and the share of new customers change after the budget change?
• Do order, returns and cost data confirm the conclusion from analytics?
• Does a fall or rise in the channel also show up in other stages of the customer journey?

If the answers are inconsistent, it is not worth cutting or scaling the channel on the basis of one report. The better decision is to go back to data quality, the role of the channel and the business goal, and only then adjust the budget. In attribution, the most expensive mistakes do not come from a lack of a model, but from excessive confidence that the report already knows the truth about the channel’s impact.