Your dashboard says the campaign is running at a $68 cost per acquisition, comfortably under your $70 limit. The most recent chunk of budget is actually buying conversions at $111. Both numbers are correct, and only one of them should be governing your next decision.
Average hides the increment
This is the whole article, so here it is with numbers.
| Monthly spend | Conversions | Average CPA | Marginal CPA |
|---|---|---|---|
| $10,000 | 200 | $50.00 | — |
| $15,000 | 265 | $56.60 | $76.92 |
| $20,000 | 320 | $62.50 | $90.91 |
| $25,000 | 365 | $68.49 | $111.11 |
| $30,000 | 400 | $75.00 | $142.86 |
Read across the $25,000 row. Average CPA is $68.49 — under your $70 limit, so every dashboard, every report and every stakeholder says the campaign is healthy and you should keep going.
But the jump from $20,000 to $25,000 bought 45 extra conversions for $5,000. That increment cost $111 each. You breached your limit two steps ago and the blended figure concealed it, because it's still being propped up by the cheap conversions you bought at the bottom of the curve.
The rule Average CPA tells you how the campaign has performed. Marginal CPA tells you whether to spend the next pound. Only one of those is a scaling decision.
The formula is trivial: change in spend ÷ change in conversions. Two columns in a spreadsheet, and almost nobody has them.
One caveat on reading it: month-to-month noise, seasonality and creative changes all move this line, so a single bad increment isn't a ceiling. Look for a sustained trend rather than reacting to one month, and be aware that the calculation assumes your conversion tracking is sound — the ground covered in setting up conversion tracking correctly.
It's arithmetic, not punishment
Worth being clear about the mechanism, because the wrong explanation leads to the wrong response.
When CPA rises with spend, the common assumption is that something broke — the algorithm is penalising you, the account needs restructuring, the bid strategy is wrong. Usually none of that is happening.
What's happening is that the people most likely to convert get reached first, and cheapest. Each additional increment of budget reaches further into a less responsive audience and requires winning more expensive auctions to do it. Returns diminish because the audience is finite, not because you're being punished.
That distinction matters practically. If you believe it's a penalty, you respond with endless account tinkering — restructuring, rebidding, chasing settings. If you understand it's audience depth, you respond correctly: either accept the higher marginal cost because the volume is worth it, or find new audience rather than pushing harder into the exhausted one.
Birch
AI-powered platform for automating and optimizing multi-platform advertising campaigns.
Best for: Performance Marketing Automation
Vertical versus horizontal
Which leads directly to the main strategic choice.
Vertical scaling — raising budget on what already works — pushes deeper into the same audience. Simple, fast, and subject to exactly the diminishing returns above.
Horizontal scaling — adding new audiences, placements, geographies, or creative angles — opens inventory that hasn't been exhausted yet. It preserves efficiency for longer because each new line starts at the cheap end of its own curve.
Horizontal is generally the better answer, with two honest costs. Each new campaign starts without accumulated learning and needs its own conversion volume to stabilise, which means splitting a fixed budget across more lines can leave all of them under-fed. And it takes more work — more creative, more setup, more to monitor.
The practical rule: scale vertically until marginal CPA approaches your limit, then scale horizontally. Not the other way round, and not both at once, because you won't be able to attribute the result to either.
How you organise those lines matters more as the count grows, which is the account architecture question handled in structuring an account for scale and control.
The learning phase, honestly
Widely discussed, frequently overstated.
The commonly repeated advice — never increase budget by more than 20% at a time — is folklore rather than published policy. You'll find it everywhere and sourced nowhere.
The underlying mechanism is real, though. Significant campaign changes can return delivery to a learning state while the system re-optimises, and during that period performance is unstable and unrepresentative. Reading results from a destabilised campaign and reacting to them produces the classic spiral: change, panic at the noise, change again, never allow anything to stabilise.
So the defensible version is behavioural rather than numerical: make changes modest enough not to destabilise delivery, then wait long enough to read a real result before changing again. Days, not hours. The specific percentage matters less than the discipline of not touching it while it settles.
Frequency of change is the more common problem than size of change. An account edited daily never produces a readable result at any budget.
Creative is the actual ceiling
The constraint most budget conversations ignore entirely.
Higher spend means the same audience sees your ads more often. Fatigue arrives faster at every level of investment — so a creative library that sustained $10,000 a month will be exhausted much sooner at $30,000.
Which means your effective scaling ceiling is set by creative production capacity, not by available budget. A team producing three new concepts a month cannot feed a spend level that consumes twelve. You can raise the budget; you just won't get proportionate output from it.
The implication for planning: if you're preparing to scale, increase creative output before increasing spend, not in response to the decay that follows it. And test systematically rather than producing variations, since the point is finding genuinely different angles — the discipline in ad creative testing.
The second ceiling is the destination. More traffic to a page that converts poorly buys the same problem at greater volume, and at scale a modest conversion rate improvement is worth more than a large efficiency gain in the account — which is why post-click optimisation becomes more valuable as spend rises, not less.
Setting the budget in the first place
Before scaling, the starting question — and it runs backwards from the outcome rather than forwards from a percentage of revenue.
- What can you afford to pay for a customer? Gross margin per customer, adjusted for expected repeat purchases, with a target payback period. This is your ceiling.
- What's your current conversion rate from click to customer? That converts a CPA limit into a cost-per-click limit.
- What volume do you need? Multiply by your CPA limit for a theoretical budget.
- Can the channel actually deliver that volume at that cost? Frequently not, and discovering it in planning is cheaper than discovering it in month three.
Step one is where most budgets go wrong, because it's set as a percentage of revenue or last year's number plus ten percent. Neither of those has any relationship to what a customer is worth to you.
The allocation question — how much goes to paid at all, versus everything else — is a level above this and covered in where marketing budgets are shifting.
Three things that distort the picture
Seasonal auction inflation. Scaling into a rising-cost period means CPAs increase regardless of what you do. If you raise budget in October and CPA rises, you cannot tell how much was your scaling and how much was everyone else bidding — which is a strong argument for establishing your marginal curve outside peak season, and for the planning discipline in Black Friday preparation.
Attribution shifting under you. Platform-reported conversions and actual incremental conversions diverge, and the gap typically widens as you scale, because more of your spend reaches people who would have converted anyway. A campaign that looks efficient in-platform may be buying customers you already had.
Cross-channel effects. Scaling one channel changes the measured performance of others — more branded search from more prospecting, for instance. Evaluating a channel in isolation misreads both, which is the case for the approaches in cross-channel measurement.
A scaling routine
- Establish your CPA limit from unit economics, not from last year's average.
- Build the marginal CPA history from your existing data. Find where it crossed the limit.
- Increase spend in modest steps, one variable at a time.
- Wait for stabilisation before reading the result. Resist editing during the wait.
- Calculate marginal CPA on the increment. Under your limit — continue. Over it — stop and go horizontal.
- Check creative supply before the next increase, not after performance decays.
- Sanity-check against business outcomes — revenue, new customers — not only platform-reported conversions.
Step five is the decision point the whole routine exists to serve. Most scaling processes never reach it because nobody calculated the marginal figure, so the decision defaults to whether the average still looks acceptable — which, as the table shows, it will long after it should have stopped you.
Worth reporting the marginal number upward too. A leadership conversation about whether the next $5,000 is worth $111 per conversion is a genuine strategic decision, whereas "CPA is $68 and holding" invites the answer "then spend more" — the reporting problem examined in building a dashboard leadership will read.
If the practical obstacle is that nobody has the time to run this properly while also making the ads, that's a capacity problem — and it's where a performance marketing partner running the curve continuously earns its fee, since the difference between stopping at the right increment and two increments later is usually larger than the management cost.
The short version
Average CPA tells you how a campaign has performed; marginal CPA tells you whether to spend the next pound, and only the second is a scaling decision. A campaign can report $68 while the latest increment buys conversions at $111, because cheap early conversions keep propping up the blended figure. The formula is change in spend divided by change in conversions — two spreadsheet columns almost nobody has. Rising costs are audience depth rather than algorithmic punishment, so the fix is new audience rather than account tinkering. Scale vertically until marginal cost approaches your limit, then horizontally. And increase creative output before you increase spend, because creative supply is the real ceiling.
Spending more each month and wondering where the efficiency went?
We track the marginal cost of every increment so you stop at the right number.
Explore Performance Marketing →Frequently asked questions
Why does cost per acquisition rise when you increase ad spend?
Mostly because you are reaching further into a less qualified audience, not because the platform is penalising you. The people most likely to convert get reached first at the lowest cost, so each additional increment of budget buys progressively less responsive impressions and requires winning more expensive auctions. That is arithmetic rather than punishment. Treating a rising cost per acquisition as an algorithmic problem to be optimised away leads to endless account tinkering when the honest answer is that you have reached the edge of the efficient audience.
What is marginal CPA and why does it matter more than average?
Marginal cost per acquisition is what the most recent increment of spend actually cost, calculated as the change in spend divided by the change in conversions. Average blends everything together and therefore hides the performance of the newest money. A campaign can show an average of around 68 while the latest increment is buying conversions at over 110, which means the reported figure looks acceptable against a limit that the incremental spend has already breached. Scaling decisions should be governed by the marginal figure.
How much can you increase an ad budget at once?
The commonly repeated twenty percent rule is folklore rather than published policy, but the mechanism behind it is real. Significant changes to a campaign can return it to a learning state while the system re-optimises, during which performance is unstable and unrepresentative. The practical approach is to make changes in increments modest enough to avoid destabilising delivery, then wait long enough to read a genuine result before changing again — which usually means days rather than hours.
Is it better to raise budgets or add new campaigns?
Adding new audiences, placements, geographies or creative preserves efficiency for longer than simply raising spend on what already exists. Increasing budget on an existing campaign pushes deeper into the same audience, where returns diminish predictably. Expanding sideways opens fresh inventory that has not yet been exhausted. The limit is that each new campaign starts without accumulated learning and needs its own volume to stabilise, so this works better when you have the creative and the budget to support several lines properly.
What actually limits how far you can scale paid advertising?
Usually creative supply rather than budget. Higher spend means the same audience sees your ads more often, so fatigue arrives faster and performance decays sooner at every level of investment. A team producing a handful of new concepts a month cannot sustain the volume that heavy spending consumes, which caps effective scale regardless of how much money is available. Landing page and offer quality set a second ceiling, since more traffic to a poorly converting destination just buys the same problem at greater volume.
