Raise a working Meta ad set by about 20% at a time, let it settle for a few days before the next step, and only raise at all when its cost per result is reliably below target on enough results to trust; bigger jumps risk sending it back into the learning phase, and every raise should be checked against the week before it rather than assumed to have worked.
An ad set is doing well. Leads are cheap, the client is happy, and the obvious move is to double the budget and get twice as many. It almost never works out that way. The cost per lead climbs, sometimes past where it started, and the ad set that was the best in the account becomes an ordinary one.
Two things are going on, and they point to the same answer: raise in small steps, and check each one.
Why doubling breaks a winner
First, the learning phase. Meta treats a significant budget change as a significant edit and can send the ad set back into learning, where delivery is deliberately less efficient while it explores. Meta does not publish an exact threshold. The working rule across the industry is that changes of around 20% or less are usually absorbed, and larger ones often are not. We explain the cost of learning in Every edit restarts the clock.
Second, the cheapest results are bought first. At a small budget Meta spends on the people most likely to respond. More money means reaching further down that list, to people who are a little less likely, and they cost more. Some rise in cost per result is the normal price of scale, not a sign something broke.
When a raise is earned
A raise should be earned by a result you can trust, not a good few days. A cost per lead measured over three leads is an anecdote: three more could double it. Wait until there are enough results that the cost is reliably below target, not just below it on average. Our own rule is at least eight leads, with the cost below the client’s target even at the worst end of what those eight could plausibly mean.
Judge it against that client’s own target or normal cost, not a number from someone else’s account. There is no good cost per lead in general, only a good one for this client.
How much, and how often
- Raise by about 20%. A ₹1,000 daily budget goes to ₹1,200, not ₹2,000.
- Wait a few days for it to settle before deciding on the next step.
- Repeat while the cost holds. Four steps of 20% is roughly double, reached in a couple of weeks without a restart.
- The same applies with Advantage campaign budget: change the campaign budget in the same steps.
Check the raise; do not assume it
A raise is a prediction: more money, about the same cost. Check it. From the third day after the raise, compare the cost per lead since the raise with the week before it. If leads are clearly more expensive, say 30% or more and on enough leads to be sure, put the budget back. If the raise has spent enough to expect a few leads at the old cost and none came, put it back sooner.
Going down works the same way: cut by about 20% at a time. The exception is a campaign costing double its normal or more. That is not something to spend through more slowly; stop and work out what is wrong.
Small steps are not timidity. They are how you find the budget where the ad set stops being worth more money, without paying for a restart to learn it.