Pretty ROAS, flat growth: the advertising metrics trap in 2026
Summarize this article with AI
ROAS, the reigning metric for fifteen years
Return On Ad Spend became the north star of performance marketing in Google Ads’ early years. The formula fits on one line: attributed revenue divided by spend. A ROAS of 5 means five euros of revenue for every euro invested. Simple. Clear. Directly negotiable with a CFO.
For fifteen years, this metric shaped budgets, bonuses, steering committees. It let marketing teams shift from a spending posture to financial logic. Platforms — Google, Meta, Amazon Ads — built their algorithms around this signal. Set a ROAS target, and the machine optimizes bids to hit it.
The problem emerged gradually. CFOs started asking an awkward question: if ROAS has climbed for three years, why isn’t total organic growth keeping pace? Why is revenue growing slower than media budget? The Looker dashboards show green curves on every channel. The consolidated line at the top of the P&L stays flat.
The answer comes down to one word: ROAS measures attribution, not causality. It attributes revenue to ads, never answering the one question that matters to a CEO: would this revenue have arrived without the ad?
This nuance, long theoretical, becomes critical in 2026. Fewer clicks happen. AI traffic reshuffles purchase journeys. Platforms consolidate their attribution power. ROAS stays useful. But confusing ROAS with growth is like confusing your speedometer with actual progress down the road.
The three structural biases that inflate ROAS
When your ROAS shows 5 and growth stalls, it’s not a bug. It’s three structural biases, baked into the very mechanics of the calculation.
Bias 1: attribution is not incrementality
ROAS attributes revenue to a campaign using a model — last-click, data-driven, multi-touch. These models allocate credit to every touchpoint. They never ask whether the touchpoint was necessary. A user types your brand name in Google, sees your paid ad, clicks, buys. Full credit to the campaign. No one asks what they would have done without the ad. Most of the time: they would have clicked the organic result right below. And bought anyway.
Bias 2: last-click overweights late-stage acquisition
E-commerce purchase journeys rarely happen in one session. A buyer discovers a brand via a YouTube video. Returns three days later via a blog article. Hesitates for another week. Then types the exact name in Google and clicks the ad. Last-click credits search brand with the full revenue. YouTube and the blog — the true engines of discovery — get zero. Automatic dashboard conclusion: shift more budget to brand search. Growth dies silently, because the top of the funnel starves.
Bias 3: halo effect stays invisible
A display campaign seen by 200,000 people generates maybe 50 clicks. From those 50, 8 conversions. Disastrous direct-view ROAS. But among the 199,950 who didn’t click, a fraction will remember the brand and buy later through another channel. This halo effect — the awareness lift that converts outside attributed journeys — never appears on a standard dashboard. A « bad » ROAS campaign can be the most profitable incrementally. The reverse is also true: a brilliant ROAS campaign can add nothing.
The cumulative effect of these three biases produces the dominant paradox of 2026: green dashboards, red growth.
Brand search cannibalization: the case study
The most documented gap between ROAS and incrementality concerns brand search campaigns — ads triggered when someone types your brand name. They systematically show the best ROAS in a Google Ads account. Often 10, 15, sometimes 20.
The reason is mechanical: a user typing « hi-commerce » already knows the brand. They intend to buy. Cost per click is low, conversion nearly guaranteed. The platform attributes the sale to the brand campaign. Dashboard celebrates.
The test eBay ran years ago, cited in the April 2026 Search Engine Land reference article, shattered this illusion. The company paused its brand ads for a test group and measured what happened. Result: organic captured nearly all the lost conversions. The net impact on revenue was minimal. The displayed ROAS was real. Incrementality was near zero.
This case isn’t isolated. It repeats across most e-commerce advertisers who push brand budget beyond strict defense. As long as a competitor doesn’t bid on your name, every euro spent on brand buys a sale already locked in.
Brand defense stays legitimate: if Amazon or a direct competitor bids on your name, cutting the ad costs real money. But the right measure isn’t « what ROAS am I getting » — it’s « what share of brand revenue truly disappears if I pause ». Most brands have never asked.
A strong brand search ROAS proves nothing. It proves your loyal customers know how to type your name. Congratulations, they could already do that before your budget.
The 2026 metrics: MER, iROAS, <a href= »https://www.hi-commerce.fr/glossaire/#cac » class= »hc-gloss-link » title= »Définition : CAC »>CAC</a> payback
Breaking free from the ROAS trap doesn’t mean discarding it. It remains useful for marginal optimization, campaign by campaign. The shift: stop steering overall strategy with ROAS alone. Add three metrics that capture what ROAS ignores.
MER — Marketing Efficiency Ratio
MER is the simplest and most honest metric: total revenue divided by total marketing spend, all channels combined. No attribution debate. If your revenue is €2M over a quarter and total marketing spend is €400K, your MER is 5. MER automatically captures halo effect, cannibalization, multi-touch journeys — because it ignores attribution. It just watches money in and money out. Comparing MER quarter by quarter immediately shows if your marketing is getting more efficient, without getting fooled by platform dashboards.
iROAS — Incremental ROAS
iROAS is ROAS after an incrementality test. It answers the only question that matters: for every euro spent, how much incremental revenue — that wouldn’t have been generated otherwise — did I get? The calculation happens via a test: two comparable groups, one exposed to the campaign, one not. The revenue difference divided by the budget gives iROAS. An iROAS of 3 means the campaign is profitable. An iROAS of 0.5 means you’re paying for revenue that would have arrived for free. On brand campaigns, typical iROAS falls between 0.1 and 1.5 depending on sector.
CAC payback period
Customer Acquisition Cost payback measures the months needed for the gross margin a new customer generates to repay their acquisition cost. This metric forces retention into the equation. An €80 CAC is excellent if the customer buys four times a year, catastrophic if it’s a one-time purchase. In 2026, with rising CPCs and tougher competition, CAC payback often matters more than ROAS itself. Healthy e-commerce target: 6 to 12 months.
These three metrics don’t replace ROAS. They frame it. ROAS says « this campaign converts ». MER says « is the whole operation getting more efficient ». iROAS says « does this campaign create real revenue ». CAC payback says « is this revenue justified over time ».
| Metric | What it measures | Cadence |
|---|---|---|
| ROAS | Platform attribution | Daily, per campaign |
| MER | Overall efficiency | Monthly, consolidated |
| iROAS | Causal incrementality | Quarterly, per test |
| CAC payback | Customer profitability | Monthly, per cohort |
The incrementality test protocol, once per quarter
Saying « let’s measure incrementality » without a protocol is talk with no action. Here’s the playbook that fits on one page. The one that separates a team that steers from a team that stares at dashboards.
Step 1 — Choose the channel to test
One test per quarter. One channel at a time. Start with the biggest budget and most suspect ROAS — usually brand search or retargeting. Both accumulate all the attribution biases. Chance of discovering overestimation: 70% on the first try.
Step 2 — Define the split method
Three methods, ranked by rigor and complexity:
- Geo split: pause ads in half your country (even-numbered departments, for example), keep running in the other half. Compare revenue movement over 4 weeks. Simplest method, needs geographic bias correction.
- Holdout audience: exclude 10% of your Meta or Google audience from campaigns for 4 weeks. Compare revenue by segment. Cleanest method, natively available via « conversion lift tests ».
- Temporal on/off: alternate weeks with and without the campaign. Roughest method, sensitive to seasonality, useful as backup.
Step 3 — Duration and volume
A useful incrementality test runs at least 4 weeks. Below that, statistical noise dominates. Sufficient conversion volume: target 500 conversions minimum per group for acceptable statistical significance. On smaller accounts, aggregate multiple weeks or test a higher-volume channel.
Step 4 — Calculate iROAS
Formula: (exposed group revenue − control group revenue) / spend on exposed group. Results often surprise. Typical documented cases from Lars Maat in Search Engine Land:
- Brand search: iROAS between 0.2 and 2, vs displayed ROAS of 10-20
- Classic retargeting: iROAS between 0.5 and 3, vs displayed ROAS of 5-10
- Paid social prospection: iROAS between 1 and 4, vs displayed ROAS of 1-3 (here, attribution often underestimates)
Step 5 — Reallocate the budget
The test’s value isn’t confirming what you thought you knew. It’s reallocation. If brand campaign hits iROAS of 0.5, cut the budget gradually — not brutally, to stop competitors moving in. Redirect to channels with iROAS above 2. This quarterly reallocation alone often unlocks 10-20% growth without raising total budget.
Correctly attribute AI traffic and generative search
New dimension in 2026: AI traffic. Users ask questions to ChatGPT, Perplexity, Gemini, Claude. Answers include brands, products, sometimes links. Clicks land on e-commerce sites with empty referers, poorly categorized sources, attribution defaulting to « direct » or « organic ».
This fog directly impacts ROAS. A user discovers your brand through a Perplexity answer, clicks, leaves, returns tomorrow via a brand ad, buys. In most attribution setups, the brand campaign takes 100% credit. AI traffic — which triggered discovery — gets zero.
Result: brand search ROAS keeps inflating in 2026. Not because the campaign works better. Because it mechanically captures work done by LLMs upstream. Growth you think comes from Google Ads actually comes from invisible AI exposure.
How to correct:
- Segment direct traffic. In GA4, create an « AI likely » segment based on visitors with no referer, mobile user-agent, session duration over 2 minutes, first visit. This heuristic captures part of LLM traffic, acceptable margin of error.
- Add qualitative questions to checkout. A « Where did you discover us? » field with an AI option gives reliable declared data at 20-30% response rate. Cross-reference with digital attribution.
- Integrate GEO into the mix. Optimizing brand visibility in LLM responses (via Schema.org, sameAs, llms.txt, Wikidata) becomes an upstream acquisition lever. ROI isn’t measured in direct ROAS, but in lift to brand search and qualified direct traffic.
The methodological issue is simple: a 2026 dashboard with no « AI » line lies about growth sources. A media budget that ignores GEO double-pays for campaigns that capture traffic created free by LLMs.
For deeper coverage of this topic, see our detailed article on converting AI traffic in 2026.
The conclusion that must reach the boardroom
ROAS says if a campaign converts. Not if the company grows. Confusing the two for five years is enough to miss a generation of buyers, let competitors claim the incremental share, and discover too late that media budget funds its own fiction.
Teams winning in 2026 did three simple things:
- They steer by MER for the board, ROAS for daily operations, iROAS for budget decisions, and CAC payback for customer strategy.
- They run one incrementality test per quarter, testing a different channel each time, and build exploitable history after a year.
- They integrate AI traffic into their attribution model, because brand campaign ROAS increasingly becomes a delayed echo of LLM visibility.
This path needs no new tools. Google Ads and Meta natively include conversion lift tests. GA4 segments traffic. A monthly Excel spreadsheet tracks MER. The investment is intellectual, not technological.
The question you gain by asking your marketing team this week: when was our last incrementality test, which channel, what result? If the answer is « never, » you’ve just identified your biggest growth opportunity for 2026.
An attribution and incrementality audit in 30 minutes
Want to know how much of your media budget finances revenue that would have arrived anyway? In 30 minutes, I audit your Google Ads, GA4, and attribution structure. You leave with a clear map of channels to test first, the exact protocol for your first incrementality test, and a realistic budget reallocation estimate for next quarter.
Book a strategic call — 45 minFrequently Asked Questions
Is ROAS still useful in 2026?
Yes, for daily operational steering of a campaign. It optimizes bids and arbitrates between keywords or audiences within a channel. What changes: you don’t steer overall strategy on ROAS alone. MER, iROAS, and CAC payback frame it to avoid biased decisions.
What’s the difference between ROAS and MER?
ROAS divides attributed revenue from one campaign by its spend — with all the attribution limits that come with it. MER (Marketing Efficiency Ratio) divides total company revenue by total marketing spend, all channels combined. No attribution debate, no double-counting. MER is the metric that goes to the board.
How do I run an incrementality test without breaking my Google Ads account?
Use native features: conversion lift tests in Google Ads, Brand Lift and Conversion Lift in Meta. These tools automatically create a control group that doesn’t see your ads, measure the conversion difference, and calculate causal lift. No manual pause needed. Minimum: 4 weeks, minimum 500 conversions per group.
Why is my brand search ROAS so high?
Because it captures customers who already know your brand and intend to buy. CPC is low, conversion strong. But most of these sales would arrive without the campaign — the user would click the organic result right below. Brand search iROAS typically falls between 0.2 and 2, vs displayed ROAS of 10-20. Test, gradually cut if iROAS is weak, keep a defensive baseline if a competitor bids on your name.
How do I include AI traffic in my attribution?
Three levers. One: create a GA4 heuristic segment « AI likely » (direct traffic with no referer, long session duration, first visit). Two: add a declarative field « Where did you discover us? » with an AI option at checkout, for 20-30% reliable response rate. Three: optimize LLM visibility via GEO (Schema.org, sameAs, llms.txt, Wikidata) and track lift on brand search and qualified direct traffic.
