A customer sees a Meta ad for a new skincare range on Monday, searches the brand on Wednesday, clicks a Google Shopping ad on Friday, then buys after an email reminder on Sunday. Which channel earned the sale? This attribution modelling example shows why the answer is rarely as simple as the last click.

For Australian businesses spending across paid media, SEO, email and marketplaces, attribution determines where the next dollar goes. Get it wrong and a channel that creates demand can look wasteful, while the channel that closes demand receives too much credit. The outcome is not merely a reporting issue. It is a budget allocation problem built to move revenue.

Key takeaways

• Attribution is a decision framework, not a perfect record of customer behaviour.
• Last-click reporting is useful for operational reporting but routinely undervalues awareness and consideration activity.
• A practical model should be compared against incrementality tests, blended revenue and profit, not treated as the sole truth.
• The best model depends on buying cycle length, channel mix, data quality and the commercial decision being made.

An attribution modelling example for an eCommerce brand

Consider an Australian direct-to-consumer retailer selling premium skincare. It generates $2 million in quarterly online revenue and invests in Google Ads, Meta, creators, organic search and email. Its performance marketing strategy needs to account for the fact that first-time buyers often make several visits before converting.

During one quarter, the retailer records 1,000 first-purchase orders worth $150,000. A simplified path analysis finds the following channel interactions:

Last-click attribution awards $73,500 of revenue to Google, $39,000 to email, $21,000 to organic search and only $16,500 to Meta. On that view, Meta appears to deliver a poor return: $16,500 in attributed revenue from $24,000 in spend.

A marketing manager acting on that report may cut Meta and move the budget to Search. That looks rational until the customer paths are examined. Meta appeared early in more than half of all converting journeys and was particularly prominent among new customers. Google Search frequently captured people who were already aware of the retailer, including branded queries that became more common after Meta activity.

The channel did not necessarily create every one of those purchases. But neither did Search create every sale it closed. That distinction is the point of attribution modelling.

How different models change the result

Attribution models apply a rule for assigning conversion credit across touchpoints.

For a broader view of channel contribution, marketing mix modelling uses aggregate business data to assess how marketing activity and other factors affect outcomes.

The rules are simple; the strategic implications are not.

Last click

Last click gives 100 per cent of conversion value to the final measurable interaction. It is easy to explain and can be useful when assessing what directly closed an order. It also favours lower-funnel channels such as branded search, remarketing and email, especially for products with longer consideration periods.

For the skincare retailer, last click says Meta generated $16,500. That conclusion is likely too harsh if Meta is introducing new buyers to the brand.

First click

First click assigns full credit to the first recorded interaction. This better reflects customer acquisition but can overstate the value of prospecting channels and ignore the work required to convert consideration into revenue.

If 400 of the 1,000 orders began with Meta, first click might credit Meta with $60,000. That is a useful signal about discovery, but it does not prove Meta alone caused $60,000 in incremental revenue.

Linear attribution

Linear attribution splits credit equally across all recorded touchpoints. A path of Meta, organic search, Google Shopping and email would allocate 25 per cent of the order value to each channel.

This is fairer than last click when channels genuinely work together, but equal credit is still an assumption. A quick email click and a product comparison session do not always have equal influence.

Data-driven attribution

Data-driven attribution uses observed conversion paths to estimate how the likelihood of conversion changes when a touchpoint is present or absent. In Google Analytics 4, this is the primary reporting model for eligible properties. It is generally more useful than fixed-rule models because it responds to a business’s actual path data.

These differences are also important when building marketing dashboards, where platform, analytics and CRM figures need to be interpreted alongside their different attribution logic.

There are limits. It only sees interactions it can measure, it relies on sufficient conversion data, and it cannot reliably resolve people who move across browsers, devices or walled platforms without consented identity. Treat its output as a sophisticated estimate, not a verdict.

A commercially useful credit allocation

After reviewing data-driven reporting, the retailer applies a practical blended view to the same $150,000 in revenue:

Meta now looks materially different. It is still less efficient than Search on attributed revenue, but it is contributing to customer acquisition and broader demand creation. Email remains highly efficient, though much of its performance depends on building an audience through other channels.

The decision is not automatically to increase Meta spend. A media planning strategy should consider gross margin, repeat purchase rate, stock availability and customer acquisition cost. If Meta-acquired customers have a stronger 90-day repeat rate, a lower first-order return can be commercially sound. If they do not, the budget should be constrained.

A more decision-ready formula is:

Contribution after marketing = revenue x gross margin – media spend – variable fulfilment costs

Return on ad spend is fast and familiar, but it can hide low-margin products, discount dependency and expensive delivery. Attribution should lead towards profitable growth, not a flattering dashboard.

Validate the model with experiments

No attribution model can fully answer the causal question: what would have happened without this marketing activity? That is where experimentation matters.

For Meta, the retailer could run a geographic holdout test. Keep creative, audience and budget consistent in selected comparable regions, then reduce or pause delivery in a test group for several weeks. Compare changes in total sales, new-customer revenue, branded search volume and direct traffic against the control group.

For Search, a business can examine whether non-brand campaigns create incremental sales or simply capture people who would have converted via organic listings. Brand campaigns require particular care. They may protect against competitors and improve conversion rates, but platform-reported revenue can substantially overstate their incremental effect when brand demand is already high.

Experiments are not always clean. Seasonality, promotions, media spillover and small sample sizes can distort results. Even so, a well-designed test is more valuable than false precision from a single dashboard.

Build an attribution operating rhythm

Attribution works best when it is part of regular commercial planning rather than a quarterly analytics project. Start by defining the decision. Are you choosing a prospecting budget, valuing new customers, assessing a promotion or setting channel targets? One model will not answer every question equally well.

Next, standardise campaign naming, UTMs and conversion definitions. A model built on inconsistent source data will produce confident-looking rubbish. Ensure revenue, refunds, new versus returning customers and key events align between your ecommerce platform, analytics setup and advertising platforms.

Then compare three views each month: platform-reported results, analytics attribution and blended business performance. Blended performance includes total revenue, total media spend, gross margin, new-customer volume and customer acquisition cost. When all three tell a similar story, confidence increases. When they diverge sharply, investigate before shifting spend.

Finally, establish a testing calendar. Reserve a portion of spend for controlled tests, especially in channels that influence demand before conversion. Businesses that only optimise to reported last-click ROAS tend to harvest existing demand until growth stalls.

When a simpler model is enough

Not every business needs advanced attribution immediately. If most sales happen within one visit, there are few channels, and the buying cycle is short, last-click analysis combined with blended revenue may be adequate. Complexity has a cost in tracking, implementation and interpretation.

More advanced modelling becomes worthwhile when customer journeys span weeks, spend is meaningful across several channels, online and offline activity interact, or executives are making significant budget reallocations. At that point, measurement is no longer a reporting function. It is part of the growth strategy.

The strongest attribution approach is not the one with the most complicated algorithm. It is the one that helps your team make a better next budget decision, then tests whether that decision created measurable incremental revenue.