ROAS Is Lying to You: Judge Ad Creative on Break-Even CPA

ROAS is revenue divided by spend. Nothing in that formula knows what a unit costs you.

Two product lines can post an identical 2.1 ROAS while one prints money and the other quietly funds its own decline. The difference is not the platform, the audience or the video. It is the margin on the thing being sold, and ROAS is structurally blind to it.

That blindness is survivable when ROAS is a blended account number over a stable product mix. It gets expensive the moment you use it to judge individual creatives, because that is a decision you make several times a week, per product. The errors run in both directions. Your fat-margin products break even at a low ROAS and your thin-margin ones need a much higher one, so a single account-wide bar lands somewhere between the two: high enough to fail fat-margin creatives that were already profitable, low enough to pass thin-margin creatives that lose money on every order.

Here is the number to judge on instead, and how to build it in about ten minutes.

Break-even CPA is just your contribution margin

For any product line, break-even CPA equals the money left from one order after every cost that scales with that order. Acquire the order for less and you make money. Pay more and you lose money, whatever the ROAS column says.

Build it per order (not per unit) if you sell bundles or multiples:

Selling price actually collected, after discounts minus COGS minus shipping and fulfilment minus payment and marketplace fees minus a returns and refund allowance at your real rate minus any other per-order variable cost = contribution margin per order = your break-even CPA

Three things decide whether that number is honest:

• Use contribution margin, not the gross margin off your P&L. Overhead, salaries and software do not belong in an acquisition ceiling. Every cost that moves with order volume does.
• Subtract returns. A 12 percent return rate in a category with restocking losses moves this line further than most sellers expect, and it is the single most commonly skipped input.
• Do not inflate it with unproven LTV. If you have measured repeat purchase, you can raise the ceiling on purpose and say so out loud. If you are assuming repeat purchase, you are just giving yourself written permission to overpay.

Once you have the dollar figure, the derived version is often easier to hold in your head: break-even ROAS = selling price ÷ contribution margin per order.

The 1.57 ROAS that loses money

Made-up numbers below, chosen so you can check the arithmetic rather than be impressed by it. Run yours.

A product line sells at a $69 average order value. After COGS, shipping, fees and returns, $30 per order survives. Break-even CPA is therefore $30, and break-even ROAS is 69 ÷ 30 = 2.30.

Last month the line spent $2,417 and returned $3,795 across 55 orders.

• ROAS: 3,795 ÷ 2,417 = 1.57. Under most account bars but above the panic line, so it survives the weekly review as "needs work" rather than "dead".
• CPA: 2,417 ÷ 55 = $43.95 against a $30 break-even line. That is 47 percent over.
• What actually happened: 55 orders × $30 contribution = $1,650, against $2,417 of spend. A $767 loss.

Same data, opposite verdicts. Judged on ROAS, that spend stays alive because 1.57 does not look like failure. Judged on break-even CPA, it dies the day you look, because $43.95 against $30 is not a judgement call, it is subtraction.

Note how cheap the CPA test is: one number per product line, computed once, revisited when costs move.

Why this specifically wrecks creative decisions

Volume of testing is a solved question (we did that math in how many ad creatives you need per week). Judgement is where accounts actually leak, because every test ends with a kill-or-scale decision and ROAS makes that decision incomparable across products.

Two SKUs, same $100 price, same 2.0 ROAS on the creative you are reviewing:

	Fat-margin SKU	Thin-margin SKU
Selling price	$100	$100
Contribution per order	$70	$25
Break-even CPA	$70	$25
Break-even ROAS	1.43	4.00
Verdict at 2.0 ROAS	comfortably profitable, scale it	losing roughly $25 per order, kill it

If your creative review is a spreadsheet sorted by ROAS across a mixed-margin catalogue, you are not ranking creatives. You are ranking products, and re-discovering your own cost structure once a week.

The fix is one extra column: CPA ÷ break-even CPA. Under 1.0 is profitable, over 1.0 is not, and the ratio is comparable across everything you sell. A creative sitting at 0.7 on a thin-margin SKU is a better creative than one at 1.2 on a fat-margin SKU, and only the ratio will ever tell you that.

Worth stating plainly: the ad platform will not do this for you. Every major performance platform, Meta and TikTok's GMV Max alike, ranks auctions on roughly bid × predicted conversion rate. On Meta you can push margin into that formula by passing contribution margin as the purchase value and optimizing for value; GMV Max gives you a target ROI on gross order value and no field for margin at all. Either way, on the platform you are most likely running, margin stays your job.

Before you kill it, check which stage broke

A CPA over the line means the money did not work. It does not tell you the video is at fault, and this is where teams throw away good creatives and pay to rebuild them.

Read the path in four stages instead of one number: thumbstop and CTR, landing page add-to-cart rate, cart-to-order rate, repeat purchase. A working rule: the creative owns stage 1 outright and shares stage 2 with the page. It does not own stages 3 or 4. Kill on economics, assign blame by stage.

The stage-by-stage diagnosis, including the misdiagnosis that sends teams into a production cycle they did not need, is in the creative didn't fail, your landing page did. (If the campaign will not even spend its budget, that is a third category of problem, covered in why GMV Max is not spending your budget.)

Write the kill line before it goes live

The decision to kill a creative is almost never made rationally after the fact, because by then you know what it cost to make and you have watched it forty times. Sunk cost does the rest. So write the line on launch day, in dollars, anchored to break-even CPA:

• One times break-even CPA in spend with zero add-to-carts: dead. There is nothing left to learn.
• Two to three times break-even CPA with no purchase: dead, unless stage data says the page broke.
• Above break-even CPA but converting, after a full data cycle: hold or fix. Do not scale. Scaling multiplies a negative.

Put it in the ad set name or the row next to the creative in your tracker, on the day it launches, so future-you is executing a rule rather than defending a decision.

One timing caveat: orders get credited back to the impression date, so the most recent day or two always under-reports (how attribution windows print zeros). Denominate kill lines in spend, never in days.

Change one variable at a time

Every kill line assumes the result is attributable. If you swap the character and the template at once, you learn nothing from the outcome: you have a new creative, not a variant.

So run one concept per test cell, change one thing per variant (the hook, the character, the opening line, the ratio), and give each variant its own budget line. That means testing on ABO, because CBO will starve exactly the low-spend, high-CTR variant you most needed to see (the mechanism and the two-track fix). Use CBO to scale what you have already validated, not to decide what to validate.

Where ROAS is still the right number

Blended, at account level, over a stable product mix, tracked month over month: that is what ROAS is for, and it is fine there. It is a portfolio metric. It is simply the wrong instrument for the decision you make most often, which is whether this specific video, on this specific product line, earns more money tomorrow.

The production side of the same math

Judging on break-even CPA raises your kill rate, which is correct, and it only ends well if replacements are cheap. A kill line is easy to honour when the creative took minutes to make and almost impossible when it took a two-week shoot and a talent fee. Part of why ROAS survives as the default is mundane: it is the only number the ad platform can compute for you, because your margin lives somewhere it cannot see. But it also happens to be the metric that lets you keep an expensive creative alive.

Riffkit attacks the cost side of that trade. You start from a short video that already won, and Riffkit rebuilds its formula (the hook, the pacing, the emotional beats) around your own product: new footage, product on screen, no filming, minutes per render. Variants are cheap enough to be treated as ammunition, so a disciplined kill line stops feeling like burning money and starts feeling like clearing the queue. Signing up is free and includes enough free seconds for a first video.

The short version

ROAS cannot see your margin, so it mis-ranks creatives across a mixed-margin catalogue. Compute contribution margin per order for each product line, treat it as your break-even CPA, and score every creative as CPA ÷ break-even CPA. Under 1.0 lives, over 1.0 dies. Read the four funnel stages before you blame the video, write the kill line in dollars on launch day, and change one variable at a time so the result means something.