Meta purchase optimization: why winning ads break when you scale
Every media buyer has seen it. An ad runs at a great CPA for two weeks, the budget goes up, and within days the results fall apart. It isn’t bad luck and it usually isn’t the ad. It is how purchase optimization works. Once you see what the system is doing with your budget, the break makes sense, and so does the fix.
What purchase optimization does
When a campaign is set to optimize for purchases, you are not buying an audience. You are asking Meta to find the people most likely to buy, among everyone your targeting allows. Every time someone opens the app there is an auction for the ad slots in their feed, and Meta decides which ad wins using three inputs.
How Meta decides which ad a person sees
The auction runs for every single impression.
What you’re willing to pay, set by your budget and bid strategy.
How likely this person is to buy after seeing this ad.
Feedback on the ad: hides, complaints, engagement.
The middle part does the heavy lifting. Meta estimates, for each person, how likely they are to buy after seeing your ad, based on their behavior and on who has bought from you before. That is why purchase data matters so much: every sale teaches the system who your buyer is.
The list your budget works through
Think of it as a ranked list of people, from very likely to buy to very unlikely. A small daily budget only needs a handful of sales, so Meta spends it on the top of the list. Those people are cheap to convert, which is why a new winner often looks amazing at low spend.
Who your budget reaches
More money has to go further down the list.
Double the budget and the top of the list isn’t twice as big. The system has to show the ad to people who are less likely to buy. They need more impressions, more reminders and a stronger reason before they order. So each extra sale costs more than the last one.
Average CPA vs. the cost of the next sale
This is where most scaling decisions go wrong. The dashboard shows the average cost per purchase across all sales. But what matters when you add budget is the cost of the extra sales that budget buys.
Why the dashboard hides the problem
The average rises slowly; the next sale gets expensive fast.
A brand can double spend, see the average CPA go from €30 to €38 and call it acceptable, while the extra sales it bought cost about €52 each, above break-even. Our e-commerce KPIs guide shows how to work out the break-even CPA that line should be compared to.
The learning phase
Meta says an ad set needs about 50 optimization events in 7 days (for a purchase campaign, 50 purchases) to leave the learning phase and deliver steadily. Until then, costs swing more. And significant edits, which Meta says include big budget changes, new creative and new targeting, send the ad set back into learning.
Two practical consequences. First, an ad set that gets 10 purchases a week won’t stabilize, whatever you change; consolidate into fewer ad sets instead. Second, tripling a budget overnight buys less likely customers and can also restart learning on the ad set that was working.
Six reasons winning ads break when scaled vertically
“Scaling vertically” means raising the budget on what already works. It is the first thing everyone tries, and it breaks for a combination of these reasons:
What goes wrong
Most accounts hit several of these at once.
More budget means reaching people further down the list, who need more ads to buy.
The best pocket of buyers sees the same ad again and again. Results fade, costs rise.
A big budget jump counts as a significant edit, so delivery re-learns and wobbles.
Every angle speaks to a limited group. Double the budget, same group, same message.
New, colder visitors need more convincing than the warm ones the page was built for.
Purchases report with a delay. Cutting after one bad day kills ads that were fine.
Notice that only one of the six is about Meta’s settings. The rest are about the message, the audience it can reach and the page behind it. That is why the fix is rarely a clever budget trick.
How to scale without losing the CPA
- Raise budgets in steps, not jumps. A common rule of thumb is around 20–30% every few days, then watch the cost of the extra sales before the next step. It is a practitioner guideline, not a Meta rule; the point is to avoid big edits that restart learning.
- Judge on the cost of the next sale. Compare what the extra spend brought in with what it cost, over at least a few days, because purchases report late.
- Protect the CPA with a cap when volume allows. Cost caps and bid caps tell Meta not to buy sales above a price. They limit spend when the cheap buyers run out, instead of letting the CPA drift.
- Scale horizontally at the same time. New angles and avatars add fresh tops of the list for the budget to reach. That is the real long-term fix, and we compare the two approaches in vertical vs. horizontal scaling.
- Make colder traffic convert. The people further down the list need more convincing: an advertorial or listicle before the product page, stronger proof, a clearer offer. See direct response funnels.
- Raise what each customer is worth. A higher order value or a second purchase lifts your break-even CPA, so you can afford the more expensive buyers. Our AOV offer library has 22 options.
The numbers to watch while scaling
| Metric | What it tells you |
|---|---|
| Cost of the extra sales | Whether the last budget increase paid off |
| Frequency (7 days) | Whether the same people are seeing the ad too often |
| CPM | Whether reaching people got more expensive |
| Click-through rate | Whether the message still lands with the new people |
| Conversion rate on the page | Whether colder visitors still buy |
| Blended MER | Whether total revenue grew in line with total spend, not only what Meta reports |
The short version
- Purchase optimization ranks people by how likely they are to buy; small budgets reach the most likely buyers first.
- More budget means reaching less likely buyers, so each extra sale costs more than the average suggests.
- Meta needs about 50 purchases per ad set per week to stabilize, and big edits such as large budget changes restart learning.
- Winning ads break when scaled because of less likely buyers, rising frequency, learning resets, saturated angles, weak pages and judging too fast.
- Scale budgets in steps, judge on the cost of the extra sales, and add new angles and a stronger funnel at the same time.
FAQ
Why do Facebook ads stop working when I increase the budget?
More budget forces Meta to reach people who are less likely to buy, frequency rises for the best buyers, and a big budget jump can restart the learning phase. The ad’s average cost per purchase rises slowly while the cost of the extra sales rises much faster.
How does Meta’s purchase optimization work?
Meta runs an auction for every impression and ranks ads by bid times estimated action rate plus ad quality. With purchase optimization, the estimated action rate is how likely that specific person is to buy after seeing that ad, learned from past purchase data.
How many conversions does Meta need to exit the learning phase?
Meta says about 50 optimization events within 7 days per ad set. Significant edits, including large budget changes, new creative or new targeting, send the ad set back into learning.
How fast can I increase a Meta ads budget?
A common rule of thumb is about 20-30% every few days, watching the cost of the extra sales before each step. It is a practitioner guideline rather than a Meta rule; the goal is to avoid large edits that reset learning.
What is the difference between vertical and horizontal scaling on Meta?
Vertical scaling raises the budget on ads and ad sets that already work. Horizontal scaling adds new angles, avatars, audiences or ad sets so the budget has new pockets of likely buyers to reach.