The dashboard says you're doing great. It's reporting an average. The dollar you're about to commit gets spent at the margin, and marginal ROAS is the number that decides whether that dollar makes money. Most budget fights at scale come down to smart people arguing over blended MER while they're deciding at the margin, and nobody in the room asking the one question that changes both: are the words still working?
My roots are in copy. These days I work as a digital futurist, the coach on the sideline with the clipboard, and you can't coach a $10M-a-month room without the math. So here's the math, and then here's what it can't see.
Where the industry's $294.6 billion went
One major industry revenue report for full-year 2025, compiled by an accounting firm and released in April 2026, put US internet ad revenue at $294.6 billion, up 13.9% on the year [1]. Social grew about 32.6% to $117.7 billion. Search grew 11% to $114.2 billion, down from close to 16% growth the year before. Digital video rose about 25% to $78 billion, commerce media rose 18% to $63.4 billion, and programmatic grew about 20% to $162.4 billion.
That's a fair map of where the industry put its money. It's also careful about what it is: revenue reported by the companies that sell the ads, compiled and unaudited.
Why the industry's curve isn't your curve
Here's my read, and it's where a lot of $10M-a-month teams get misled. Aggregate growth by channel is a record of where the marginal dollar used to be cheap. Social grew fastest because, for a lot of advertisers, the next dollar there was still working. That doesn't mean the next dollar there is still working for you.
Your account doesn't live on the industry curve. It lives on its own response curve, one per channel, and every one of those curves bends. The first $500,000 in a channel finds the people closest to buying. The next $500,000 finds people a little further away. Eventually you're paying to reach people who were never going to buy, or people who were going to buy anyway.
The market reportshows the tide. Yourresponse curves showthe shore.
One bending curve, two diagnoses
The technician looks at a bending curve and sees a spend problem. Pull budget, shift budget, re-bid, rebuild the model. But it treats the curve like weather, something that happens to you.
The strategist looks at the same curve and sees a message problem first. A curve bends when the words stop reaching new people. The first $500,000 found the buyers your current message already speaks to. The next $500,000 is asking that same message to convince people it was never written for. Of course the return falls. You're shouting the same sentence louder at people who didn't care the first time.
Change the sentence and you change who leans in. A new angle, a new buying situation named out loud, a sharper offer, and suddenly there's a fresh pocket of people the old message never touched. The curve flattens out later. Marginal ROAS climbs back. Same channel, same bid strategy, same budget. Different words.
The dashboardmeasures themessage. Somebodystill has to writeit.
The technician's model tells you where the money stops working. The strategist's job is to push that line further out. You need both. But only one of them can make the number go up without spending more.
The Headroom Rule: money goes where the gap is widest
I use the Headroom Rule to find where the biggest gain hides in a funnel. Applied across channels it says one thing: money belongs where the gap between the next dollar's return and your break-even is widest, and nowhere else.
That gap is headroom. A channel with a beautiful blended ROAS can have no headroom left. A channel with an ugly blended number can have a lot.
Two numbers make it work:
- Marginal ROAS: incremental revenue divided by incremental spend, measured as spend moves. It's the change in total revenue, read outside the platform's own attribution.
- Break-even marginal ROAS: 1 divided by your contribution margin. At a 50% contribution margin, break-even is 2.0. Any marginal dollar returning less than 2.0 in revenue is losing money, even if the blended picture still looks great.
Reading headroom every week, before the budget meeting forces the question, is the habit the iFix Lab is built to drill.
The math: how a smooth MER hides a $500,000 loss (illustrative)
These numbers are a model built for illustration. They're a way to think about your account, and they won't forecast it. Assume a 50% contribution margin, so break-even marginal ROAS is 2.0.
One account, three spend levels:
- At $8M a month total, revenue is $28M. MER is 3.5.
- At $10M, revenue is $33M. MER drops to 3.3. The extra $2M brought $5M, so marginal ROAS is 2.5. Contribution on that step is $2.5M against $2M of spend. You made $500,000. Worth it.
- At $12M, revenue is $36M. MER is 3.0, which still looks fine on a slide. But that last $2M brought $3M, a marginal ROAS of 1.5. Contribution is $1.5M against $2M of spend. You lost $500,000 to grow.
MER fell from 3.5 to 3.0 across both steps, and it fell smoothly. Nothing on the blended dashboard says the first step made money and the second one lost it. That's the whole problem in one table.
Two channels, same total spend:
- Channel A runs at $4M a month, blended ROAS 3.0, marginal ROAS 1.4.
- Channel B runs at $1M a month, blended ROAS 2.2, marginal ROAS 2.6.
Move $500,000 from A to B. Because A's curve is bending, the last $500,000 there was earning less than 1.4, call it around $650,000 in revenue. In B, the next $500,000 earns a bit under 2.6 as you climb its curve, call it $1.2M. Same spend. Roughly $550,000 more revenue, and about $275,000 more contribution at a 50% margin.
The channel with theworse blended numberwas the better placefor the money.
Now the part the spend model leaves out. Say instead of moving that $500,000, you put a new message into Channel A, one built for a buying situation the account has never named. In this illustration, if the new angle lifts A's marginal ROAS from 1.4 to 2.2, the same $500,000 now earns about $1.1M without leaving the channel. The words did what the reallocation did, and you kept the option to reallocate too.
MER drifts down smoothly. The last $2M lost $500,000 and the blended line never said so.
Ran out of audience, or ran out of message?
When marginal ROAS falls in a channel, ask one thing before anyone touches the budget: did we run out of audience, or did we run out of message?
You can tell the difference. If frequency on your best concepts is climbing and new concepts revive marginal returns, you ran out of message. That calls for a copy brief before anyone touches the budget. If fresh concepts with genuinely different angles, and I mean different promises to different people, because new colors on the same promise don't count, still can't move marginal ROAS, you ran out of audience at this price, and the money belongs somewhere else.
Most teams never run that second test honestly. They ship variations of the winning ad, watch them fail, and call it saturation. The team ends up testing everything and losing clarity, when what it's really looking at is one sentence wearing ten outfits.
Three ways averages punish the growth that pays
- MER targets punish growth that pays. A finance team that sets a hard MER floor will block the $8M to $10M step in my example, because MER fell. That step made money. Floors on averages kill profitable scale.
- Platform-attributed ROAS overstates marginal returns most where you spend most. The more you spend, the more of your existing demand the platform gets to claim. Attribution can look flat while true marginal return collapses.
- Channel owners defend averages. If each channel has its own lead or agency, nobody's paid to say their marginal dollar is weak. And nobody's paid to say the message is tired. That's an incentive design problem, and it's fixable.
Why scaling your winners stops paying between $1M and $10M a month
The popular advice says scale your winners. At $1M a month that mostly works, because you're still low on every curve and almost every channel has headroom.
Between $3M and $10M, scaling winners turns into overfeeding the channel with the best history and the message with the best history. The blended number stays strong because the early, cheap demand is still inside it. The marginal number goes soft months before anyone sees it. By the time MER finally drops enough to trigger a meeting, you've been spending past break-even for a quarter, on words that stopped landing two quarters ago. And nobody in the room can say why the playbook that built the account stopped paying.
Past $10M a month, the hard part is the room
At this level the math isn't the hard part. The organization is.
- Every channel is near saturation somewhere. Headroom stops being a channel question and becomes a segment, geography and message question. You'll find more headroom inside a channel, in an underserved region or a buying situation nobody's written to yet, than by moving whole budgets between channels.
- You need curves you can defend to finance. Platform numbers won't survive a CFO who's been burned. Geo-based holdouts, run on a calendar, become the curve estimates the room accepts. Budget follows whichever measurement finance trusts.
- Small marginal errors are big money. If you're reallocating $3M a month on curves that are off by 0.2 points of marginal ROAS, that's $600,000 a month of revenue landing in the wrong place. The cheapest way to buy back that 0.2 is usually a better message.
The one-page marginal ROAS sheet
Whether you're the in-house growth lead defending the plan, the agency owner reporting up, or the CFO who signs the increase, the whole budget fight fits on one page. Three numbers carry it: break-even marginal ROAS, each channel's last real marginal read, and the date that channel last got a genuinely new message. A room that can't fill in the third number hasn't earned the word “saturated.”
The full sheet, with the funding order and the stop line, is a template members get inside the iFix Lab.
Questions senior buyers ask
The dashboard says we're doing great. How would I know we're already spending past break-even?
Not from the dashboard. Blended numbers stay healthy for months because the early, cheap demand is still inside them. Compare the last real spend step's marginal ROAS with 1 divided by your contribution margin. If it's under, you're buying growth at a loss.
What's a good MER at scale?
It depends on your margin and your mix of new and repeat buyers. The sharper question is whether the next dollar clears break-even marginal ROAS.
Isn't MER the honest number because it can't be gamed by attribution?
It's honest about the past and blind about the next dollar. Use MER as the health check and marginal ROAS as the steering wheel. They answer different questions.
How do we measure marginal ROAS without a perfect model?
Change spend deliberately and watch total revenue, setting attributed revenue aside. A step change in one channel, held long enough to cover your purchase cycle, or a geo holdout where some regions get the increase and matched regions don't. Rough and recent beats precise and stale.
How do I know a falling curve is the message and not the market?
Run two or three concepts that make a different promise to a different buyer, at meaningful spend. Variations of the old ad don't count. If marginal ROAS recovers, it was the message. If it doesn't, it's the market at this price.
When is spending past break-even justified?
When you're deliberately buying customers whose later purchases you can prove, or buying share in a market you expect to own. Then set a payback window, convert it to a target, and treat it as an investment line with its own owner.
Before you approve the next increase
Write break-even marginal ROAS at the top of the page. Then ask each channel owner two things: what the last dollar returned, and when that channel last got a new message. If the room goes quiet on the second question, you just found your cheapest growth.
Learning to ask that question every week is what the iFix Lab trains. The weekly Signal Briefing gives you what moved across the field, each item with a verdict (hype, watch or working) and a move, so you walk in reading the margin while everyone else defends the average. If you're already past break-even, every week before the next budget meeting costs real money. Join the iFix Lab for $97 a month. Cancel any time, and month one carries a 14-day refund. The next note goes after the measurement problem underneath all of this: why your best ad can make your incrementality test lie.