A sale that appears as a Google Ads conversion may have started with a Meta video, continued through an email, and closed after a branded search two weeks later. Ecommerce attribution modelling is the process of deciding how much credit each of those touchpoints deserves. Get it wrong and you can cut the campaigns creating future demand while increasing spend on the channels merely collecting it.
For ecommerce operators, this is not an analytics exercise for its own sake. Attribution affects budget allocation, reported ROAS, new customer acquisition and ultimately net margin. The aim is not to find one perfect number. It is to make better decisions with imperfect but useful evidence.
Why ecommerce attribution modelling matters
Most advertising platforms want to demonstrate value. Google Ads, Meta and TikTok each use their own tracking methods, lookback windows and conversion logic. As a result, the same £10,000 of online revenue can be claimed more than once across different platforms.
That does not necessarily mean the data is wrong. It means the data answers different questions. A platform report may tell you which users it influenced according to its own rules. Your ecommerce platform may show the final recorded source of a transaction. Neither view alone explains the full customer journey.
The commercial risk is clear. If a retailer relies only on last-click reporting, branded search and direct traffic often look exceptionally strong. Prospecting campaigns on paid social, YouTube, Demand Gen or display can look weak because they introduce customers earlier in the journey. Switch those campaigns off, however, and branded search volume may fall a few weeks later.
The opposite problem also occurs. Taking platform-reported revenue at face value can make every channel look profitable at once. Total claimed revenue then exceeds actual store revenue, and the business increases budgets on the basis of duplicated credit.
A sensible attribution approach helps answer four practical questions:
- Which channels are generating incremental sales, rather than claiming sales that would have happened anyway?
- Which campaigns attract genuinely new customers?
- Where is spend producing profitable revenue after product margin, returns and fulfilment costs?
- What should receive the next pound of budget, and what should be reduced or tested?
The main attribution models and their trade-offs
There is no universal best model. The appropriate choice depends on your sales cycle, average order value, repeat purchase rate, media mix and data quality.
Last click
Last click gives all credit to the final non-direct interaction before purchase. It is simple and easy to explain, which is why it remains common in ecommerce reporting. It can be useful for assessing campaigns intended to capture existing demand, such as high-intent non-brand search or shopping activity.
Its weakness is that it undervalues activity higher up the funnel. A customer may first discover a product through a paid social advert, compare options over several days, then return through a brand search. Last click gives the credit to search and none to the campaign that created awareness.
First click
First click does the reverse, assigning full credit to the first known interaction. It can reveal which channels introduce customers to the brand, particularly where awareness is a genuine constraint.
However, first click can overstate the value of early-stage activity. Generating an initial visit is not the same as moving someone towards a transaction. It also says little about which campaigns are effective at closing considered purchases.
Linear and position-based models
Linear attribution shares credit equally across each interaction. Position-based models give more weight to the first and final touchpoints, with the remainder shared between the middle interactions. Both offer a more balanced view than single-touch models.
The limitation is that the credit weighting is still an assumption. Why should every interaction receive equal value? Why should a first click receive 40 per cent rather than 25 per cent? These models can be helpful for directional reporting, but they should not be treated as proof of causation.
Data-driven attribution
Data-driven attribution uses observed conversion paths to estimate the contribution of different touchpoints. In Google Ads and GA4, it can provide a more sophisticated view than fixed rules, especially for accounts with meaningful conversion volume and a varied channel mix.
It is often a sensible operational setting for bidding because it uses more signals than last click. Yet it remains modelled data inside an ecosystem with incomplete visibility. Privacy controls, consent choices, cross-device behaviour, ad blockers and offline influences all create gaps. Data-driven attribution is useful evidence, not an independent audit of performance.
Build reporting around business reality
The best starting point is not an attribution model. It is a clear definition of what a profitable order looks like.
For many retailers, headline ROAS is too blunt. A campaign selling discounted, low-margin products may report strong revenue while contributing little profit. A campaign acquiring a high-value customer who returns three times may look expensive on the first order but be commercially attractive over time.
Your measurement framework should therefore separate the figures needed for daily optimisation from the figures needed for management decisions. Media teams may optimise using conversion value, CPA and campaign-level ROAS. Directors should also see blended revenue, blended MER, new customer revenue, contribution margin and customer acquisition cost.
Blended MER, or marketing efficiency ratio, compares total revenue with total marketing spend. It will not diagnose every campaign, but it is a valuable sense check. If individual channels report excellent ROAS while blended efficiency deteriorates, attribution inflation, rising discounting or falling conversion quality may be the cause.
It is also worth reporting paid and organic branded search separately. Brand campaigns can be valuable for protecting visibility and capturing high-intent demand, particularly in competitive categories. But treating all branded revenue as incremental paid media growth will exaggerate performance.
A practical approach to ecommerce attribution modelling
Start by making sure the foundation is sound. Purchase events should pass accurate order values, currencies and transaction IDs. Refunds and cancellations need to be considered where they materially affect revenue. Consent mode, enhanced conversions and server-side tracking can improve measurement, but no technical setup removes all uncertainty.
Next, establish one primary source of truth for total sales. For most ecommerce businesses, that is the store platform or finance system, not a collection of ad-platform dashboards. Reconcile paid-media revenue against total store revenue regularly. The figures will not match perfectly, but large or growing gaps deserve investigation.
Then use attribution at three levels. At platform level, use the platform data to optimise campaigns and identify audience, creative, product and search-term patterns. At analytics level, compare channel paths, assisted conversions and new versus returning customer behaviour. At business level, judge whether total marketing investment is producing profitable growth.
A monthly decision meeting should bring those levels together. Avoid changing budgets because of one reporting view or a short period of volatility. Look for patterns across at least several weeks, adjusting for seasonality, promotions, stock availability and changes to pricing.
Test incrementality where the budget justifies it
Attribution models estimate credit. Incrementality testing asks a harder question: did the advertising cause additional sales?
The cleanest approach is a controlled test. You might hold out a region, audience segment or matched group from a campaign and compare performance against an exposed group. Geographic tests can work well for businesses with sufficient volume and relatively consistent demand. Audience holdouts can be useful on paid social, although platform options and practical constraints vary.
Not every retailer needs a complex experiment every month. Tests require enough data, a stable trading period and the willingness to accept a temporary performance trade-off. But they are particularly valuable before scaling spend substantially, reducing upper-funnel activity or making strong claims about brand campaign efficiency.
There are also lower-cost diagnostic tests. Pause a small portion of branded search coverage for a carefully controlled period, where commercially safe. Compare new customer rates by channel. Review what happens to direct, organic and brand-search sales when prospecting investment changes. None is perfect alone, but together they build a more credible picture.
Common mistakes that waste budget
The most damaging mistake is choosing a favourite attribution view because it makes results look best. A channel manager may prefer platform reporting; a search team may prefer last click. The business needs a reporting framework that challenges both.
Another is using different conversion definitions across campaigns. If Google Ads optimises towards purchase value while paid social is judged on landing-page views or add-to-baskets, budget decisions become distorted. Campaign objectives should reflect the role each channel is expected to play, while senior reporting returns to comparable commercial outcomes.
Finally, avoid treating attribution as a one-off analytics project. Tracking settings change, product ranges change, consent rates fluctuate and platform models evolve. Ecommerce measurement needs ongoing quality control, not an annual spreadsheet exercise.
Maple Forest Marketing approaches attribution as part of accountable PPC management: pairing platform optimisation with clear commercial reporting, so reported success stands up against actual revenue and profitability.
The most useful attribution model is the one that makes your next budget decision less speculative. If the reporting cannot tell you whether an extra £1,000 should go into prospecting, shopping, search or conversion rate optimisation, it is time to simplify the dashboard, validate the data and test the assumption behind the number.



