Incrementality testing has gone from a frequently-discussed best practice to actual operational discipline at a growing number of major advertisers over the past 24 months. The pressure of cookie deprecation, the maturity of the available tooling, and the increasing scrutiny of marketing investment efficiency have all combined to push incrementality from theoretical to practical.

Looking at where the discipline actually sits in mid-2026, the picture is one of genuine progress, persistent operational difficulty, and a clearer understanding of what incrementality testing can and can’t tell you.

The high-level story is positive. Major Australian advertisers across retail, financial services, telecommunications, and consumer products now run regular incrementality tests as a matter of course. The methodologies have stabilised. The integration with marketing decision processes has matured. The conversations between marketing and finance about media investment efficiency are better-grounded in data than they were three years ago.

What hasn’t fully resolved is the gap between incrementality measurement at the channel level (which is now reasonably tractable) and incrementality measurement at the campaign or creative level (which remains genuinely difficult).

What’s actually working

Channel-level incrementality testing for major paid media channels is producing usable answers. The major platforms — Meta, Google, TikTok at the larger end — all support some form of conversion lift testing, holdout testing, or geo-experiment frameworks. The methodologies vary, the limitations are real, but the answers they produce are decision-useful.

The sophisticated advertisers are running these tests regularly across their channel mix and using the results to inform budget allocation. The pattern that’s emerging is that the incrementality of individual paid channels is often substantially lower than the platform-attributed conversions would suggest, particularly for prospecting and lower-funnel campaigns. The implications for budget allocation have been significant in many cases.

Geo-experiment testing for offline channels. The methodology of running campaigns in some geographic markets and not others, controlling for baseline differences, has matured into operational practice for retailers and brand advertisers running offline media. The integration with point-of-sale data and store-level reporting produces usable answers about offline media incrementality. The advertisers who’ve invested in this capability have a clearer picture of their full marketing efficiency than the ones who haven’t.

Match-market testing for new-channel decisions. When evaluating whether to invest in a new media channel or significant expansion of an existing one, structured match-market testing produces decision-useful evidence. The discipline of identifying matched markets, running structured tests over adequate periods, and analysing results properly is becoming more common. The platforms supporting this — both first-party platforms and third-party measurement services — have improved.

What remains hard

Cross-channel interaction effects. The interaction between channels — how exposure on one channel affects response to messaging on another — is genuinely difficult to measure and has not been adequately solved by the available tooling. Advertisers running coordinated multi-channel campaigns can demonstrate channel-level incrementality but struggle to demonstrate the coordination benefit beyond what each channel produces in isolation.

Brand and long-term effects. Most incrementality testing has a measurement window of weeks to a few months. The brand-building and long-term effects that justify substantial portions of marketing investment are not adequately captured by these tests. The advertisers who under-invest in brand for several years to maximise measured short-term incrementality may produce decisions that look right in the data but produce business outcomes that disappoint over longer time horizons. The methodological work to capture longer-term effects is real but is not yet operationally simple enough for most advertiser practice.

Creative-level testing. Identifying which creative is producing what incrementality, within an active campaign, requires test designs that are operationally complex and that often conflict with the optimization objectives of the platforms running the campaigns. The major platforms’ machine learning systems are designed to optimize delivery within campaigns based on response, which is operationally helpful but creates measurement challenges for creative-level incrementality testing.

Small advertiser application. The methodology and tooling for incrementality testing have generally been built around the data volumes and operational scale available to large advertisers. Smaller advertisers, who arguably need this kind of measurement more because they have less margin for error, find the tooling and methodology harder to apply. The market has gradually adapted but the gap remains real.

What the better practitioners are doing

A few patterns I observe in advertisers running incrementality testing as actual practice rather than as occasional one-off projects:

They’ve integrated test design into the marketing planning cycle, not as an afterthought. Quarterly planning includes the testing programme: what tests will run, what budget is allocated to them, what decisions the results will inform. The test programme is treated as a first-class part of the work, not as overhead on top of the work.

They’ve built statistical literacy into the marketing team. The interpretation of test results, the recognition of statistical significance versus practical significance, the understanding of what tests can and can’t tell you — all of this requires baseline numerical capability that not all marketing teams have. The teams that have invested in this capability produce better decisions from their tests.

They’ve negotiated their measurement infrastructure to support the discipline. The first-party data infrastructure, the platform-level access, the reporting tooling — all of this needs to support the kind of testing that’s being done. Teams that try to run incrementality tests on top of inadequate measurement infrastructure produce inconclusive results that don’t change decisions.

They communicate test results in business decision terms, not in statistical terms. The right level of communication for executive audiences is “this channel produced $X of incremental revenue per $Y of spend, with confidence interval Z” — not detailed methodological discussions. The teams that translate technical results into business language get better executive support and better follow-on investment decisions.

What I’d avoid

Three patterns that produce bad outcomes:

Treating incrementality test results as universal truths. A test result is valid for the conditions under which the test was run — the audience, the time period, the offer, the competitive context. Generalising too broadly from a single test produces over-confidence in the results and may produce decisions that don’t hold up.

Running tests that don’t have the statistical power to produce useful answers. The marketing science literature is clear about minimum sample sizes for various effect sizes, but the practical pressure to “run a test” sometimes produces tests that are too small or too short to detect the effects that exist. The result is “no significant difference” findings that get interpreted as “no effect” when the actual interpretation should be “test was inconclusive”.

Ignoring the cost of running tests. Holdout audiences forgo conversions. Geo-experiments forgo revenue in test markets. These costs are real and need to be in the test design economics. Tests that produce useful answers but cost more than the answers are worth aren’t a good use of resources.

What I’d watch

Three things over the next two quarters.

The regulatory direction on first-party data and consent frameworks. The Australian Privacy Act reforms continue to evolve, and the implications for measurement infrastructure are real. The advertisers building measurement capability that depends on patterns of consent that may not be sustainable should be considering the medium-term direction.

The maturity of clean-room measurement frameworks. The ability to combine advertiser and platform data without exposing either side’s primary data has been advancing. Whether this produces measurement capability that improves the cross-channel and creative-level testing problems remains to be seen.

The integration of AI tooling into measurement workflow. The use of AI for test design, result interpretation, and ongoing measurement has been advancing. The capability is genuine but the validation discipline around AI-assisted measurement is still being established.

The honest summary: incrementality testing in mid-2026 is real operational practice at sophisticated advertisers, and the discipline is producing better decisions. The gap between the best-in-class and average practice remains wide. The advertisers who are still measuring marketing effectiveness primarily through last-click attribution or platform-reported metrics are operating with a substantially weaker information base than their better-equipped competitors.