When CPA rises, do you raise the budget? Not until you know what moved.

I am Samet Durgun, a fractional Head of UA. I run paid UA for subscription apps and mobile games and write up what I find in the accounts I manage. This piece sits under paid UA for subscription apps; more about me.

I do not raise a budget to bring a rising CPA down. CPA is the cost per acquisition. When it climbs on Meta, I first check whether the number is counted right, and then what moved. I raise in two cases. One is when cost per paying customer is under my CPA ceiling, the most I can afford to pay, on cohorts of the same age (customers who started equally long ago). Two is when the campaign gets so few purchases or trials that its cost is noise. In that case the raise is a test with a stop rule. Otherwise I hold, fix what moved, cut or pause.

I am not afraid of the budget edit, and I follow no fixed percentage for each raise on Meta. I also do not believe more budget fixes a weak campaign. On one campaign that was starved of events I proposed a higher budget. Other things changed in the same weeks, and the results did not get cheaper.

I run paid user acquisition for subscription apps and mobile games, mostly on Meta. The question behind this piece was asked in public by an advertiser spending about $50 a day on sales ads, with CPA moving from about $25 toward $50. I have run tests that small on apps and games, never for a web shop. By my own sizing rule that budget is too small to read, and the last section does the sum.

Does raising the budget bring a rising CPA down?

Not in my experience. There is one exception, and it is a test.

I have written both sides of this. In a LinkedIn post in March 2024 I wrote that CPA will likely rise in a roughly straight line as daily spend goes up, and faster on big jumps. The chart used hypothetical data. I have also argued the other way. Costs settle down with volume, and teams stop before the campaign has enough events. With 50 or more events a week and positive economics, I see no issue raising the budget.

My read is that both are true, at different volumes. Below my event floor, more volume can bring cost down. The floor is the fewest events a day I need before I trust a campaign to learn. Above it, more spend pushes cost per result up. Neither fixes a campaign that was weak before. And neither statement covers a campaign that was converting and then got more expensive. So the checks come first.

Meta’s help pages support one part of this: the cheaper results are bought first. In its answer on why campaign prices increase over time, it says the delivery system seeks the highest volume opportunities first, and it calls those the lower cost ones. When they run out, the system may move on to more expensive options, depending on your bid strategy. That page is about prices over the life of a campaign, not about a budget raise. Meta’s developer documentation on bid strategies goes one step further. For the highest volume strategy it says costs can rise as the cheapest opportunities run out or as you increase the budget. My own picture is that the last customers cost the most.

From my app work, I expect a healthy campaign not to get much more expensive between about €100 and €1,000 a day. I have not measured that. In 2024 I warned of steep climbs on big raises. Today I would say returns fall as spend rises, but not off a cliff. I have no published account data for either view.

My own accounts do not agree with each other. One account that ran install campaigns gave me 125 days of daily data, and cost per install rose with daily spend. The link was moderate, a correlation of about 0.48. A subscription app went the other way. Cost per trial fell by about a fifth while daily spend about doubled. That campaign already had enough events before the raise, and I did not yet know how many trials would pay, so it proved less than it looked.

I have also reversed myself. On one account I gave three different budget instructions in three weeks. First, after a cost rise, I cut the budget and blamed tired ads and wasted segments. Then, 11 days later, I said I saw no positive link between daily spend and CPM (the cost per 1,000 impressions), and that the budget could go up. Another nine days later, with more data, I said daily spend was the biggest driver of install cost and return, and I told the team to go back down and step up again later.

If you raised the budget and CPA went up, work out what the added money bought.

Illustration, not an account:

Week A Week B
Spend $3,000 $6,000
Purchases 75 125
Average CPA $40.00 $48.00
What the extra $3,000 bought 50 purchases at $60.00 each

The dashboard shows $48. The decision is about the $60. Plot daily spend against cost per result on your own account before you believe any of my stories.

What do I check before I touch the budget?

Whether the number is real, then what changed. I have not always kept to that order myself.

First I ask whether the old low number was ever real. A cost per Subscribe in Ads Manager can mix new payers with renewals, so it can read below what a new payer really costs. If Meta’s count and your billing tool disagree, my post on trials missing from Meta also covers a count that is too high. I have seen a low starting cost that was never a real paid cost, because free installs were diluting it. For a shop, the same check covers test orders, repeat buyers and a changed attribution setting. I have never run ads for a web shop, so that list and the website inputs below are my app checks adapted. I do not run them day to day.

Then I write down every change from the last two to three weeks. That list is my change log. With it I find which input moved. On a website that means CPM, click rate, conversion rate on the site and order value. I also check whether cost is up everywhere at once. A rise in every country or campaign at the same time points to seasonality or tired ads more than to one campaign.

More than once I have opened the placement breakdown and found one placement taking spend with no purchases. That placement has not always explained the whole rise. Ads Manager has also recommended a higher budget in an account I ran. I noted it and still ran the same checks.

Sometimes the checks rule the budget out. On one account, install cost had spiked. I changed the optimization event and lowered the spend at the same time, and the cost fell to about half. Before I credited the lower spend, I checked. The new campaign had spent as much as the old one on some days, so my read was that the budget was not the cause. What I could not say was whether the old campaign would have recovered if I had left it alone.

The full order is in the CPI and ROAS diagnostic.

Raise, hold, cut or pause the budget: which one when?

I decide by what the account shows, on weekly numbers.

What I see What I do Why
Too few events to read, cost swinging, nothing in the change log Hold. Combine ad sets or campaigns first. Raise only as a test, with the stop rule written before the raise The number is noise until volume is there. On one starved campaign this test failed for me, with other changes in the same weeks
Cost up, still under the CPA ceiling at the same cohort age Hold and read again next week I watch it at steady spend until the next read
Cost up in every country or campaign at once Check seasonality and tired ads. Hold, or drop to a maintenance budget and test new ads A rise everywhere at once is rarely one campaign’s fault
Cost up, and the change log explains it (release, tracking, placement, new ads) Fix that. Leave the budget One change at a time, or nobody can say what worked
Cost up right after a raise Do not judge it the next day. Give it a few days, a full week if the agreed loss limit allows Results swing after an edit, and a solid campaign settles
Cost still up once the raise has settled, and the added customers cost more than the ceiling Go back to the last level where cost was affordable, bring in new ads, step up again later. To grow, I would rather add a platform, country or network than push more money into the same audience The extra spend bought the most expensive customers. The cut also buys fewer events
Cohorts of the same age miss the ceiling on data reconciled with billing, nothing in the change log explains it Cut, or pause if the cut would starve the campaign of events. Switch off what is weak before trimming everything A halved budget that cannot feed the optimization is worse than a pause
Under the ceiling at the same cohort age, events above the floor Raise. My published rule waits for a second good week. I have raised sooner Cost can still rise after a raise

A stop rule is one line written before the raise: the day I review, and the cost per result above which the budget goes back down.

After a raise I have waited anywhere from two days to a week, depending on how fast events come in and how much loss the team and I agreed on beforehand. When CPA rises on a campaign that was converting, I judge it by the week. Meta’s page on the cost per result goal says the same for that bid type: evaluate on weekly averages.

A cut is not free. A lower budget can make every ratio look better while it buys fewer customers and fewer events, and the business has to choose where it wants to sit on that tradeoff. On one account my call before Christmas was to drop to a maintenance budget and use those weeks to test new ads, because the season was pushing costs up for everyone. On a fitness app the call was a full pause. The diagnostic lists what a cut does not fix.

When do I raise the budget, and how did those raises go?

In two cases, and the second one is a test.

The first case I have already published: I raise the budget only while cost per paying subscriber, at the same cohort age, stays under the ceiling. I also wrote that I wait for a second good week, and I have raised sooner than that. With a ROAS (return on ad spend) target in place I worry less. I raise the budget on what performs and tighten the target on what lags.

The second case is a campaign starved of events. When Meta marks an ad set learning limited, short of the events it needs to learn, its help page lists a higher budget as one fix. It stands next to combining ad sets and campaigns, a wider audience and an optimization event that happens more often. When the budget is fixed, I combine first, because the budget stays the same. Before adding money I reach for two other moves: a more frequent event, as long as it goes together with paying later, and the same money in half the time. More money comes last.

My event floor is softer than it sounds. Meta’s learning phase page puts the usual exit from learning at about 50 results a week. The same page says an ad set leaves learning as soon as its delivery is stable, so 50 is the usual figure, and Meta does not present it as a hard line. I count per day on purpose, because a week can hide dead days. Meta’s guidance works out to about 7 a day. The least I plan for is 5, and in a talk to subscription app teams I asked for 10 a day in every ad set. The whole floor is a hypothesis I act on, and I have called it a huge one myself. I base it on the apps I have run.

On one subscription app the campaign got too few purchases a week to learn from. I proposed spending two weeks of budget in one week. We doubled it, and the ads and the targeting changed around the same time, so it was not a clean test. It came nowhere near 50 events and the results did not get cheaper. My read was that the bottleneck was the share of installs that go on to pay. I no longer thought it was the event count. That is still a read.

This is how I put the choice to another team with a starved campaign: stop it now because it is expensive, or triple the budget for two days and see whether purchases really come. I cannot show you a clean win from a test of this kind. I have also not always written the stop rule first.

People often ask why they should not spend 3X if spending X works. A bid cap or a cost per result goal, which I call a cost cap, is one way to try 3X without betting the budget. The catch is that it may not spend. Meta’s developer documentation says a cost cap may not spend the full budget once it hits the cap, and on TikTok I have started a campaign on a cost target and watched it spend nothing. When I proposed a bid cap to answer the 3X question, I gave it a 50% chance of working. In my one published bid cap run the cap went quiet and never recovered. I have proposed a cost cap too. Still, I tell subscription apps with several price points to skip a cost cap, because where customers differ in value it tends to find the cheapest ones. With a single yearly package I think it is fine. Meta says its cost per result goal works best with at least 50 to 100 conversions a week.

How much can I raise at once, and will the edit break the campaign?

Meta publishes no percentage and neither do I.

Meta puts the budget amount on a list of edits that “may or may not be significant, depending on the magnitude of the change” (Meta Business Help Center). A significant edit is one that sends the ad set back into learning. Its example: $100 to $101 is unlikely to send ad sets back into learning, and $100 to $1,000 may. Meta also advises against frequent budget changes. For campaigns on a cost per result goal that are meeting expectations, it says to raise the budget gradually, which is the opposite of my habit.

I wrote that habit down in the Meta playbook: “My answer is that Meta can take it. I double budgets comfortably.” The lines after it add one condition, that the campaign is solid. A doubled budget on a campaign that is only days old and still learning may not spend in full, and I expect that.

I do not believe a budget must go up only about 10% at a time. I said so about one account with few events a week, so read it with that volume in mind. I have not seen a raise break a campaign for good. I think a small dip often follows a raise, and then the campaign settles again.

Three things are true here, and how they fit is my reading. The fear of the edit is overdone on a solid campaign. You cannot compare the days before an edit with the days after it until the campaign has settled. Waiting does not fix a campaign that was weak before the edit.

I have no single step size. I have doubled in one step and proposed tripling for two days. What I have written down is smaller. Once it was a step such as 10%. I called it a test step and wrote that no platform guarantees it. In my own notes it was 20% to 30% once every two days when things work really well. The 20% I have proposed was for ROAS targets, raised every two to three days.

On one account the budget had been raised many times in a short stretch, and results seemed to drop after each raise. The data could not tell whether the raises caused the drops or whether their effect was small and passing. There I said I would not change the budget more than once a week. That is not a weekly rule of mine. My own practice has been faster when a test called for it.

I prefer a campaign budget with ad set minimums, as long as the campaign has several ad sets to choose between. In the playbook I wrote that my preferred way to scale is to duplicate the best campaign and raise the copy’s budget. My read is that below the event floor a copy splits a signal that is already thin, which is my own fewer campaigns argument. On one of my accounts the share of installs that become payers differs by weekday, so compare the same weekdays before and after a change.

What if the budget is too small to read at all?

Then a small raise will not make it readable, and I would say that first.

Take the case behind this piece, with its approximate numbers. $50 a day is $350 a week. That buys about 14 sales at $25 and seven at $50, so 1 to 2 a day. A few sales either way is the whole difference. So I would not call one week at $50 a trend. Before any budget question, work out what a sale leaves you after product cost, fees and refunds. Until you know that, $25 or $50 is neither good nor bad. My method for apps is here.

My sizing rule, from my app work, is the cost of the event times the events I need per day. At 5 events a day, a cost of $25 per sale means $125 a day, and $50 per sale means $250. At 10 a day it is $250 and $500. So about $50 a day is between one tenth and two fifths of my floor, and $55 or $60 changes nothing. The plain sum points the same way. If the cost stays at $50, $55 a day buys about 1.1 sales a day. Getting back to 2 a day at that cost takes about $100 a day, and that assumes the added sales cost no more than the first ones did.

On one Meta account, an ad with fewer than 20 purchases in a week told me little about its next week. That was ROAS per ad, and the question here is CPA per campaign, so take only the direction. My post on creative testing does the same kind of sum per ad, to find how many ads a budget can judge in a week.

I have not seen this account. So I list options and recommend none.

  • Fund a short test at the floor, with the stop rule written first.
  • Put the same money into fewer days, which I prefer in general to spreading the money thin.
  • Keep one campaign and one ad set and read it over several weeks.
  • If the sum cannot work at any budget, go back to the product.

Sources and scope

I opened each Meta page in the list below on 14 November 2026 and checked every line I paraphrase against it. None of them shows a date, so open them again before you lean on one. The Week A and Week B table and the sums on the $50 a day case are illustrations, and the arithmetic is mine. The account observations are anonymous. They come from live accounts, were never set up as experiments, and several had other changes in the same week. Some of them describe the same account more than once, so there are fewer accounts than stories. My practice is in subscription apps and mobile games. The event floors and waiting times are my planning numbers. The 50 a week is Meta’s.

Questions people ask

Does increasing the budget lower CPA on Meta?

Not as a fix for a cost that rose. The exception is a campaign with too few events to learn from, where a raise is a test. On one such campaign that test failed for me, though it was not a clean one. With enough volume, more budget buys more of the same audience and the last customers cost the most. Meta's developer documentation says costs can rise with budget on its highest volume strategy. The rest is my read from apps and games. My own accounts do not all agree, and I have no published account data for it.

Why did my CPA go up after I raised the budget?

Results swing for a few days after an edit, and the added spend reaches more expensive customers. I do not judge the raise the next day. Once it has settled, I work out what the added budget bought and compare that cost with the most I can afford.

How much can I raise a Meta budget at once, and does it restart learning?

Meta publishes no percentage. Its help page says a budget change may or may not send an ad set back into learning, depending on the size of the change. I have doubled budgets in one step on solid campaigns. I have also once proposed a smaller step, 10%, and called it a test step. I have no single number.

When should I cut the budget instead of raising it?

I cut when cohorts of the same age miss the CPA ceiling, the most a customer may cost me, and nothing in my list of recent changes explains the miss. I check that on numbers reconciled between the ad platform and billing. If the cut would starve the campaign of events, I pause. A cut also buys fewer customers, so I switch off what is weak before I trim everything.

I get fewer than 50 conversions a week. Should I raise the budget to get out of learning?

Not as the first step. Meta lists a higher budget as one fix among several. I combine ad sets first, because the budget stays the same. If I raise, it is a short test, and before I start I write down the cost at which the budget goes back down. That test has failed for me, and my read was that the event count was not the real bottleneck.