ROAS benchmarks by industry are, with very few exceptions, invented. Not exaggerated, not merely out of date. Invented, in the sense that no dataset sits behind the number.
I went looking for the sources under the tables ranking on page one for this search. What came back was one undisclosed vendor feeding most of them, a widely copied table whose two cited sources contain no benchmark data whatsoever, and a handful of figures that are real but arrive with caveats their publishers never repeat.
This is for anyone who has been sent a link to an industry table and asked why their account sits below it.
Where ROAS benchmarks by industry actually come from
Almost none come from a survey, and almost all come from each other.
Search for ROAS by industry and you get a dozen tidy tables, each with a row per sector and a figure to two decimal places. Cross reference them and the same values keep reappearing, which is the giveaway. Independent measurement produces different answers, while copying produces identical ones.
The chain usually runs like this. A vendor publishes a median with no method attached. Then a content site republishes it with the current year in the title. A second site splits it by sector using nothing but judgement, and a third quotes the second.
By that point the figure has a date, a breakdown and four citations, and still nobody has measured a single account.
The table that cites two pages containing no data
One example is worth walking through, because the pattern repeats everywhere.
A page ranking for industry return on ad spend presents seventeen rows, one per sector, each with a neat figure. The page calls its own numbers directional ranges. It gives no sample size, no data provider, no date range and no attribution setting.
Then look at what it cites. Two sources: Google’s documentation for target ROAS bidding, and Shopify’s page about return on ad spend. Neither document contains a single benchmark figure.
Google’s page explains how a bid strategy works. Shopify’s page says an acceptable return depends on your profit margins, operating costs and advertising goals, then declines to name a number. So the largest authority in ecommerce refuses to answer the question, and a page citing that refusal produces seventeen precise answers.
One undisclosed vendor sits underneath most of the tables
Follow enough tables upstream and you keep arriving at the same company.
Varos publishes cross platform medians, most often quoted as 3.31x on Google and 2.19x on Meta. Those two figures appear on page after page, usually with no attribution and usually presented as this year’s data.
Here is what accompanies them: no sample size, no methodology, no date range and no geography. When I went looking for the underlying benchmark pages, they returned no data at all.
So the most repeated ROAS benchmarks in the category are also the least documented. That is not an argument against measuring things. It is a warning about what happens when one undocumented figure becomes the seed for an entire content category.
The one ROAS benchmarks figure I will quote, and its caveat
There is a single number here I am willing to repeat, and it comes with a warning attached.
Triple Whale reports a platform wide median of 2.04, published as a 2024 figure, with Meta accounting for 68.3% of the ad spend it tracks. The wider report behind that analysis covers $18.4 billion of ad spend across more than 33,000 brands, which is a serious base.
Then read the caveat. That median ROAS of 2.04 appears with no stated sample or date range on the page carrying it. So it is better sourced than anything else in the category and still not something to set a target from.
It is useful for exactly one purpose. If the middle of tens of thousands of tracked ecommerce brands sits near 2.0, a shop on a 25% margin needing 4.0x to break even will not be saved by matching the average.
Two of my own accounts that ROAS benchmarks cannot survive
Put two accounts I run side by side and the whole idea falls apart.
The first is a US flag and banner ecommerce brand on Meta, June to August 2026. Across its six sales ad sets it returned 1.58, or 1.99 with one failed location test removed. Those figures are dated, measured against the shop’s own revenue, and completely unremarkable.
The second is a US automotive equipment retailer whose dashboard reports 44.37. That is not a result, it is a fault. The conversion setup is counting purchases it should not, because no ecommerce business quietly earns forty four times its ad spend without anybody noticing. Its traffic metrics are perfectly sane: an $8.28 CPM, a $0.28 CPC, a 2.92% click through rate and 43,415 link clicks.
Both accounts would supply a row to somebody’s table. One row would be measured and one would be nonsense, and no table distinguishes between them.
Why the numbers could not be comparable even if they were real
Suppose every table were honestly measured. ROAS benchmarks would still not be comparable between two shops.
Attribution alone decides the answer. A seven day click and one day view window reports a bigger figure than a one day click window on exactly the same sales, and almost nobody states which one they used.
Channel mix does the same job. An account whose spend sits mostly in branded search reports a return that a prospecting account cannot get near.
Then there is margin, which never appears in any table, along with returns, discount depth, repeat purchase rate and currency. Two shops in one category can differ on all six at once. An average built across those differences is an average of things that were never the same measurement.
Ninety seconds to test any benchmark page
Four questions kill most of them, and it takes about a minute and a half.
How many accounts, over what dates, in which countries, and on which attribution setting. Any figure that cannot answer all four is a number somebody typed.
Then click the citations. Not to confirm they exist, but to read what they actually say. A surprising share of links point at pages containing no figures at all, or at an entirely different statistic.
Finally, check whether the publisher sells something the number supports. Vendor research is not automatically wrong, and it does need reading with the commercial interest in view. I run this test on every statistic before it reaches a client deck, and most published ROAS benchmarks fail on the first question.
What to use instead of ROAS benchmarks
Three sources beat every table, and all three describe your own business.
Start with your break even point, which is one divided by gross margin. Across my ecommerce clients that margin runs from 20% to 50%, so the floor lands between 5.0x and 2.0x. A good ROAS is whatever clears your own floor with enough room left to pay the overheads.
Then use your own trailing ninety days as the comparison. Last quarter in the same account, on the same tracking setup, is the only genuinely like for like benchmark in existence.
Third, use platform documentation for mechanical thresholds. Google states that Search and Shopping need at least 15 conversions in the past 30 days before target ROAS bidding is available, with Demand Gen needing 50 in 35 days. Those are facts about the system, not claims about performance.
Where benchmark data is still worth having
Input costs are a different case from outcomes, and there the numbers are usable.
Industry benchmarks for auction prices, session costs and conversion rates describe conditions that everybody shares. Your gross margin does not change what a thousand impressions cost to buy.
IRP Commerce, for instance, published UK and EU ecommerce figures for June 2026 showing a 2.03% conversion rate, up 9.74% year on year, alongside a cost per session of £0.16, up 45.82%. They also state their own limitation plainly: these are informed estimates rather than audited values, drawn from last click data across SME and mid market stores.
That is how a benchmark should read. A number, a period, a population and a limitation, all in the same paragraph.
What to change this week
Three steps.
Delete the industry ROAS benchmarks from your reporting deck and replace them with your own break even figure next to last quarter’s actual performance.
Then take the single most quoted statistic in your current strategy document and spend five minutes tracing it. Find the sample, the date and the method, or stop using it.
Then agree one dated internal benchmark with your client, written down, so next quarter has something honest to be measured against.
The same test applied to lead generation figures shows how far the problem spreads beyond ecommerce, the purchase costs I can show with volumes attached is what a dated number looks like, what an acceptable return really depends on covers why no industry answer exists, and the floor to calculate first gives you the arithmetic that replaces the table. For how I work through this on live accounts, the services page sets it out. To have your own figures checked against something defensible, book a teardown.