You have probably heard that correlation is not causation. Yet a lot of marketing measurement still treats the two as interchangeable. Attribution models show which touchpoints appeared in the customer journey, not which ones actually caused the purchase. Incrementality is the discipline that helps close that gap: it estimates what would not have happened without your marketing, which is the question that matters most when you are defending a budget. It is a core part of a credible marketing measurement approach, alongside attribution and modelling.
Incrementality is the additional outcome your marketing actually caused, over and above what would have happened anyway. Attribution cannot see it, because it credits touchpoints that were present rather than testing what was causal. You estimate incrementality with controlled experiments: randomised tests where you can split people and geo-lift tests where you cannot. The strongest programmes pair experiments with marketing mix modelling, using test results to calibrate the model so its conclusions hold across channels you have not tested. None of this is free, experiments cost reach and take time, so the discipline is testing where the uncertainty is most expensive and letting a calibrated model carry the rest.
Incrementality is the causal lift your marketing creates: the difference between what happened with a campaign and what would have happened without it. Straightforward to state, harder to measure well.
Attribution offers a tidier story, this click led to this conversion, here is the return, but it sets aside a basic question: would that customer have bought anyway? Estimating incrementality means comparing exposed and unexposed groups to gauge the true effect of spend, rather than crediting whichever touchpoint sat nearest the sale.
Attribution assigns credit based on presence, not cause. Last-click gives all the value to the final touchpoint; multi-touch spreads it across the journey. Both assume every interaction mattered. What they tend to miss:
The result is optimising for presence in journeys rather than for created demand, which is part of why click-based attribution holds performance understanding back.
Measuring incrementality means running a controlled comparison: a test group exposed to your marketing and a control group that is not. The difference between them is your estimated incremental lift.
Randomised controlled trials provide the strongest causal evidence. You split your audience at random into test and control, run the campaign to the test group only, then compare outcomes. Key requirements:
The trade-off is real: you are deliberately withholding marketing from potential customers, which is hard to justify against quarterly targets. Academic work on causal inference sets out why a proper counterfactual, not just good data, is what separates cause from correlation.
When you cannot randomise individuals, you can randomise geographies. Geo-lift tests compare markets where you change spend against matched control markets where you do not; validating assumptions with a geo-lift test on branded search is one worked example. The approach suits TV, out-of-home and radio, where individual targeting is not possible.
Experiments in marketing measurement helps validate channel assumptions that attribution cannot address on its own.
Privacy changes have made deterministic attribution less reliable. Platform restrictions, the decline of third-party cookies and regulation mean you can observe less of the journey and with less accuracy. With budgets under pressure (Gartner found 59% of CMOs say theirs is too small to deliver their strategy), being able to show causal impact is what protects spend.
Third-party cookies have not disappeared altogether, Google has kept them in Chrome, but relying on persistent user-level tracking as the backbone of measurement has become less attractive. Incrementality does not need individual tracking: test and control groups can be measured in aggregate, so you read lift at that level rather than stitching together individual conversions. Preparing for measurement in a privacy-first world covers the wider shift.
Where incrementality helps:
Incrementality testing and marketing mix modelling are natural partners. Experiments give you strong causal evidence for specific questions under the condition tested; marketing mix modelling gives you a portfolio view across all marketing. Run geo-lift tests on your major channels, then use the results to calibrate the model, checking its estimates against causal evidence where you have it and correcting the response curves that are off. That calibration raises confidence in how the model is built, which makes its estimates for untested channels more credible, though those remain modelled estimates rather than experimentally proven. The combination helps in three ways: it anchors model estimates to experimental evidence, it needs fewer expensive tests and it produces both channel-level insight and portfolio-level guidance. differ in how well they incorporate experimental results.Marketing mix modelling solutions
A model estimates response curves for each channel from historical spend and performance. Those curves can be wrong, picking up correlation rather than cause or missing a confounding factor. Feeding incrementality results in as priors or validation points corrects for that. If an experiment and the MMM materially disagree, you have found something worth investigating. Look at differences in population, timing, treatment definition, spillovers, model specification and the uncertainty around both estimates before deciding which evidence should carry more weight.
When the model and the experiments agree, you can lean on the model for channels you have not tested, bearing in mind those figures are still modelled estimates rather than directly measured. In our case, combines customisable Bayesian models with a Media Scenario Planner, so calibrated response curves feed straight into budget planning.Objective Platform
Many tests are underpowered. Detecting a realistic lift needs large samples: a 5% lift can require tens of thousands of users per group to estimate reliably. Run power calculations before launching. If you cannot reach around 80% power at your expected effect size, increase the sample, extend the test or accept that smaller effects may not show.
Control groups need to be genuinely unexposed, but spillover works against that. National TV reaches control markets, social ads get shared and branded search captures demand created elsewhere.
Geographic controls tend to work better for broadcast media; user-level controls for targetable digital. Choose the design that matches your contamination risk.
Test length matters. Too short and you miss delayed conversions or long consideration cycles; too long and external factors creep in. Match the test to your sales cycle: a six-month B2B journey needs a longer window than an impulse purchase and seasonal businesses need to account for cyclicality. Predictive media planning helps set the right test windows for your category.
Measuring incrementality only pays off if it changes decisions, so you need a way to turn findings into budget choices.
You cannot test everything at once. Prioritise by spend size and uncertainty: test your largest channels first, then the ones where attribution is most questionable. Two to four major tests a year is usually enough to keep learning without overwhelming the organisation or contaminating results.
Incremental lift tells you what the tested activity added, not the optimal spend. A channel can be clearly incremental and still be poorly optimised. A channel showing 20% incrementality at today’s spend might show much less once spend is doubled because of diminishing returns. Estimating lift at different spend levels helps map the response curve; combining that evidence with MMM lets you project performance across a broader range of spend and make more informed reallocation decisions.
Incrementality often unsettles comfortable narratives. Channels that look strong in attribution can show modest incremental lift. Teams measured on attributed conversions may resist a method that lowers their apparent contribution. Implementing it well tends to need senior sponsorship: someone who values an accurate answer over a flattering one and will protect teams running tests that might return uncomfortable results. It also needs some education, because experimental design is unfamiliar to many marketers. Sharing results clearly and showing how incrementality improves decisions and protects budgets when revenue tightens does more than any single workshop.
Retail media is a hard case: you advertise on the same platform where people buy, so attribution looks strong because it is catching high-intent shoppers, many of whom would have bought anyway. Holdout tests, where some high-value customers do not see the ads, estimate the real difference in purchase rates and basket sizes, which is often well below the attributed figure. Measuring the true incremental impact of retail media goes into this in depth.
Brand building
Brand activity rarely gets last-click credit, but sustained geo-holdouts, where some markets never see the brand campaigns, can estimate its contribution through differences in awareness, consideration and sales over six to twelve months. This is central to balancing brand and performance.
Incrementality at current spend does not predict incrementality at higher or lower spend, so it is worth testing the margins: pulse spend up or down in test markets while holding controls constant. These tests show where diminishing returns set in and where you may be underspending, which is the evidence behind a budget plan that holds up all year.
Both have a place. You do not drop attribution because incrementality is more rigorous; you use each for what it does best.
Attribution suits: real-time optimisation within channels, understanding journey patterns, allocating credit for internal reporting and quick directional feedback on new creative.
Incrementality suits: budget allocation across channels, estimating true marketing contribution, validating attribution assumptions and making the case for marketing to the board.
Use attribution for fast, granular signals within addressable channels, experiments for causal checks on specific questions and MMM for the portfolio view across channels and external business drivers. The strongest measurement setups combine the three rather than asking one methodology to provide every answer. Run experiments quarterly or annually and use attribution for day-to-day optimisation in between.
Incrementality separates marketing that creates demand from marketing that captures it. As attribution becomes less reliable and budgets face more scrutiny, estimating true causal lift matters more, not less. Objective Platform combines customisable Bayesian marketing mix modelling with scenario planning, so incrementality insights translate into budget decisions you can defend across the organisation.
What is incrementality in marketing?
Incrementality is the additional outcome your marketing actually caused, over and above what would have happened without it. It is measured by comparing an exposed test group with an unexposed control group and reading the difference, rather than crediting whichever touchpoint was closest to the sale.
What is the difference between incrementality and attribution?
Attribution assigns credit to the touchpoints present in a journey, so it tends to over-credit lower-funnel activity and cannot see demand that would have converted anyway. Incrementality tests what was causal. Attribution is useful for fast, tactical optimisation; incrementality is the more defensible basis for budget decisions across channels.
How do you measure incrementality?
With a controlled experiment. Randomised controlled trials split individuals into test and control groups; geo-lift tests vary spend across matched markets when you cannot split people. You then compare outcomes between exposed and unexposed groups to estimate the lift, with enough sample size and run time to detect a realistic effect.
What is a holdout test?
A holdout test deliberately withholds marketing from part of your audience or some geographies, so the difference in outcomes against the exposed group estimates what that marketing actually added. It is one of the most practical ways to check whether apparently effective activity is genuinely incremental or mostly capturing existing demand.
How does incrementality work with marketing mix modelling?
They complement each other. Experiments give strong causal evidence for specific channels; marketing mix modelling gives a portfolio view across all of them. Feeding experimental results into the model as priors or validation points calibrates its response curves on the channels you tested, which makes its estimates for the rest more trustworthy, while keeping clear those are modelled rather than directly measured.
