ROAS vs ROI: why a 4x campaign can still lose money

ROAS is revenue divided by ad spend. ROI is profit divided by everything it cost to earn it. A 4x ROAS means $4 of revenue per ad dollar; whether that's a profit or a loss depends on your margin, and at a 25% margin it's exactly break-even. That one substitution, revenue standing in for profit, is how most ad accounts fool their owners.

This post walks the two formulas and what the platforms actually report, the break-even math that turns ROAS into a usable number, the 2025 medians, the four ways ROAS misleads, and a division of labor for who should look at which metric. It pairs with our cost benchmarks and the budget math in how much to spend on ads.

The two formulas, and what the platforms actually report

ROAS is attributed revenue divided by ad spend, usually written as a multiple. Meta's metric is called Purchases ROAS: website purchase conversion value divided by amount spent, counted inside your attribution window (7-day click, 1-day view by default) and, where data is partial, filled in by statistical modeling.¹ Google's column is conv. value/cost, and its bidding product expresses it as a percentage: in Google's own worked example, $5 in sales per $1 of spend is a 500% target ROAS.² Meta reports a multiple, Google a percentage, and neither reports profit.

ROI is profit divided by cost: (gain − cost) ÷ cost. In advertising that means the numerator subtracts COGS, shipping, payment fees, refunds and the ad spend itself, and the denominator carries the costs of running the program, including agency fees, creative production and tools. In practice most teams report marketing ROI: profit after variable costs, divided by the full marketing cost. Either convention works; name your denominator when you report it, because half of all ROI arguments are two people using different ones. Nothing on any ads dashboard computes any version of this for you, because the platforms don't know your costs.

So the two metrics answer different questions. ROAS answers "how efficiently does this campaign turn ad dollars into attributed revenue," which is a fine question for comparing campaigns. ROI answers "did this make us money," which is the question the business actually runs on. Trouble starts when one is used to answer the other.

Break-even ROAS: the only benchmark that matters

The conversion between the two worlds is one division: break-even ROAS = 1 ÷ contribution margin. Everything above it is profit, everything below it is loss, and no benchmark from anyone else's account changes where your line sits.

The margin you divide by matters as much as the formula. Most people plug in gross margin (price minus COGS) and get a flattering answer. The honest input is contribution margin: what's left after COGS, shipping, payment fees and expected refunds, before ad spend. Take an $80 product with $32 COGS. Gross margin says 60%, so break-even reads 1.67. Now subtract $7 of fulfillment, $3 of payment processing and $6 of expected refund cost (a 15% return rate with half the value recovered): contribution is $32, margin 40%, and real break-even is 2.5. An account holding a steady 2.2 ROAS looks profitable under the first calculation and loses money under the second, which is why "we grew revenue all year and have nothing to show for it" is such a common ecommerce autopsy.

If you pay an agency a percentage of spend, that goes in too: at a 12% fee, every media dollar costs $1.12, and your break-even rises with it. Our free ROAS and break-even calculator does this arithmetic, agency-fee toggle included.

One more decision before you use it: which ROAS faces which break-even. Platform ROAS against break-even is a campaign steering test. Blended MER faces a blended break-even, all revenue over all spend. The expensive mistake is scaling to blended break-even: the blend includes repeat revenue you would have earned anyway, so acquisition can be deeply underwater while the blend looks fine. Scale to new-customer ROAS, against a break-even that only credits the repeat purchases you can actually model. And check what your pixel calls revenue while you're at it, because pre-discount versus post-discount changes every number in this article.

Work out your break-even ROAS before you look at any benchmark. It converts every dashboard number from "sounds good" into a verdict.

What ROAS actually looks like in 2026

The most-quoted benchmark in this space is that "a 4:1 ROAS is good." It traces back to an undated commerce-platform blog post with no dataset behind it, and it survives because it's round. The measured numbers are lower. Across roughly 35,000 ecommerce brands in Triple Whale's full-year 2025 data, the median platform-reported Meta ROAS was 1.86; across about 18,000 brands, the Google Ads median was 3.52, down about 10% year over year.³ ⁴ TikTok sits somewhere between them, mid-1s to low-2s depending on whose panel you trust.⁵

The Google-vs-Meta gap is structural, not a quality verdict: search harvests declared intent while Meta creates demand, so the same brand base posts roughly 1.9x the ROAS on Google. The uncomfortable reading arrives when you set the medians against category margins. Three caveats belong inside this table rather than under it: the ROAS medians are platform-attributed, first-purchase numbers from one vendor's DTC-skewed customer base; the margins are our estimates from published category data; and, per the attribution section below, the dashboard number can be off in either direction. So read the last column as "the dashboard alone can't demonstrate profitability at the median," not "the median brand is losing money."

*Margins are our estimates from published category gross margins with typical shipping, fees and refunds subtracted; your own number outranks them.⁶ Push the estimates around and some verdicts flip: beauty at a 62% margin clears, while electronics doesn't clear at any believable margin. The shape holds though: at the median, first-purchase Meta economics can't be shown to clear break-even in several big verticals. The brands making the math work are either well above median, profitable on the blend with Google and email, or counting on repeat purchases the first-order math can't see. Brands retain about 28% of customers for a second order on average, so the repeat story is real for some categories and wishful for others.⁷ If your plan needs LTV to work, that's a model to build, not a sentence to say.

Four ways ROAS misleads while ROI tells the truth

1. It counts revenue you don't keep. Here's the same $80 product from the break-even section, sold through a campaign holding a 4x ROAS:⁹

A 4x ROAS on this order returns $12.00 of contribution, or $0.60 per ad dollar. At 3x, the same order returns $5.33, or $0.20 per ad dollar. Fixed costs, salaries and tools still come out of both, and they stay out of the table deliberately: they don't change with the next ad dollar, so they don't belong in the scaling decision. And this is a decent cost structure. Run the same table at electronics margins, around 20% contribution before ads, and a 4x campaign loses money outright: 4 × 20% hands back $0.80 for every ad dollar spent. That's the title of this post, worked out.

2. It books refunds as revenue. Platforms report conversion value at purchase time and never claw it back. The NRF projects 15.8% of US retail sales returned in 2025, and roughly 19% of online orders.⁸ (Our examples assume a 15% return rate; apparel runs far above it.) So a fashion brand's true ROAS is materially below the dashboard number every single day, and worse: value-based bidding keeps optimizing toward the revenue you give back. Google supports refund-adjusted conversions by order ID; almost nobody feeds them.¹⁰

3. Its attribution errs in both directions, and you don't know which. The over-crediting direction is the famous one. When eBay paused its paid brand-search ads for a landmark 2015 study, 99.5% of the click traffic came back through organic search anyway.¹¹ That's a decade old now, but modern holdout tests keep finding the same shape: practitioners running geo holdouts report retargeting delivering 25 to 50% of its dashboard ROAS and branded search 10 to 25%, meaning reported figures 2 to 10x reality.¹² Taylor Holiday's working rule for Google Performance Max is to assume incremental ROAS at one third of the reported number until you've tested it yourself.¹³

The direction most people miss: platforms also under-report. Haus ran 640 incrementality experiments on Meta and found that for every $100 of platform-attributed DTC revenue, about $115 was actually incremental, with upper-funnel campaigns under-credited by as much as 6x.¹⁴ On YouTube the same shop measured 3.4x more incremental lift than Google Ads reported.¹⁵ Put the two directions together and the conclusion isn't "ROAS is inflated," it's worse: the error is large and you can't predict its sign. A metric with that property can steer a campaign; it cannot be the number the business runs on. The dashboards are, after all, each platform grading its own homework. And none of this is evidence that an agency cooked anyone's books: nobody's media buyer invents attributed ROAS, the gap is a property of the measurement everywhere, and the fix is a holdout test that client and agency run together, not an accusation.

4. The average hides what the next dollar earns. Response curves flatten: the first dollars buy your easiest customers, and a channel showing a 4x average ROAS can be earning well under 1x on its marginal dollar.¹⁶ This cuts both ways. Cutting budget almost always raises average ROAS while shrinking total profit, which is why "maximize ROAS" as an instruction quietly means "shrink." And spending past break-even destroys money at a healthy-looking average: Common Thread Collective published a case where an apparel brand spent $2.14M at a 1.63 acquisition MER when the break-even math said stop around $1.6M, roughly $400K of negative contribution margin hidden inside a plausible blended number.¹⁷

The metrics between ROAS and ROI

A family of metrics exists precisely to patch these holes. Worth knowing which hole each one patches:

One caution on that last row: the LTV:CAC "3:1 rule" is David Skok's rule of thumb from mature public SaaS circa 2013, not a derivation, and Bill Gurley was warning about overconfident hands on the LTV formula back in 2012.¹⁸ If your LTV assumption is doing the heavy lifting in a profitability story, the assumption is the story.

The strongest signal that revenue-ROAS is on its way out comes from the platforms themselves. Google now lets accounts feed COGS and bid on gross profit, claiming about 15% more campaign profit than revenue bidding (their number, so season accordingly).¹⁹ Meta launched Incremental Attribution in 2025, optimizing toward conversions its model believes wouldn't have happened without the ad, and by mid-2026 Haus's independent testing found it outperforming standard attribution.²⁰ When the scorekeepers start selling corrections to their own scoreboard, the argument is over.

B2B: mostly skip ROAS

Everything above assumes a purchase happens near the click. B2B lead gen doesn't work that way, and ROAS math collapses on contact with it. The median B2B sales cycle runs about 84 days, and larger deals run far longer, so a 30-day ROAS window catches a sliver of the return: lead-gen campaigns reading 0.1 to 0.5x at 30 days are often performing exactly to plan.²¹ HockeyStack's benchmark panel of 70+ B2B SaaS companies reported LinkedIn pipeline ROI reaching about 6x only by the third quarter after spend.²² Judging that spend at day 30 on a revenue ratio would have killed it at its most productive moment.

The metrics that work instead: cost per qualified opportunity rather than cost per lead (First Page Sage's client data puts the average B2B cost per SQL at $1,357 against a $198 CPL, a 7x gap that cheap-lead campaigns hide behind²³), pipeline dollars per ad dollar, and CAC payback: the 2026 Benchmarkit data on 342 SaaS companies puts the median at 16 months, with the top quartile under 6.²⁴ The CFO framing is the most honest one here: long payback with positive unit economics is a financing question, while negative unit economics is the disease tech CFO CJ Gustafson names when he writes "you can't trade a buck for eighty-five cents forever."²⁵ We've written up what the funnel math looks like in practice in what $5K a month buys a B2B SaaS.

Who looks at which number

None of this means ROAS is useless. It means ROAS is a steering wheel, not a scoreboard. The stable arrangement we've seen work assigns each metric to the person whose decision it actually supports:

• The buyer, daily: platform ROAS. For comparing campaigns and creatives inside one platform, same objective, the attribution error mostly cancels out. It's the fastest read on where delivery is working. Steering, not scoring.
• The operator, weekly: MER against break-even, plus NC-ROAS. Total revenue over total spend can't be argued with by an attribution model, and the gap between blended MER and new-customer MER shows how much of "performance" is loyalty getting re-billed as acquisition. One caveat MER carries: it credits ads with revenue you'd earn anyway, so judge its trend against your own history rather than its level. And judge scaling decisions in contribution-margin dollars, not in the ratio.
• Finance, monthly: marketing ROI with the whole program in the denominator. Media, agency fees, creative, tools. A $30K media month at a reported 6.7x, with $4,500 of agency fees and $3,000 of creative, on a 45% contribution margin: roughly $90K of contribution against $37.5K of program cost, a 141% marketing ROI. Still good, and a long way below what the 6.7x implied. Finance also owns the thing no ratio shows: cash timing. Ad spend clears this week, contribution arrives after platform payouts, and the inventory was paid for months ago, so two accounts with identical ROI can have very different maximum safe spend.

For agencies, this split is retention strategy, not accounting hygiene. Agency operator Jordan Glickman's line rings true: "Clients rarely leave agencies because performance was bad. They leave because they could not tell whether performance was good or bad, and nobody on the agency side helped them figure it out."²⁶ An agency reporting platform ROAS alone is betting the relationship on a number it doesn't control and can't fully defend. Reporting MER against the client's break-even, with platform ROAS as the steering detail, is a report that survives the CFO's second look. That's the shape we've built our cross-platform reports around: Meta, Google and LinkedIn in one view, judged against one baseline instead of three self-reported scoreboards.

And if you're carrying one sentence out of this post, take the judgment call: any number that decides whether more money gets spent must contain your margin. ROAS never does. That's not a flaw to fix; it's a reason to demote it.

Frequently asked questions

Sources

1. Purchases ROAS metric definition and modeling caveat — Meta Business Help Center
2. Conv. value/cost and the 500% worked example — Google Ads Help, About Target ROAS bidding
3. Meta median ROAS 1.86 and vertical medians, full-year 2025, ~35,000 ecommerce brands — Triple Whale, Facebook Ad Benchmarks
4. Google Ads median ROAS 3.52, ~18,000 brands, 2025 — Triple Whale, Google Ads Benchmarks
5. TikTok ROAS medians, 2.21 vs 1.41 — Triple Whale, TikTok Benchmarks and Varos April 2025 panel via Superscale
6. Category gross margin ranges used for contribution estimates — NYU Stern (Damodaran) margins by sector and Level CFO ecommerce benchmarks
7. Average second-purchase retention ~28.2% — Rivo, Repeat Purchase Rate Benchmarks
8. 15.8% of 2025 US retail sales returned; ~19% of online orders — NRF × Happy Returns, 2025 Retail Returns Landscape
9. Per-order P&L framing (our table uses its own consistent assumptions) — Niblin, High ROAS But Not Profitable (Mar 2026)
10. Refund-adjusted conversions in Google Ads — Space Ads, Returns and Conversion Adjustments (2025)
11. 99.5% of branded paid-search clicks retained when ads paused (2015 study, flagged for age) — Blake, Nosko & Tadelis, Consumer Heterogeneity and Paid Search Effectiveness (Econometrica)
12. Practitioner holdout ranges: retargeting iROAS 25–50% of reported, branded search 10–25% — BreakevenHQ, Incremental ROAS (May 2026)
13. "Assume PMax incremental ROAS at one third of reported" — Taylor Holiday (Jul 2024)
14. 640 Meta incrementality experiments: $115 incremental per $100 attributed, upper funnel under-credited up to 6x — Haus, The Meta Report (Jul 2025)
15. YouTube 3.4x more incremental lift than reported, 190 tests — Haus, Do YouTube Ads Perform? (Mar 2025)
16. Average vs marginal ROAS and response curves — Mutt Data, Optimizing for ROAS vs Marginal ROAS (Jun 2025)
17. Apparel brand: $2.14M spent at 1.63 aMER vs ~$1.6M break-even — Common Thread Collective, Spend, aMER and Spending Power (Aug 2025)
18. LTV:CAC 3:1 origin and the LTV caution (2012–2013, flagged for age) — David Skok, SaaS Metrics 2.0 and Bill Gurley, The Dangerous Seduction of the LTV Formula
19. Google gross-profit bidding with COGS data, +15% claimed — Wolfgang Digital (2025) and Google, COGS conversion data
20. Meta Incremental Attribution rollout and independent read — Social Media Today (Sep 2025) and Haus follow-up (Jul 2026)
21. 84-day median B2B sales cycle; 0.1–0.5x ROAS at 30 days is normal — Optifai sales-cycle compilation and GrowthSpree (2025)
22. LinkedIn pipeline ROI ~6x by the third quarter, 70+ B2B SaaS companies — HockeyStack Labs, 2025 LinkedIn Ads Benchmark Report
23. Average B2B cost per SQL $1,357 vs $198 CPL (agency client data) — First Page Sage via Omnibound
24. Median CAC payback 16 months, top quartile ≤6, FY2025 actuals of 342 companies — Aleph × Benchmarkit 2026
25. "You can't trade a buck for eighty-five cents forever" — CJ Gustafson, Mostly Metrics
26. Agency churn as a reporting-legibility problem — Jordan Glickman (May 2026)