Ecommerce Attribution Models: Which One to Use in Australia

Ecommerce Attribution Models: Which One to Use in Australia

Isometric touchpoints converging on a purchase

An attribution model is the rule that decides which ad, email or search gets the credit for a sale in your online store. For most Australian stores spending on Google and Meta, the answer is to start with a simple position-based or time-decay view, move to data-driven once you have enough conversions, and check whatever the platforms report against the sales in your own store backend. The model you pick changes which campaigns look good, so it also changes where your budget goes.

I’m Jesse, a senior paid media specialist at Senior Ad Managers. Below I explain each model in plain terms, show what each one misses, and give you a way to choose.

Senior Ad Managers
Measure Ads Against Real Revenue
We fix conversion tracking and measure Google and Meta campaigns against closed revenue in your CRM, not just platform-reported results.
  • ✓Google Ads campaign management
  • ✓Meta prospecting and retargeting
  • ✓Conversion tracking fixes
  • ✓Campaign measurement against CRM revenue
Meta (Facebook & Instagram) Ads Management

TL;DR

  • Single-touch models (first-touch, last-touch) give 100% of the credit to one interaction. Multi-touch models (linear, time-decay, U-shaped, W-shaped, data-driven) share it across several.
  • Short buying cycles and few channels suit a simple model. Longer journeys and more channels suit data-driven, once you have the volume.
  • Google recommends at least 200 conversions and 2,000 ad interactions in 30 days before relying on data-driven attribution in Google Ads.
  • Google, Meta and TikTok each count conversions in their own window and do not deduplicate against each other, so their totals can add up to more than your store sold.
  • Whichever model you choose, test it against your actual orders and revenue.

Table of Contents

What an attribution model does

A customer rarely buys on the first visit. They might see a Meta ad on Monday, click a Shopping ad on Thursday, open an email on Saturday and buy on Sunday. An attribution model is the rule for splitting credit for that sale between those touches.

Google Analytics describes attribution as assigning credit for key actions to the ads, clicks and other factors along the path to the action, and says a model can be a single rule, a set of rules or a data-driven algorithm.

Two things matter before you pick one.

First, a model only redistributes credit among touches it can see. It does not prove that an ad caused the sale. Someone who was going to buy anyway and searched your brand name on the way will hand that credit to your brand search ad in a last-click model.

Second, the model is a lens. Change the lens and the same sales make different campaigns look like winners. That is why two reports on the same account can disagree and both be “correct” under their own rules.

The main models, and what each one misses

First-touch

First-touch gives 100% of the credit to the first interaction. Nielsen’s guide to multi-touch attribution notes it is often used to measure marketing meant to reach new consumers for the first time.

It suits a store that wants to know which channels introduce new customers. It misses everything that happens after that introduction, so it flatters prospecting campaigns and hides the retargeting and search work that closed the sale.

Last-touch

Last-touch gives 100% to the final click before purchase. It suits short buying cycles where you mostly want to know which channel closes the sale. It is also the easiest to read and the cheapest to run: the American Marketing Association points out that single-touch models are easier and less expensive to implement, though they ignore the other events leading to conversion.

The usual failure is brand search. If most of your last clicks come from people typing your store name, last-touch hands credit to a campaign that is collecting demand created elsewhere.

Linear

Linear splits credit equally across every touch. Nielsen says even weighting is a good starting point for marketers new to multi-touch attribution who do not yet understand the typical customer journey.

Its weakness is that it treats a passing impression the same as the click that led to the order.

Time-decay

Time-decay gives more credit to touches closer to the purchase. Nielsen notes it is often used for marketing with short consideration cycles, such as promotional campaigns. That makes it a sensible fit for a sale weekend or an EOFY push, where the ads closest to the order did the heaviest lifting.

It undervalues the early touches that put your brand in front of people in the first place.

U-shaped (position-based)

In a U-shaped model the first and last touches each get 40% of the credit, with the remaining 20% shared across the middle. It works for stores where discovery and closing both matter and the middle is mostly reminders.

W-shaped

W-shaped gives 30% each to the first interaction, the moment someone becomes a lead, and the final conversion, with the rest shared between other touches. It was built for sales journeys with a clear lead stage, such as a quote request. For a typical checkout-based store, the middle milestone is hard to define, so it is rarely the best fit.

Data-driven

Data-driven attribution uses machine learning to assign credit from real paths. Google’s version evaluates both converting and non-converting paths, looking at factors such as time to conversion, device, number of ad interactions, order of exposure and creative type.

It is the most informative option when you have the data, and the least useful when you do not. It is also a black box: you see the output, not the rule. We cover how it works and when to trust it in our guide to data-driven attribution.

A quick comparison

Model How credit is split Best for What it hides
First-touch 100% to the first touch Finding what brings in new customers Everything that closes the sale
Last-touch 100% to the last touch Short cycles, simple reporting Discovery and nurturing, and it flatters brand search
Linear Equal across all touches A first multi-touch view Which touches mattered most
Time-decay More to recent touches Promotions and short cycles Early awareness work
U-shaped 40% first, 40% last, 20% middle Stores with 3 or more touches Middle-of-journey influence
W-shaped 30% first, lead and last Journeys with a lead stage Weak fit for plain checkout journeys
Data-driven Algorithm from real paths Accounts with enough volume How the credit was decided

How to choose for your store

Shopify Australia suggests starting with the business question rather than the model. Its guidance, which matches how we approach it, runs like this:

  • One or two touches before purchase, short cycle: single-touch is enough.
  • Three or more touches: start with linear to see your channel mix, then move to U-shaped.
  • Long or complex cycles: start with time-decay or W-shaped, then move to data-driven once volume allows.

Here is how that plays out in practice for a typical Australian store.

A DTC brand selling a $60 to $150 product, mostly on Meta. Buyers often decide within days. Start with U-shaped or time-decay so you can see both the ad that introduced the product and the one that closed it. Keep an eye on whether Meta prospecting looks weak simply because retargeting and brand search get the last click.

A store with a larger basket or a considered purchase, such as furniture or equipment. Journeys run for weeks and include several sessions on different devices. Data-driven suits this better, provided your account has the conversions to support it.

A small store with a few dozen orders a month. A complex model has too little data to learn from. Pick one simple rule, stay consistent for a few months, and compare channels on cost per order against your store’s own order count.

When data-driven is worth switching on

Google Ads data-driven attribution is available for all conversion actions regardless of volume. However, Google recommends at least 200 conversions and 2,000 ad interactions in a 30-day period for the model to analyse the data and give the most effective attribution.

That gives you a simple rule of thumb:

  1. Under that volume, the model has little to learn from. Use a rule-based view and be cautious about big budget moves based on it.
  2. Above it, data-driven is worth running, because bidding can use the paths it sees.
  3. Whatever you run, keep a simple second view, such as position-based, so you can spot when the algorithm’s story drifts from your order data.

Data-driven also depends on clean inputs. If your purchase event fires twice, or fires on a thank-you page reload, the model learns from wrong data. That is a tracking job first and a modelling job second. Our notes on GA4 ecommerce tracking and reconciling before you scale ads cover the usual faults.

Attribution windows: why platforms disagree

The model decides how credit is split. The window decides which touches are eligible at all.

A window is the time after an ad interaction during which a conversion still counts. Google Ads explains that a 7-day window excludes conversions occurring more than seven days later, and recommends checking the time-lag report against your real buying cycle.

You will also see settings written as “7-day click, 1-day view”. That means a purchase is credited to the ad if it happens within seven days of someone clicking, or within one day of someone seeing it without clicking. View-through credit is the part that tends to inflate numbers, because a person who scrolled past your ad and bought later through another channel still counts.

The bigger issue is that each platform counts in its own window, by its own rules. Google, Meta and TikTok do not deduplicate against each other, so the same order can be claimed by all three. Add the platform dashboards together and the total can exceed what your store sold.

Attribution windows: why platforms disagree

One of our clients, a real estate agency, had a reported cost per new client of $300 while the true cost was $800, because most of the conversions were people searching the agency’s own name. After we fixed that, the cost came down to $68 within about three months and held for 10 months. The platform report had told a good story. The numbers in the CRM told the real one. Results like that depend on the account and are never guaranteed.

For an online store, the equivalent check is simple:

  1. Export orders and net revenue for a month from your store platform.
  2. Add up the purchases and revenue each ad platform claims for the same month.
  3. Compare. Some gap is normal, because windows and browser tracking differ. A gap that is large or that suddenly jumps usually points to a broken or duplicated purchase event, such as a Meta Pixel and a server-side event both firing for the same order.

When the gap is explained, you can trust the model’s output. When it isn’t, no model will save you.

Perspective

Most Australian stores we look at are not short of attribution models. They are short of clean inputs and a consistent way of reading them.

Our approach is to fix conversion tracking first, so each order is counted once. Then we split brand searches from new demand, because brand clicks flatter last-touch reports and make prospecting look weaker than it is. Only after that do we choose a model, and we pick the simplest one the volume supports. On Google we run exact and phrase match keywords with tight negatives and a weekly search term review, so the clicks feeding the model are the ones you want to pay for.

The test of any model is whether it agrees with the sales in your own system. If platform-reported revenue and your store’s revenue tell different stories, we trust the store, and we work out why the platform is wrong.

Jesse Lees

Choosing a model is the easy part

Once your data is clean, there are a few habits that make attribution useful rather than decorative.

Use one model for budget decisions and keep it steady. Switching models every month makes trends meaningless. Pick a lens, hold it for a quarter, and compare like with like.

Compare channels on cost per order, not just ROAS. Platform ROAS reflects each platform’s own window and model. Cost per order from your store’s data gives you a figure that does not depend on either.

Look at brand and non-brand separately. If your Google account blends the two, the average hides how much of the result came from people who already knew you. We go into this in more depth when we manage Google Ads for eCommerce, including Shopping and Performance Max.

Treat the model as a guide to where to test, not proof of cause. Attribution measures which touches sat on the path to a sale. It does not show what would have happened without the ad. For big budget shifts, a holdout or geographic test is a stronger check than any rule-based model. For wider, top-down views of spend across channels, see our primer on marketing mix modelling for Australian businesses.

Capture click IDs so your own data can close the loop. If you can tie ad clicks to completed orders or sales in your own system, you can check every model against revenue. Our piece on capturing GCLID and fbclid for closed-loop attribution explains how.

Choosing a model is the easy part

What we do for stores that want clearer numbers

We start new clients with a $99 AUD Google Ads audit. It looks at how your conversions are tracked, where spend is going and which search terms are producing orders, so you know whether your current numbers can be trusted before you choose a model.

Senior Ad Managers

From there, our Google Ads management and Meta Ads management are each $1,725 AUD per month per platform, with no setup fee, no percentage of ad spend and a 3 month minimum. Senior specialists run every account, and results are checked against the sales in your own system. If the tracking itself needs work, our website tracking service covers it.

FAQ

What is the best attribution model for an ecommerce store?

There is no single best one. For most small to mid-sized Australian stores with short buying cycles, U-shaped or time-decay is a good start. Move to data-driven when your account has enough conversions, and keep a simple model alongside it as a cross-check.

What does 7-day click, 1-day view attribution mean?

It means a purchase is credited to an ad if it happens within seven days of a click on that ad, or within one day of someone viewing it without clicking. The view part can add conversions that the ad did not drive, so compare it with your store’s order data.

Why do Google and Meta report more sales than my store?

Each platform counts conversions in its own window, using its own rules, and they do not deduplicate against each other. One order can be claimed by several platforms. Duplicate purchase events, for example from a Pixel and a server-side event, can also inflate totals. Your store’s order data is the figure to check against.

Is last-click attribution still worth using?

Yes, for short buying cycles and simple reporting, because it is easy to read. Its weakness is brand search, which collects credit for demand created elsewhere. Split brand from non-brand before you judge any campaign on last-click results.

How much data do I need for data-driven attribution?

Google Ads makes it available on all conversion actions, but recommends at least 200 conversions and 2,000 ad interactions within 30 days for the most effective results. Below that, a rule-based model is easier to trust.

If your reports and your sales do not match, start with the $99 AUD Google Ads audit.

Senior Ad Managers
Discuss Ecommerce Attribution
Email us about your store's attribution and ad reporting, and we can discuss measuring campaigns against actual sales and revenue.

Sources

J
Jesse · Senior Ad Managers
Founder of Senior Ad Managers, a Melbourne performance marketing agency. Manages paid search across AU, NZ & US accounts.

Want a $99 AUD audit of your account? →

Next step

Find out where your ad spend is going.

Get a $99 AUD Google Ads audit. We'll show you where money is being wasted and which leads you're missing out on.

Show me my current ad mistakes →
CallContact Us