Lift vs Attribution
AnalyticsAlso: Incrementality vs Attribution · Causal vs Modelled Credit
Quick definition
Lift and attribution answer different questions about marketing effectiveness. Attribution decides which channels get credit for conversions that happened. Lift (also called incrementality) tests whether those conversions would have happened without the marketing at all. Most businesses run attribution. Very few run lift tests.
A customer touches all four, then converts.
Credit distributed
Last-click gives all credit to direct. This is where most lift tests reveal the biggest gap: direct often has low incrementality because those visitors were coming anyway.
How it varies across Australia
Lift testing is uncommon outside large Australian advertisers. Most mid-market businesses rely entirely on attribution, which means they are measuring credit distribution rather than causal impact. The gap between attributed conversions and truly incremental conversions tends to be significant, though the size varies sharply by channel and category.
See data and tracking maturity across Australian industries →The two questions
Divides credit for conversions that happened across the touchpoints in the journey.
Answers: who gets credit?Compares a treated group to a control group to measure whether marketing caused the conversion.
Answers: did this cause it?What it actually means
Imagine you run a campaign and sales go up. Attribution says 'here is how we divide the credit among the channels that touched those buyers.' Lift testing asks a different question: 'would those buyers have bought anyway, even without the campaign?'
They sound related because they are. Both try to understand what marketing is doing. But they use completely different methods and answer completely different questions.
Attribution is a credit-allocation model. It takes the conversions that happened and distributes them across the touchpoints in the journey using a set of rules (last-click, linear, time-decay, data-driven). The problem is that the conversion already happened. Attribution doesn't know whether the marketing caused it.
Lift testing uses a control group. You show the campaign to a randomly selected group and withhold it from another. The difference in conversion rate between the two groups is the incremental lift. That difference is as close as marketing gets to a causal answer.
This distinction matters enormously for budget decisions. If your retargeting campaign is showing strong attributed conversions but zero incremental lift, it means you are paying to show ads to people who would have purchased anyway. The attribution model looks good. The spend is waste.
Channels like direct, branded search and email look disproportionately good in attribution models because they sit late in the journey. Lift tests often show those channels have lower incrementality than their attribution share suggests, while upper-funnel channels like display or video are contributing more causally than any attribution model gives them credit for.
Attribution tells you who got the credit. Lift tells you whether there was anything to credit in the first place.
How it shows up
Attribution shows up in your ad platform dashboards, in GA4 channel reports, and in any CRM field that records lead source. It is running all the time, whether you set it up deliberately or not.
Lift testing has to be set up deliberately. The most common formats are geo holdouts (run a campaign in some Australian states but not others, compare results), platform-native lift studies (Meta, Google, TikTok each offer versions), and ghost-ad studies (serve a fake PSA to the control group at the same bid as the real ad).
The output of a lift test is an incremental conversion number and a cost-per-incremental-conversion. Compare that to the attributed CPA from the same campaign and you have a direct measure of how much your attribution model is overstating the channel.
The Australian context
Australia's smaller media market creates specific complications for lift testing. Geo holdouts are harder to run cleanly when your target audience is concentrated in Sydney and Melbourne and there is no clean comparable region to use as control. Platform-native lift studies from Meta and Google require minimum spend thresholds that many mid-market Australian advertisers don't hit in a single campaign.
The practical workaround for most Australian businesses is time-based holdouts (pause a channel for a defined period and measure the revenue impact) or ghost-ad studies run through platforms that support them. Neither is as clean as a randomised geo holdout but both are meaningfully more informative than attribution alone.
Where people get this wrong
Related terms
Common questions
Do I need lift testing if I already have attribution set up?
Attribution and lift answer different questions, so yes. Attribution tells you how credit was distributed across channels. Lift tells you whether the spend actually caused incremental conversions. Running attribution without any lift testing means you're optimising credit allocation without knowing which channels are genuinely driving growth.
How do you run a lift test on a small Australian budget?
Time-based holdouts are the most practical option. Pause a channel for four weeks and measure what happens to conversions. Geo holdouts work if you have clean regional splits. Platform-native lift studies from Meta or Google require minimum spend thresholds, so they suit larger campaigns. Imperfect tests still beat no test.
Which channels usually show the biggest gap between attributed and incremental performance?
Retargeting, branded search and email typically show strong attributed performance because they sit late in the journey. Lift tests often find lower incrementality there than the attribution numbers suggest, and higher incrementality for upper-funnel channels like display and video that attribution models under-credit.
Can attribution models approximate incrementality?
Data-driven attribution gets closer than rule-based models but still can't answer the causal question directly. It weights channels based on observed conversion patterns, not on a controlled experiment. Treat data-driven attribution as a better credit-allocation tool, not a substitute for genuine lift testing.
Keep exploring
About New Rebellion
New Rebellion is a marketing intelligence consultancy. We build tools, score Australian businesses on how their marketing actually performs, and publish Debrief every day. This dictionary is part of how we work in the open.
How we think →