No attribution model tells you what caused a sale. Every model is an accounting rule for splitting credit between touchpoints, and the rule you pick decides which channel looks like the hero in your report.
If you run an e-commerce brand, that matters because budget follows the report. Pick last-click and search gets the credit. Pick first-click and prospecting gets it. Same orders, same journeys, different winners. This guide covers the models, what each one rewards, how multi-touch and data-driven attribution differ, and how to choose one without fooling yourself.
Key Takeaways
- A model is a credit rule, not proof of cause. IAB Australia’s playbook says attribution alone may not show which conversions were incremental, and recommends checking it with experiments.
- Last-click rewards whoever closes the sale. It cannot explain what created the demand in the first place, a point B&T made in 2026.
- Google Ads data-driven attribution compares converting and non-converting paths. Google recommends at least 200 conversions and 2,000 ad interactions in 30 days for better performance.
- Run two or three models side by side and read the gaps. A channel that looks great in one view and weak in another is where your next test belongs.
- Pair attribution with an incrementality test or a marketing mix model before moving a large share of budget.
Table of Contents
- What a marketing attribution model actually does
- The main attribution models, side by side
- One journey, four different answers
- Multi-touch attribution: fuller picture, softer ground
- Data-driven attribution: what it needs and what it learns
- What attribution cannot tell you
- How to choose a model for your business
- Where attribution goes wrong in practice
- Putting attribution to work with one team
- Our view on attribution
- FAQ
- Sources
What a marketing attribution model actually does
Marketing attribution assigns credit for a conversion to the marketing interactions that came before it. IAB Australia’s Digital Ad Effectiveness Playbook (December 2017) defines digital attribution as connecting the advertising interactions that precede an outcome, and describes last interaction as a common standard.
That is the whole job. A model takes a path (say, a TikTok ad, then a Google search, then an email) and divides one sale across it. It does not add sales. It does not remove sales. It moves credit around.
Why does this trip people up? Because a dashboard number looks like a fact. It is a view. Two models can describe the same 500 orders and disagree about every channel. Treat each as a lens, and ask which lens answers the question you actually have.
The main attribution models, side by side
There are five rule-based models you will meet everywhere, plus the algorithmic one. Shopify Australia’s 2026 multi-touch guide describes them as follows: first-touch gives all credit to the first interaction, last-touch gives it to the final one, linear shares it evenly, time-decay weights later interactions more, and position-based (U-shaped) favours the first and last touches. Data-driven attribution estimates credit from observed conversion paths instead.
| Model | How it splits credit | What it rewards | Where it misleads |
|---|---|---|---|
| First-touch | 100% to the first interaction | Prospecting, awareness | Ignores everything that closed the sale |
| Last-touch | 100% to the final interaction | Brand search, retargeting, email | Credits whoever harvests demand, not whoever made it |
| Linear | Equal split across touchpoints | Long, multi-channel journeys | A banner impression counts the same as a product demo |
| Time-decay | More credit the closer to purchase | Short sales cycles, promos | Under-credits the early ad that started the journey |
| Position-based (U-shaped) | Heavy on first and last, light in the middle | Brands that value both intro and close | The weights are a guess, not a measurement |
| Data-driven | Credit estimated from path data | Whatever the data associates with conversion | Needs volume, and only sees what is tracked |
(Fixed weights ≠ measured influence.)
IAB Australia also notes the lookback window matters. A last-interaction approach may look back 30 days for clicks and, if none are found, 14 days for impressions, giving credit to the most recent eligible interaction. Change the window and you change the winners, with no change in customer behaviour.
One journey, four different answers
Take a customer who clicks a Google ad, later clicks a Facebook ad, then types your URL and buys. Shopify Australia’s example of this exact path (Google, Facebook, Direct) shows how far the answers swing:
- Last non-direct click: 100% to Facebook.
- Last click: 100% to Direct.
- First click: 100% to Google.
- Linear: one-third to each.
That is one order and three different heroes. These are alternative accounting views of the same journey, not separate sales.
Now the trap. The same guide warns that its “any click” model gives 100% credit to every channel that was clicked, so total credited conversions exceed actual orders. It is useful for examining one channel. It is useless for splitting a budget across the mix. If your channel reports add up to more orders than your store processed, this is usually why.
Multi-touch attribution: fuller picture, softer ground
Multi-touch attribution (MTA) collects the interactions between discovery and purchase and spreads credit across them. Shopify Australia says it uses software to estimate each interaction’s influence. The weights come either from fixed rules (linear, U-shaped, time-decay) or from an algorithm.
What you gain over last-click is visibility. Upper-funnel video and social ads that never get the final click at least show up in the report.
What you do not gain is proof. IAB Australia’s playbook is plain on this: rule weights may not reflect actual influence, and attribution alone may not identify incremental conversions. Why? Because the model only sees people who touched your ads and then bought. It cannot see the ones who would have bought anyway. A retargeting campaign aimed at people already in your checkout will look brilliant under almost any model.
So MTA answers “who was in the path?” It does not answer “who changed the outcome?” Keep those two questions apart and most attribution arguments go away.
Data-driven attribution: what it needs and what it learns
Data-driven attribution (DDA) is the algorithmic option. In Google Ads, it compares the paths of customers who convert with those who do not, then gives more credit to interactions associated with a higher likelihood of conversion. The weights are learned from your data, not set by a rule someone chose in 2012.
Google’s help page says the model considers clicks and video engagements across Search (including Shopping), YouTube, Display and Demand Gen ads, along with eligible website, store-visit and Google Analytics conversions. Note the scope: it sees Google’s inventory. A Meta or TikTok touchpoint is outside that view unless it is imported some other way.
On volume, the IAB Australia playbook reports Google’s guidance: at least 200 conversions and 2,000 ad interactions in supported networks within 30 days for better performance, though it works with less. If you sell 40 orders a month, a data-driven model is learning from a thin sample. Read it as directional.
Caveat: DDA is still attribution. It is better at describing paths than a fixed split, but it inherits the same blind spot. It rewards what correlates with conversion, and retargeting correlates very well. If you are changing bids on the back of these columns, our guide on Google Ads attribution columns shows what to check first.
What attribution cannot tell you
Three questions get lumped together, and one tool cannot answer all of them. B&T’s 2026 piece on attribution separates campaign-optimisation attribution from experiments that test whether advertising caused uplift, and from brand tracking that measures changes in perception. It also makes the point that last-click can show where existing demand was captured but not what created it.
That gives you a simple split:
- Which ads and audiences should I tweak this week? Use platform attribution and path reports.
- Did this spend cause extra sales? Use an experiment, such as a geo holdout or a conversion lift test. We cover the Meta version in our piece on incrementality testing for Meta ads.
- How should I split the whole budget across channels over a year? Use marketing mix modelling (MMM), which uses historical data to assess how marketing and outside factors affect total sales over time, as Shopify Australia describes. Our note on when media mix modelling pays off covers when it is worth the cost.
MTA works at the level of individual journeys. MMM works at the level of aggregate sales. They disagree sometimes. When they do, that is information, not a bug.
How to choose a model for your business
Shopify Australia’s advice boils down to three inputs: the business question, the complexity of the sales cycle, and the data you have. Here is how that plays out for a DTC brand.
Short cycle, one or two channels, under a few hundred orders a month. Last-click plus a second view (first-click or linear) is enough. A complicated model on thin data produces confident noise.
Consideration of days or weeks, paid social feeding search. Run position-based or time-decay next to last-click. Look at the gap. If Meta looks twice as valuable under U-shaped as under last-click, that is your cue to test it with a holdout, not to move the budget on the spot.
High volume on Google. Data-driven attribution is a sensible default in Google Ads, given Google’s stated data thresholds above. Still compare it with a last-click view for a month before you trust bid changes built on it.
Heavy brand spend or offline channels. Path-based models struggle here. Lean on MMM and lift tests.
A practical routine:
- Write down the decision the report must support (cut a channel, raise a budget, change a creative).
- Pull the same period under last-click, one multi-touch rule and, where available, data-driven.
- Mark the channels whose ranking flips between models.
- Test those channels with a holdout or lift study before moving money.
- Recheck when your mix, promo calendar or tracking changes.
Fixing step 2 is half the battle, and it depends on your data. Tracking gaps from consent choices and browser limits mean every model sees only part of the path. A solid first-party data strategy closes some of that gap. On privacy, the OAIC’s analytics guidance (published 21 March 2018) says analytics using personal information can fall under the Privacy Act, while many analytics activities do not need personal information at all. Have whoever owns privacy at your business confirm which side your setup sits on.
And map the journey before you model it. If you cannot name the usual first touch and the usual last touch for your best customers, no model will name them for you. Our customer journey mapping guide for e-commerce is a good place to start.
Where attribution goes wrong in practice
Four habits cost brands money. We name the habits, not the people.
Reporting from one platform’s dashboard. Each platform credits itself, and each uses its own window. Add them up and you get more orders than you shipped. Reconcile against your store’s orders first.
Cutting top-of-funnel because last-click says it does nothing. Last-click gives it nothing by design. The model has no slot for work that happened earlier. Brands that cut prospecting often see branded search shrink a few weeks later, because the demand it harvested is gone.
Swapping models until the report looks good. If you change the rule every time a channel disappoints, you are not measuring. You are editing.
Treating creative as a footnote. Attribution tells you which channel got credit. It does not tell you which ad did the persuading. Within a channel, creative is usually the lever you can pull fastest, and platforms group near-identical ads together, so ten versions of one script behave like one ad. We cover that in our guide to Meta ad creatives that convert.
Directional note: nobody has a universal benchmark for “how much credit prospecting deserves”. Anyone quoting a single figure for all brands made it up. Your own holdout test is the only number that counts.
Putting attribution to work with one team
Measurement breaks when the people reading the numbers are not the people making the ads. The analyst finds that a creative angle pulls new customers, then hands a note to a creative team that never sees the data again.
We run strategy, creative production, paid media and conversion optimisation as one senior team, so the strategists who read the numbers also brief the creators and sign off the edit. That covers Google Ads, Meta and TikTok, YouTube, conversion optimisation and email and SMS. If your reports disagree and you want one view tied to profit, not just platform-reported revenue, talk to us about a growth programme. We aim to deliver at least 3 times your investment in us.
If you want to see how that looks in practice, read the Bonaire case study or the Honeysuckle Swim Co. case study.
Our view on attribution
Most attribution trouble is a hand-off problem. One person builds the report, another buys the media, a third makes the creative, and nobody owns the gap between what the model says and what the customer did.
We treat a model as a question generator. When last-click and a multi-touch view disagree about a channel, that is a test waiting to be run, not a verdict. We want the person who sees the disagreement to be the same person who changes the brief.
That is why we keep strategy, creative, paid media and conversion work in one senior team. The model points at a weak spot. The same team rewrites the ad, adjusts the landing page and reads the result.
Our target is to deliver at least 3 times your investment in us, measured against your bottom line and not only platform revenue. Attribution is how we keep that honest week to week.
Harry Sekhon
FAQ
What are the main types of marketing attribution models?
The common rule-based ones are first-touch, last-touch, linear, time-decay and position-based (U-shaped). Data-driven attribution is the algorithmic alternative, which estimates credit from observed conversion paths. Each is a different way to split credit for the same sale.
Which attribution model is best?
None wins everywhere. Choose by the question you need answered, how long your sales cycle is and how much conversion data you have. Most brands do better comparing two or three models than committing to one.
What is the difference between multi-touch attribution and marketing mix modelling?
Multi-touch attribution splits credit across touchpoints in individual customer journeys. Marketing mix modelling uses historical data to assess how marketing and outside factors move overall sales over time. MTA suits weekly optimisation. MMM suits annual budget splits.
How much data does data-driven attribution need?
Google recommends at least 200 conversions and 2,000 ad interactions in supported networks within 30 days for better performance. It can work with less, but treat results from low-volume accounts as directional.
Does attribution prove my ads caused the sale?
No. It shows which interactions sat in the path. To test whether ads caused extra sales, run an experiment such as a holdout or conversion lift test, as IAB Australia recommends.
Sources
- IAB Australia, The Digital Ad Effectiveness Playbook (December 2017)
- Shopify Australia, Multitouch Attribution: Models and Tips (2026)
- Shopify Australia, Measure multichannel marketing with Campaigns (2023)
- Google Ads Help, About data-driven attribution
- B&T, A Perfect Attribution Model? You’re Dreaming (2026)
- OAIC, Guide to data analytics and the Australian Privacy Principles (2018)






