Guide · Benchmarks
What is a good ROAS? Depends whose margin is asking.
Written by Saoirse Brennan · 3 February 2026 · Refreshed 29 July 2026
Three sourced numbers frame the question: the 2025 ecommerce field average of 2.87, the 2024 median of 2.04, and the much-quoted 4:1 "healthy" bar, which is really top-quartile territory. None of them is your answer. Your answer is a margin calculation, done per location, and this guide walks the math, the traps in the measurement, and the cases where ROAS is the wrong yardstick altogether.
What do the published benchmarks actually say?
| Benchmark | Value | Reading |
|---|---|---|
| Ecommerce average, 2025 | 2.87 | mid-field |
| Ecommerce median, 2024 | 2.04 | the field's halfway line |
| Common "healthy" bar | 4.0 | top-quartile turf, margin-dependent |
The average-median gap is the first honest lesson: the field data skews upward on a minority of high performers, so "average" already means better than most. Anyone promising 8:1 as a baseline is selling something other than arithmetic.
Benchmarks also age fast in this market. The broader 2026 search data moved in advertisers' favor: cost per lead fell year over year for the first time in five years, and 87% of industries posted better conversion rates. A return that looked strong against 2023 conditions may be merely average now, which is one more reason the target belongs to your margin, never to a screenshot of someone else's dashboard.
How does the margin math work?
Break-even ROAS is 1 divided by contribution margin. A 50%-margin business breaks even at 2.0; a 33%-margin one at 3.0; a 25%-margin one at 4.0. That single line explains why the same 2.9 return was a win for the garden-centre chain in this diary (margins above 35% on collect orders) and would be a loss for a thin-margin electronics reseller.
Contribution margin means revenue minus every cost that scales with the order: goods, payment fees, shipping or fulfilment, returns. Most brands hand us a gross margin and forget returns, which in apparel alone can move the break-even half a point. Run the calculation per product family too: the garden chain's starter kits carried a different margin than refill-style consumables, so one blended target would have mispriced both. Set the working target 15 to 20% above break-even so the campaign funds its own management fee and the occasional bad fortnight.
Which measurement traps corrupt the number?
- Brand demand inside the count. People searching your name convert regardless; counting them inflates ROAS with revenue you already owned. Report brand-excluded or report fiction.
- Uncounted conversion paths. Calls and click-and-collect are the classics: the garden chain had roughly 40% of revenue arriving as collect orders that never reached Google Ads. Phone-heavy branches suffer the same fate, which is why call tracking comes before target-setting.
- Platform-attributed revenue. Google grades its own homework with modeled, last-touch-flavored numbers. Sanity-check quarterly against your order system; gaps above roughly 15% mean the target is being negotiated with a fiction.
- Blended channel reporting. Since November 2025, Performance Max shows a channel performance report splitting Search, Shopping, YouTube, Display, Gmail, and Maps. Use it: a "3.5 ROAS" PMax that is 80% brand-search is a different animal than one prospecting on Shopping tiles.
Why per-location targets beat one number
Branches differ in margin mix, capacity, and repeat behavior, so their break-evens differ. One blended target recreates the blending problem: strong branches subsidize weak ones and the report smiles. Across our audits the median best-to-worst branch spread is 2.7× on cost per lead; a single ROAS target papers straight over it. The per-branch ledger from the structure guide is the fix, with each line judged against its own break-even.
Capacity belongs in the target too. A branch running at 90% utilization gains nothing from more leads it cannot serve; its "good ROAS" is whatever maintains full books at minimum spend, which usually means pulling budget back and watching the return climb. The dental group's saturated city practices ran exactly that play while their freed budget moved to suburban practices with empty chairs, and the blended return rose from 2.6 to 4.2 largely on that reallocation.
When is ROAS the wrong metric entirely?
Lead-generation branches, dentists, gyms, trades, sell appointments rather than carts, and their honest metric is cost per booked outcome against lifetime value: the 2026 medians run $72.97 per lead for dentists and $67.36 for health and fitness. Forcing a ROAS frame onto them requires revenue-per-booking data, which is exactly what offline imports provide. A gym trial at €19 means nothing until the join rate and twelve-month membership value stand next to it; with them, the branch ledger can price a trial like a retailer prices a cart. Measure what the branch actually sells; rename the column if needed.
Fair questions
What is a good ROAS in 2026?
The 2025 ecommerce field average was 2.87, the 2024 median just 2.04, and the widely quoted 'healthy' bar of 4:1 is top-quartile territory. Good is relative to your contribution margin: a brand on 50% margins can grow at two-to-one while one on 25% needs almost four just to break even.
How do I calculate my break-even ROAS?
Divide 1 by your contribution margin. At a 33% margin, 1 ÷ 0.33 gives 3.0: every euro of ad spend must return €3 of revenue before the campaign contributes anything. Set your target above that line with room for the costs the margin figure missed, like payment fees and returns.
Should every location have the same ROAS target?
No. Margins, capacity, and repeat rates differ by branch, so break-even differs too. A blended target lets strong branches subsidize weak ones invisibly, which is the failure our per-location ledgers exist to expose. Set per-branch targets from per-branch economics and review them quarterly.
Why did my ROAS drop when tracking improved?
Usually it rose in truth and fell on paper. Removing brand demand from the count, or adding uncounted conversions like calls and click-and-collect, changes the denominator and numerator honestly. The garden-centre diary shows reported ROAS jumping from 1.7 to 2.2 on counting alone; the reverse happens when inflated counting gets cleaned.
Is ROAS better than CPA as a target?
For carts, usually; for bookings, rarely. ROAS assumes order values vary and matter, which fits retail. A dental practice or gym sells appointments whose value arrives later, so cost per booked outcome against lifetime value reads truer. Pick per branch based on what the branch actually sells.