"You need to be able to survive, I think, a 1.5X ROAS." Sean Frank, the CEO of Ridge, said that on a podcast about building a business that lasts. Most operators hear it as a floor for the account. It is a floor for the next dollar.
Here is the Shopify account that trips over it. Average ROAS holds at 3x for the month. Feels safe. Then the founder looks at the last raise, from $1,000 to $1,500 a day, and the extra $500 brought back revenue at 1.4x. That is the marginal ROAS, and it is already under Frank's line while the account still reads 3x.
This is for founder-operators and growth leads at $500K–$30M who are mid-scaling-push and can't work out why the last budget increase did worse than the average said it would. Marginal ROAS is the number that answers it, on Meta and on Google. The calculator below works it out from four figures you already have.
Key Takeaways
- → Marginal ROAS = extra revenue ÷ extra spend between two spend levels. It tells you what the next dollar earns. Average ROAS tells you what every dollar earned so far, blended together.
- → It falls as you raise budget because the cheapest demand gets bought first: auction depth, audience saturation and creative reach all bite on the increment, not the average.
- → An account can read 3x on average while the last step earned 1.4x. Under a 40% contribution margin (break-even 2.5x), that step lost money and the dashboard never blinked.
- → The move depends on which of eight budget states the marginal reading puts you in. "Raise 20%" is not one of them.
Average ROAS Lies About the Next Dollar
Average ROAS is every dollar of revenue divided by every dollar of spend. Inside that average sits the cheapest demand you were ever going to get: people searching your brand name, retargeting to last week's cart abandoners, the first $300 a day that reaches your warmest lookalike. Those dollars earn 6x, 8x, sometimes 12x. They hold the average up long after the increments stopped paying.
Every raise buys the next block of demand, and the next block is more expensive than the last. The average dilutes slowly. The marginal number shows it immediately. So the operator who watches the average sees a gentle slide from 3.4x to 3.1x and calls it noise, while the operator who watches the marginal number sees the last step earn 1.4x and knows exactly what happened.
Andrew Lolk pulled this apart on one client's Performance Max account. Brand searches inside the campaign ran at 5.5x ROAS; non-brand ran at 2.2x. Blended, the campaign looked like profitable scale. It was mostly cheap branded demand, people who were already buying, booked as prospecting. Same disease, different axis: the average hid what the marginal unit was actually earning.
You have probably said a version of this yourself. "every time i push past $1k/day my cpa doubles. back off and it heals. stuck." (That is a composite line from our operator research, and it is the marginal curve talking.)
What Marginal ROAS Is, With the Worked Example
Marginal ROAS = (revenue at the new spend level − revenue at the old spend level) ÷ (new spend − old spend). Same account scope, same length of window, revenue from Shopify orders rather than a platform's own conversion column.
Walk through an account stepping up from $500 a day.
| Daily spend | Daily revenue (Shopify) | Average ROAS | Marginal ROAS on the last $500 |
|---|---|---|---|
| $500 | $2,000 | 4.0x | 4.0x (first step) |
| $1,000 | $3,400 | 3.4x | ($3,400 − $2,000) ÷ $500 = 2.8x |
| $1,500 | $4,100 | 2.7x | ($4,100 − $3,400) ÷ $500 = 1.4x |
Read the two right-hand columns together. The average never fell below 2.7x. The marginal figure went 4.0x, 2.8x, 1.4x. If this brand runs a 40% contribution margin before marketing, its break-even is 2.5x (that arithmetic is in the contribution-margin check). The second step still cleared it. The third step lost money on every one of its $500, and the account "looked fine" the whole time.
Marginal CPA is the same idea for cost per order: extra spend divided by extra orders. Use whichever your team already argues about.
Why Marginal ROAS Falls When You Raise the Budget
Three mechanisms, and they stack.
Auction depth. At $500 a day the platform bought you the impressions it could win cheaply for your creative. The next $500 has to win impressions that cost more, against advertisers who value them more. And the floor keeps rising: Meta's average price per ad was up 12% year over year in Q1 2026. Google Search behaves the same way through impression share: the last 10 points of share cost more than the first 60.
Audience saturation. More budget on the same targeting reaches the same people more often. Frequency climbs. The additional person you reach is, on average, less likely to buy than the person you reached at half the spend. Past that point you are paying to re-show the ad rather than buying new demand.
Creative reach limits. A concept works for the people it works for. Once it has reached them, extra budget buys the people it does not work for. This is why a budget raise after a fresh winner often holds, and the same raise on a concept that is 30 days old collapses. Budget scales a creative; it cannot replace one.
Two platform effects muddy the read. A large single-step raise can reset learning on Meta, so the first few days after a raise are noisier than the marginal number suggests. And conversions lag: read the step after a full conversion cycle rather than after 48 hours. The daily noise is a separate question, and we cover it in the signal-or-noise piece.
One more thing the curve tells you. If you back off and efficiency heals, you did not break anything. You found the shape of the curve. "Back off and it heals" is the tell of saturation, and a saturating account needs a different growth path (new creative, new product, new geography) rather than another go at the same slider.
How to Estimate Yours Without an Econometrics Team
You do not need incrementality tests for this. You need two spend levels and matched windows.
- → Take two periods from the last six to eight weeks separated by a real budget step (20% or more), each at least one full conversion cycle long, same days of the week, no promotion in either.
- → Pull total ad spend for the scope you changed (the whole account if you raised the whole account; one campaign if you raised one campaign).
- → Pull Shopify revenue for the same windows. Net of refunds and discounts. Not the platform's conversion value: Meta and Google both self-attribute, and their totals will disagree with the orders that exist.
- → Subtract, divide, compare with the average. If you only raised one campaign, platform-reported revenue for that campaign is fine for direction, as long as you label it platform-reported.
Or type the four numbers in here.
Marginal ROAS calculator
Use the same window length for both periods (a week against a week, a fortnight against a fortnight). Any currency.
The calculator works from spend and revenue only; bring the margins you share with Drew to the decision, not to the tool. The budget-state read is published method, from the table below.
A word on precision. This is a two-point estimate of a curve, on data with day-to-day noise. It is not a lift study. But it is the difference between scaling on a number that describes the past and scaling on one that describes the decision in front of you, and two points beat zero.
The Eight Budget States, and Which Marginal Reading Maps to Which
Operators who scale well do not think in percentages. They think in states. Each state has a marginal signature, an evidence requirement and a move. "Raise 20% every three days" is not on the list because it ignores all three.
| State | What the marginal reading looks like | The move |
|---|---|---|
| Invest aggressively | Marginal close to average, well above break-even, on enough conversions to trust it, with stock and creative to spend into | Raise in measured steps; re-read after each conversion cycle |
| Scale cautiously | Marginal above break-even but sliding, or the evidence is thin (few conversions, short window, recent edit) | Smaller steps, longer windows, fix the evidence problem first |
| Maintain | Average is fine, marginal is unclear or flat around break-even | Hold. Spend the effort on creative, offer or conversion rate instead |
| Constrain | Marginal has dropped to a fraction of average; the account is useful but the current spend sits at its limit | Cap the budget; stop testing the ceiling with money |
| Harvest | Demand is finite and mostly captured (brand search, retargeting); more budget buys re-shows, not buyers | Fund it to capture existing demand and no further |
| Explore | There is no marginal reading yet; the spend is buying information (new concept, new market, new product) | Fixed test budget, fixed window, judge it on learning, not ROAS |
| Reduce | Marginal below break-even on a clean read, and the cause is saturation or a fading creative rather than a tracking break | Step back to the last level that paid; move the difference to the best marginal opportunity |
| Pause | Continued spend has negative expected value, or it breaks a rule you set (stock, cash, margin floor) | Stop, with a written condition for restarting |
Notice what the table is missing: a ranking of campaigns by ROAS. Moving money by sorting a column is how brands over-feed brand search and starve the concept that would have carried Q4.
What a marginal decision actually weighs, in one breath: the expected contribution of the next unit of spend, how confident you are in that estimate, how much headroom the demand and creative have left, whether the campaign matters strategically, whether the customers or products it brings in are the ones you want (margin you know, stock you have, repeat behaviour you have measured at product level), minus the downside if you are wrong. Platform ROAS is one input to the first term. That is all it is.
Where Should the Next Unit of Budget Go?
Once you have marginal readings for two or three places (the core Meta campaign, the Google Shopping set, the new concept in testing), the allocation question answers itself: the next dollar goes to the highest marginal contribution you can trust, until its marginal reading falls to meet the others. Meta versus Google is the same question at channel scale, and it gets its own piece in this series.
One marginal-specific warning, because it is the mistake underneath most channel arguments. Ranking channels by their average platform ROAS tells you nothing about what the next dollar earns in either one. Google Search at 6x average may be fully harvested with a marginal reading near 1x, while Meta at 2.4x average has a marginal reading of 2.2x and room to grow. The averages point one way; the money should go the other. (And neither platform's ROAS is comparable with the other's in the first place, since each attributes to itself. That is a which-number-is-real question, and Shopify orders are the answer.)
Marginal vs Blended vs Platform ROAS: Which Number Answers Which Question
Three numbers, three jobs.
- → Blended ROAS (MER): total Shopify revenue ÷ total ad spend. Is the whole business still efficient? An honest ratio, credits every channel with all revenue, so it cannot allocate. What it tells you and what it can't, and how to pull it across Meta, Google and Shopify.
- → Marginal ROAS: what the next dollar earns, in one scope. The scaling number. This piece.
- → Platform ROAS: direction inside one channel, self-attributed, never additive across channels.
Incremental ROAS, the lift-test sense of "what would not have happened without the ads", is a fourth number with its own methods and price tag. The full router across all four is next in this series.
How Drew uses this. Ask Drew where your next ad dollar should go. Its budget recommendations weigh the margins you share with it, your inventory and your repeat behaviour, not platform ROAS alone, and they come with the reasoning: what to scale, reduce, pause or leave alone this week, how much, and why. Drew reads yesterday's Shopify orders and Meta and Google spend every day, so the marginal read is on real revenue rather than the platform's claim. The eight states above are method; Drew doesn't stamp your account with one on its own. It brings you the numbers and the argument, and the raise is still your call, sized to the evidence rather than to a rule.
To see the shape of this on your own account, pricing is published and flat and the free plan doesn't need a card.
Key Takeaway
Marginal ROAS is extra revenue divided by extra spend between two budget levels. It is the only ROAS figure that describes the decision you are about to make, because it measures the increment rather than the history. It falls as you scale, through auction depth, audience saturation and creative reach, and it can sit below your break-even while the average still reads 3x. Estimate it from two matched windows and Shopify revenue, read it against 1 ÷ contribution margin, and let the eight budget states decide the move. An account that "holds 3x" while the next dollar earns 1.4x is already under Sean Frank's 1.5x survival line. The average will be the last number to tell you.
Frequently Asked Questions
What is marginal ROAS?
Marginal ROAS is the revenue earned by the last increment of ad spend, divided by that increment: (revenue after − revenue before) ÷ (spend after − spend before), measured over matched windows. Average ROAS describes every dollar spent so far. Marginal ROAS describes the next one, which is the only dollar a budget decision can affect.
How do I calculate marginal ROAS?
Take two periods separated by a real budget step, each at least one full conversion cycle long, on the same days of the week and without a promotion. Subtract the earlier period's Shopify revenue from the later one's, subtract the earlier spend from the later spend, and divide. Use Shopify revenue net of refunds rather than platform-reported conversion value, because Meta and Google each attribute the same orders to themselves.
Why does my ROAS drop every time I raise the budget?
Because the platform bought your cheapest demand first. Each raise buys impressions that cost more, reaches people who are less likely to buy, and stretches a creative concept past the audience it works for. The average slides slowly while the marginal reading falls fast. If efficiency recovers when you back off, you have found the saturation point of the current setup, and the next level of spend needs new creative, a new product or a new market rather than more budget on the same thing.
Is marginal ROAS the same as incremental ROAS?
No. Marginal ROAS is an observed estimate of what the last spend step earned, taken from two points on your own account. Incremental ROAS is a causal claim, what would not have happened without the ads, and needs a holdout, a geo test or a model to measure. Marginal is the cheap, weekly scaling read. Incremental is the expensive, occasional truth check.
What marginal ROAS should I stop scaling at?
At your break-even, which is 1 ÷ contribution margin before marketing. A brand at 40% margin breaks even at 2.5x; at 30% it needs 3.33x. Stop raising when the marginal reading approaches that floor, unless you are deliberately buying customers below first-order break-even because their repeat behaviour, measured at product or cohort level, pays it back.
My ROAS drops every time I raise the budget. How do I scale without breaking the campaign?
Raise in steps sized to the evidence rather than to a percentage rule, read each step after a full conversion cycle against Shopify revenue, and stop at the step where marginal ROAS meets your break-even. When it does, scale sideways instead of up: a new creative concept, a product with better margin and stock, a new geography or a stronger offer. The eight budget states in the article map each reading to its move.
Written by Sumit Bansal, co-founder of Datadrew. Published 3 September 2026. Cost and quote sources are linked inline. Datadrew deals in realised numbers with Shopify orders as the source of truth; we don't adjudicate which channel deserves credit for a sale.