General Terms

Incrementality

Measurement methodology that isolates and quantifies the true causal impact of marketing activities by comparing against a baseline.

Definition

Incrementality is a scientific measurement approach that determines the true incremental value generated by a marketing activity by comparing outcomes against a statistically valid control group. Unlike basic attribution models, incrementality testing uses randomized controlled trials and sophisticated causal inference techniques to identify what would have happened without the marketing intervention, enabling marketers to understand the real marginal impact of their spending and optimize toward truly incremental growth.

Examples

Measuring incremental revenue from a campaign by comparing exposed vs unexposed audiences

Testing true incremental impact of retargeting by withholding ads from a control group

Calculating incremental ROAS by comparing conversion rates between test and control

Determining incremental brand lift through randomized holdout experiments

Calculation

How to Calculate

Subtract the control group's baseline conversions (scaled to treatment-group size) from the treatment group's conversions to get incremental conversions. Divide by control to express as a percentage lift. Incremental ROAS (iROAS) is the same shape but divides incremental revenue by treatment spend: iROAS = (Treatment Revenue − Control Revenue) / Treatment Spend. The control group must be selected via true randomization (not 'similar' demographics) for the math to hold.

Formula

Incremental Lift % = ((Treatment Conversions − Control Conversions) / Control Conversions) × 100

Unit of Measurement

Operation Type

composite

Formula Variables

Treatment ConversionsConversions in the audience exposed to the ad

Control ConversionsConversions in the matched holdout audience, NOT exposed to the ad

Lift Confidence IntervalRange around the lift estimate at the chosen confidence level (typically 90% or 95%)

Industry Benchmarks for Incrementality

Typical performance ranges by industry segment. Benchmarks vary by platform, audience maturity, and attribution window — treat these as starting points, not targets.

Meta — DTC prospecting (cold audiences)
Typical range
55% – 80% incrementality (platform ROAS overstates by 20–45%)
Median
~65%
Cold prospecting is the cleanest test — least overlap with organic demand, highest measured incrementality.
Meta — retargeting (warm audiences)
Typical range
15% – 35% incrementality (platform ROAS overstates by 65–85%)
Median
~25%
Warm audiences already intended to buy — most retargeting 'conversions' would have happened without the ad. The most over-credited channel.
Branded search (Google)
Typical range
10% – 20% incrementality (platform ROAS overstates by 80–90%)
Median
~15%
The canonical 'channel that takes credit for organic demand' — most branded-search conversions happen organically when ads are paused.
Non-brand Google Search
Typical range
40% – 70% incrementality
Median
~55%
Generic queries cleanly measurable — the conversions are largely demand the brand wouldn't have otherwise captured.
TikTok — Spark Ads
Typical range
50% – 75% incrementality
Median
~62%
TikTok measurement is less mature than Meta but holdout tests show comparable incrementality on cold audiences.
Display / programmatic
Typical range
20% – 45% incrementality
Median
~30%
Wide variance — high-quality contextual display reaches 45%; remarketing display often below 20%.
Email / SMS to existing CRM
Typical range
25% – 50% incrementality
Median
~35%
Subscribers already have purchase intent — incrementality is real but smaller than the attribution model claims.
Minimum sample for significant lift detection (DTC)
Typical range
~5,000 – 15,000 users per arm (treatment + control)
Median
~8,000
Below this floor, incrementality tests can't distinguish a 20% lift from noise. Most accounts under $50K/mo on Meta can't run a properly-powered geo test.

Segment	Typical Range	Median	Notes
Meta — DTC prospecting (cold audiences)	55% – 80% incrementality (platform ROAS overstates by 20–45%)	~65%	Cold prospecting is the cleanest test — least overlap with organic demand, highest measured incrementality.
Meta — retargeting (warm audiences)	15% – 35% incrementality (platform ROAS overstates by 65–85%)	~25%	Warm audiences already intended to buy — most retargeting 'conversions' would have happened without the ad. The most over-credited channel.
Branded search (Google)	10% – 20% incrementality (platform ROAS overstates by 80–90%)	~15%	The canonical 'channel that takes credit for organic demand' — most branded-search conversions happen organically when ads are paused.
Non-brand Google Search	40% – 70% incrementality	~55%	Generic queries cleanly measurable — the conversions are largely demand the brand wouldn't have otherwise captured.
TikTok — Spark Ads	50% – 75% incrementality	~62%	TikTok measurement is less mature than Meta but holdout tests show comparable incrementality on cold audiences.
Display / programmatic	20% – 45% incrementality	~30%	Wide variance — high-quality contextual display reaches 45%; remarketing display often below 20%.
Email / SMS to existing CRM	25% – 50% incrementality	~35%	Subscribers already have purchase intent — incrementality is real but smaller than the attribution model claims.
Minimum sample for significant lift detection (DTC)	~5,000 – 15,000 users per arm (treatment + control)	~8,000	Below this floor, incrementality tests can't distinguish a 20% lift from noise. Most accounts under $50K/mo on Meta can't run a properly-powered geo test.

Sources: Meta Conversion Lift studies 2024–2025, Common Thread Collective DTC Incrementality Report 2024, Meta CL benchmark library 2024, Northbeam DTC Index 2025, eBay 2014 branded-search holdout (Tadelis et al.), Google iROAS 2023, Google Causal Inference Lift docs, Northbeam 2025, TikTok Marketing Mix Studies 2024, Nielsen Display Lift Studies 2023, The Trade Desk benchmark library, Klaviyo Benchmarks 2024, Triple Whale CRM Lift 2025, Meta CL methodology, ChannelMix sample-size guidance 2024

Best Practices

✓Use proper randomization for test and control group selection — never 'comparable demographics'
✓Ensure sufficient sample sizes for statistical significance — power analysis BEFORE the test
✓Control for external factors and seasonality by overlapping test/control windows
✓Test across different audience segments and channels to detect interaction effects
✓Document all test parameters and conditions — including the pre-registered MDE
✓Hold the test for at least one full purchase cycle (7 days DTC, 28+ days B2B)

How AdSights helps you track Incrementality

Incrementality testing exposes the gap between what platforms claim and what actually drove conversions — but the test itself doesn't tell you WHY the lift was what it was. AdSights closes that loop by tagging every creative variant, audience, and placement during the test window, then surfacing which creative-level patterns drove the incremental conversions vs which just absorbed organic demand. Teams use this to (1) identify which specific hooks and formats generate real incremental lift on cold audiences vs which only win on retargeting (where lift is structurally low), (2) brief the next creative round against the patterns proven incremental in their own account rather than chasing platform-reported ROAS, and (3) defend creative quality investments to finance by showing the audited gap between reported and incremental ROAS that the account would carry without ongoing creative work. AdSights doesn't run the holdout test itself — that lives in Meta CL, Northbeam, or your warehouse — but it materially improves the diagnostic interpretation of any lift result.

Want AI to track Incrementality across your creative automatically?

Request early access

Supplemental Resources

📚
A/B Test Significance Calculator
Calculate statistical significance for incrementality tests
AdSights Tool
📚
Incrementality Calculator
Measure the true impact of your marketing campaigns by calculating incrementality
AdSights Tool

Frequently asked questions

Common questions about Incrementality, answered.

What's a good incrementality score?

It depends entirely on the channel and audience. For cold prospecting on Meta or TikTok, 50–80% incrementality is healthy — meaning at least half the conversions the platform reports were truly caused by the ad. For retargeting on Meta, 15–35% is typical and not a problem — those audiences would have converted at high rates anyway, so the channel is correctly described as 'finishing' demand rather than creating it. The honest target isn't a single threshold; it's that you actually run the test and budget against incremental ROAS rather than platform-reported ROAS. Most DTC brands discover their incrementality-adjusted ROAS is 30–50% lower than reported across the portfolio.

How is incrementality different from attribution?

Attribution allocates credit for a conversion that already happened — first-touch, last-touch, multi-touch, data-driven. It answers 'who should get the credit?' Incrementality asks the prior question: 'would this conversion have happened anyway?' If a user was going to buy regardless of seeing an ad, EVERY attribution model is wrong by construction — the conversion has zero true incremental value, but every model will still hand credit to whichever touchpoint matches its rule. Attribution is a credit-allocation engine; incrementality is a reality check. Sophisticated teams use both: attribution for in-channel optimization (which ad set scaled?) and incrementality for budget decisions (which channel keeps its budget next quarter?).

How do I run an incrementality test on Meta?

Three options, in increasing order of rigor. (1) Meta Conversion Lift studies — Meta runs the holdout for you, randomizing exposure within your target audience. Free for accounts above a spend threshold (~$30K/month). Easiest to run, but you're trusting Meta's measurement. (2) Geo-holdout tests — pause ads in matched market regions, compare conversion rates. More rigorous but requires geo-tagged conversion data and 6+ matched market pairs. (3) Synthetic control / time-series methods (Causal Impact, Geo-Lift) — model the counterfactual statistically. Most rigorous, most complex. For most DTC accounts, run a Meta CL study quarterly per major audience, and budget against the lift it shows.

Why is platform ROAS always higher than incremental ROAS?

Two main reasons. First, click-through and view-through attribution windows credit the platform with conversions that overlap with organic demand — someone Googles your brand, sees the ad, clicks, buys; the platform takes credit, but the purchase would have happened anyway from the Google search. Second, the same conversion gets credited by multiple platforms simultaneously — Meta, Google, TikTok, and Klaviyo each take 100% credit, so the sum of platform ROAS often overstates true performance by 30–50%. Incrementality testing strips both effects. The gap between reported and incremental ROAS is the 'attribution tax' that every paid-social budget pays.

How long should an incrementality test run?

At minimum, one full purchase cycle plus a one-week buffer — typically 14 days for DTC, 28+ days for B2B with longer consideration windows. The test needs to be long enough that the treatment group's conversions reflect the steady-state effect, not just the launch-week novelty. The sample-size requirement matters more than duration: ~5,000–15,000 users per arm at typical DTC conversion rates. Tests run too short produce wide confidence intervals — you might see a 'lift' between -10% and +60%, which is statistically inconclusive even if directionally positive. Run the test once with adequate power rather than three times with insufficient power.

Does branded search really have low incrementality?

Yes — this is one of the most replicated findings in performance-marketing research. The eBay 2014 holdout (Tadelis et al.) found brand-keyword search ads had effectively zero incremental impact for established brands: users who searched the brand name were going to buy whether the paid ad ran or the organic listing took the click. Subsequent studies at Microsoft, Google, and dozens of DTC brands replicated similar findings — typically 10–20% incrementality, sometimes negative when the paid ad cannibalized organic. The honest takeaway isn't 'kill branded search' — it's 'price the budget against ~15% incremental rather than 800% reported.' Branded search remains worth running for defensive reasons (competitor bids on your name) but at materially lower spend than reported ROAS would suggest.

Can I trust Meta's Conversion Lift results?

Mostly yes, with caveats. Meta's CL studies use proper randomization and the math is sound. Two caveats: (1) Meta's pixel reporting bias — Meta sees more of its own attributed conversions than holdout-group's actual conversions, which can slightly inflate measured lift; counteract by pairing with server-side or warehouse-level conversion data. (2) Selection effects — Meta's CL studies typically don't include audiences who would have been excluded by typical optimization, so the lift estimate applies to the targeted audience, not the universe. Use Meta CL as your primary lift number, validate periodically with a warehouse-based geo-holdout, and budget against the lower of the two.

Featured in topic hubs

Explore this term in context — alongside the related metrics, calculators, and guides curated in these hubs.

Attribution & Measurement Experimentation & Statistics