On this page
iROAS (Incremental Return on Ad Spend)
iROAS (Incremental Return on Ad Spend)
iROAS measures the revenue an advertiser would not have received without running ads — it strips out sales that would have occurred organically regardless of ad exposure. It is the causal answer to "what did this spend actually cause?" and is distinct from ROAS, which reports total attributed revenue divided by ad spend and routinely over-credits by including organic conversions.
How it works
Formula:
iROAS = incremental revenue / ad spend
= (test group revenue − control group revenue) / ad spend
Google Ads defines an iROAS of 2 as meaning "for every $1 USD invested, your business generated $2 USD net new in conversion value that otherwise wouldn't have existed" (as-of live Google Ads Help doc, https://support.google.com/google-ads/answer/14102986).
Why traditional ROAS falls short: each walled garden (Amazon, Google, Meta, TikTok) uses different attribution windows. Skai's 2025 worked example: electronics brand — platform ROAS = $4.50; incrementality testing reveals 30% of conversions were organic; therefore iROAS = $3.15 — ROAS over-credits by ~30% (as-of 2025-11-11, Skai vendor source).
Related metrics — not the same thing:
| Metric | Question answered | Use case |
|---|---|---|
| ROAS | Total attributed revenue / spend | Daily monitoring |
| iROAS | Incremental revenue / spend | Campaign validation |
| miROAS | Revenue from the next marginal dollar | Budget optimisation |
| iCMAM | Incremental contribution margin / spend | Profitability check |
Prescient AI further distinguishes three ROAS variants by use case: traditional ROAS for quick daily monitoring, modelled ROAS for ongoing measurement, and iROAS specifically for validating measurement systems at defined points — not an always-on metric [1].
Measurement methods
IAB Europe (2025-03-06) ranks seven methods, from most to least experimental rigour:
| Method | Strength | Challenge |
|---|---|---|
| A/B testing / RCT | Precise point-in-time estimate | Not all platforms allow it; hard to scale |
| Match-market testing | Precise estimate | Risk of contamination from external factors |
| ML counterfactual models | High accuracy and flexibility | Data complexity; trust issues with non-experts |
| Synthetic control | — | Bias and quality of synthetic data |
| Shadow-mode testing | — | Requires additional technical systems |
| Ghost ads | Real-time; reduces ad waste | Very few retail media platforms allow this |
| Marketing Mix Modelling (MMM) | Comprehensive, ongoing, cross-channel | Limited granularity and timeliness |
Source: https://iabeurope.eu/unlocking-the-power-of-retail-media-a-deep-dive-into-sales-incrementality-measurement/, IAB Europe 2025-03-06.
Geo holdout design is the most commonly cited practitioner method. Requirements per Measured.com: at least 6 months of clean historical data, 80% statistical power, holdout ~25% of baseline revenue across matched markets, run for 4–6 weeks [2].
Privacy resilience: geo-based methods split by geography rather than individual user tracking, enabling measurement against first-party offline sales data without cookie dependency [3].
Minimum viable scale: Eightx (2026) reports single-digit adoption of incrementality testing at sub-$5M DTC revenue (holdout costs outweigh decision value), rising to majority adoption above $150M [4]. eMarketer/TransUnion (July 2025): 52% of US marketers use incrementality tests; 36% plan further investment within 12 months (as-of 2025-07, via Skai).
Platform tools (2026)
Google:
- Conversion Lift (in Google Ads): two designs — user-based holdouts (smaller budget, can run alongside Brand Lift) and geo-based holdouts (any data source, no cookie dependency; uses open-source Trimmed Match and Time-Based Regression tools). Source: Think with Google, https://business.google.com/en-all/think/measurement/incrementality-testing/, updated 2026-01-19.
- Meridian GeoX (previewed May 2026): open-source geo-incrementality solution, publisher-agnostic — can measure any channel (Meta, TikTok, podcasts) against first-party backend revenue. Distinct from Conversion Lift, which measures Google Ads within the Google walled garden. Source: Stella, https://www.stellaheystella.com/blog/google-ads-incrementality-test-with-meridian-geox-setup-guide, 2026. (as-of 2026-05, in preview)
- Meridian Studio (pilot July–December 2026): Google Cloud UI for Meridian; integrates incrementality, MMM, and MTA into a unified view via Analytics 360. Source: Google Marketing Live 2026 via Brainlabs, https://www.brainlabsdigital.com/google-marketing-live-2026-brainlabs-review/, 2026-05-20.
- Demand Gen Uplift Experiments (announced GML 2026): automated A/B framework to measure incremental uplift of adding Demand Gen to a campaign mix; Google reports +10% higher ROAS and +12% higher sales effectiveness (as-of 2026-05, Google self-reported — volatile, self-serving).
Meta: Conversion Lift uses randomised controlled trials (RCT) to determine whether ad exposure caused purchases that would not otherwise have occurred. First-party documentation not directly accessible for this run (JavaScript-rendered page). Source: multiple third-party references.
Tinuiti / Bliss Point:
- YouTube Intelligence Suite (launched May 2026): reports 47% average incremental conversion impact for brands (as-of 2026-05-14, Tinuiti proprietary data — volatile, vendor source).
- NOBULL ghost bidding case (streaming TV via NBCUniversal): +19.8% site visits, +5.5% purchases, +3.7% total revenue from adding streaming TV ads [5].
Skai: keyword-level iROAS insights for retail media; Bayer achieved >32% improvement in incremental ROI after integrating Incremental's measurement signals into Skai's platform (as-of date unknown, Skai vendor case study — promotional bias, Skai blog).
Pacvue Incrementality Console (launched Sep 2024): modelled iROAS dashboard running regression models to estimate incrementality without live holdout experiments — presented as scalable alternative for retail media [6].
Benchmarks
All benchmark figures below are volatile. Treat as directional only until independent non-vendor sources can confirm.
| Benchmark | Figure | Source | Date | Caveat |
|---|---|---|---|---|
| Meta median iROAS | $2.30 | Measured.com (274 experiments) | Date unknown | Vendor; conflict of interest |
| Google median iROAS | $2.39 | Measured.com (274 experiments) | Date unknown | Vendor; conflict of interest |
| CTV median iROAS | $2.88 | Measured.com (274 experiments) | Date unknown | Vendor; conflict of interest |
| Sponsored Brands Video vs Sponsored Products | ~2.5× iROAS | WARC Data | Date unknown | Paywalled; extracted from search snippet only |
| IKEA iROAS boost using AI Max | +28% incremental ROAS | Google (GML 2026) | 2026 | Single advertiser; Google-reported |
| HexClad: brand video budget shift | +40% incremental results from 20% budget shift to brand video | Connor Rolain, Marketing Operators (2026-03-17) | 2026 | Single brand; practitioner-reported |
| Google Search (self-reported) | $6 per $1 invested | Think with Google | 2026 | Self-serving; Google first-party claim |
| Google PMax example (beauty brand) | £6 iROAS | Think with Google | 2023 | Illustrative scenario; not third-party validated |
| Google YouTube example (financial) | £1.10 iROAS | Think with Google | 2023 | Illustrative scenario; positive but marginal |
| LiftLab FY-2025 (Apparel/CPG/Footwear/Lifestyle) | iROAS +5–10% YoY | LiftLab FY-2025 Benchmark Report | 2026 (FY-2025 data) | US client dataset; vendor source |
| LiftLab FY-2025 (Home/Tech Services) | iROAS −4–7% YoY | LiftLab FY-2025 Benchmark Report | 2026 (FY-2025 data) | US client dataset; vendor source |
What practitioners report
Marketing Operators podcast (E103, 2026-03-17): HexClad, Ridge, and Jones Road Beauty practitioners describe running consecutive holdout-style incrementality tests as standard operating practice — "string together consecutive channel tests, using each result to build the next hypothesis rather than testing in isolation." TikTok Shop validated as incrementally additive for HexClad (reaching a genuinely new audience, not cannibalising existing channels).
IAB Europe (2025-09-09) forecasts commerce media ad spend will exceed $150 billion across US and Europe, framing incrementality as "one of the most critical metrics for success" in commerce media.
Tinuiti notes that linear TV, podcast, and direct mail cannot be measured via user-level holdouts — geo-lift or MMM is required (Tinuiti, 2026-03-22).
Path to Purchase Institute + Skai joint study (2024): 70% of advertisers struggle to measure the incremental performance of their retail media (as-of 2024, vendor-partnered study).
Industry standards (2026)
IAB Europe and IAB released:
- Guidelines for Incremental Measurement in Commerce Media (November 2025) — available free at https://iabeurope.eu/wp-content/uploads/IAB_and_IAB_Europe_Guidelines_Incremental_Measurement_Commerce_Media_November_2025.pdf
- Commerce Media Measurement Standards V2 (January 2026) covering incrementality, new-to-brand, and sales reporting; V2.1 released May 2026 at https://iabeurope.eu/wp-content/uploads/IAB-Europes-Commerce-Media-Measurement-Standards-V2.1-May-2026.pdf
- Certification transition: V2 became mandatory for certification after end of July 2026 (as-of 2026-01-22).
- IAB EU measurement standards (April 2024): define iROAS as a standard for retail media; define Halo Attribution (IAB Europe 2024-04, via ROAS run).
Contradictions
iROAS as a structural problem vs. a disclosure problem: Ovative / Albertsons / Kellogg School of Management (March 2026) analysed 42 real ad campaigns with identical media, audience, creative, and spend, and found iROAS figures varied by an average of 6.5× (median 2.5×), with results flipping positive to negative in 83% of campaigns depending solely on measurement methodology — framing iROAS as currently unreliable enough to be actively gamed [7]. VS Skai argues the metric is sound but requires transparency standards — disclosure of methodology, control group construction, features used, and known limitations — framing the problem as a retailer accountability gap, not a methodology flaw [8].
Always-on experiments vs. periodic validation: Haus advocates for always-on geo holdout experiments as the primary causal measurement approach, citing 31% iROAS improvement from iterative consecutive testing [9]. VS Prescient AI argues iROAS should be used only as a periodic validation tool (not always-on), with Marketing Mix Modelling (MMM) as the preferred continuous optimisation signal, noting that iROAS consistently undercounts cross-channel halo effects from upper-funnel activity [1]. Both are vendors with competing products; neither is a neutral arbiter.
Geo holdout practicality by brand size: Marketing Operators E103 (2026-03-17) describes HexClad, Ridge, and Jones Road Beauty practitioners running consecutive holdout experiments as standard practice. VS a practitioner YouTube video [10] argues geo holdouts are too slow, costly, and fragile for routine DTC use — regions are rarely comparable, revenue is sacrificed in suppressed geos, and a 2–4 week window is too short for upper-funnel effects. The contradiction may be a scale effect: the 9operators guests are large-scale operators with sufficient volume; the "fails" argument targets smaller brands. Neither source establishes revenue thresholds.
Modelled iROAS vs. experiment-based iROAS: Pacvue Incrementality Console (2024-09-20) presents modelled iROAS (regression-based, no live holdout) as a reliable alternative to experiment-based measurement for retail media advertisers. VS the "How to Actually Measure Marketing Effectiveness" video [11] implies only in-market experimentation with test and control groups establishes true causality. This is the core modelled-vs-experimental debate in the measurement community; neither source was independently audited.
iROAS overstatement range: Measured.com claims platform ROAS overstates true iROAS by 30–60% (1.5×–3× overstatement range) — specifically on branded search and retargeting [2]. No independent academic validation of this specific figure was found. The 30–60% range originates from vendors who sell incrementality measurement, creating an incentive to overstate the overstatement.
Limitations and critiques
- Geographic spillover: Haus identifies this as a structural flaw in geo holdout tests, particularly in dense European cities like London, where holdout residents move into test zones and are exposed to suppressed ads — calling standard postal-code holdout regions a likely source of "extreme spillover effects" [9].
- Holdout opportunity cost: holding out 20–50% of audience for weeks or months creates direct revenue risk (Haus, ibid).
- Cross-channel halo blind spot: iROAS consistently undercounts upper-funnel contribution (CTV awareness → search conversion); MMM is better suited for capturing these long-horizon effects [1].
- Privacy and signal loss: iOS 14.5+, third-party cookie deprecation, and European privacy regulation have complicated user-level tracking, making platform-side conversion lift studies harder and pushing brands toward geo-based methods (Haus, ibid).
- Scale dependency: geo holdout testing requires sufficient baseline revenue to achieve 80% statistical power across matched markets within the test window [2].
Key terms
| Term | Meaning |
|---|---|
| iROAS | Incremental Return on Ad Spend — revenue caused by ads / ad spend |
| miROAS | Marginal iROAS — return from the next additional dollar of spend (used for budget optimisation) |
| iCMAM | Incremental Contribution Margin Allocation Model — profitability-adjusted variant of iROAS |
| Geo holdout | Incrementality experiment that suppresses ads in matched geographic regions |
| Ghost bids | Phantom auction bids placed without ad delivery — enables user-level holdout without sacrificing impression opportunities |
| Conversion Lift | Platform-level RCT tool (Google, Meta, TikTok) measuring incremental conversions from ad exposure |
| Meridian GeoX | Google open-source geo-lift tool, publisher-agnostic (previewed May 2026) |
| Blended ROAS | Total revenue across all channels / total ad spend — monitors efficiency but cannot isolate causal channel contribution |
| Halo Attribution | IAB Europe standard for attributing indirect brand halo effects in retail media measurement |
Next frontier
- miROAS — marginal iROAS; needed for spend optimisation; different metric from iROAS; no dedicated page
- iCMAM — profitability-adjusted incrementality metric; surface in one 2024 video; no dedicated page
- Halo Attribution — IAB Europe retail media standard; newly elevated by this run
- Conversion Lift — platform-level tool; referenced heavily; no dedicated page
- Geo Holdout Testing — practitioner methodology; multiple contradictions surfaced; warrants own page
- Marketing Mix Modelling (MMM) — counterpart measurement approach; existing page status unknown
References
- Prescient AI, date unknown, vendor source — prescientai.com/blog/iroas-limitations
- Measured.com, date unknown, vendor source — www.measured.com/faq/incremental-lift-analysis-practical-guide-iroas-confidence-intervals
- Lifesight, 2026 — lifesight.io/blog/geo-based-incrementality-testing
- Eightx, directional; not from a formal survey — eightx.co/blog/what-is-incrementality-testing
- Tinuiti, 2026-03-22 — tinuiti.com/blog/measurement/incrementality-in-marketing
- Pacvue, 2024-09-20, vendor product demo — www.youtube.com/watch?v=HaC8_TOvPfE
- ovative.com/wp-content/uploads/2026/03/Retail-Media-iROAS-Demystified.pdf
- date unknown, vendor source — skai.io/blog/iroas-core-requirements
- date unknown, vendor source — www.haus.io/article/measuring-iroas-is-it-worth-it
- "Why Geo Holdout Tests Fail", date unknown — www.youtube.com/watch?v=dgy0q5DebFE
- 2026-05-12 — www.youtube.com/watch?v=XhMbeN7cH34