On this pageJump to a section
You finally have a campaign that works. Leads are coming in at a cost you like. So you do the obvious thing: you double the budget.
Two days later the cost per lead has doubled too, delivery is erratic, and the campaign that printed money is limping. You did not run out of audience. You shocked the system.
This is the most common way I see good accounts get worse. Here is how scaling really works, what the learning phase is, and how to add spend without throwing away what the algorithm already knows.
What is the learning phase, and why does it matter when you scale?
When you launch a new ad set, Meta's system does not yet know who is most likely to take your action. The learning phase is the stretch where it is finding out. Delivery is less stable, and costs usually swing more than they will later.
The 50-event rule of thumb
Meta's guidance is that an ad set typically leaves learning after about 50 optimization events following the last significant edit.1 "Optimization event" means whatever you told the ad set to optimize for: a lead, a booked call, a purchase. If you optimize for a rare event, you give the system very few chances to learn.
Why scaling can undo your best work
Every time you make a significant edit, the system treats it as a fresh start. Scale carelessly and you can pay twice: once for the money you spent to find what worked, and again for the money you spend to find it again.
What counts as a significant edit?
Meta lists pausing an ad set, changing the optimization event, changing the audience and changing the creative as significant. It adds that bid or budget changes may also count, depending on the magnitude of the change.2
What I treat as off limits on a winner
- The optimization event.
- The audience definition.
- Replacing the creative inside an ad set that is working.
- Pausing and unpausing an ad set to "reset" it.
Dara Denney keeps her creative tests in their own campaign for this reason. In a video she posted on February 21, 2024, she explained why:
you don't want to unintentionally tank the performance of an adset or a campaign that's working well
Same logic for scaling. Put the experiment next to the winner, not inside it.
VideoDara Denney
How to Test Facebook Ads Creative in 2024
Her account structure, test budgets and hypothesis template, in about seventeen minutes.
Budget is the gray area
Meta does not give a single safe percentage that I can point to in its published help pages. Plenty of buyers use steps of around 20% as a working habit. Treat that as a habit to test, not a rule from Meta. The more honest rule is this: the bigger the jump, the bigger the risk.
How do you scale a budget vertically without a reset?
Vertical scaling means putting more money behind what already works. The goal is to raise spend while keeping the system's confidence.
- Confirm the winner is stable. Look at several days of steady cost per result, not one good afternoon. If it is still swinging, it is not ready for more money.
- Raise in small steps. Increase the budget a modest amount, then leave it alone. Do not stack a budget change with a creative or audience change on the same day.
- Wait for stability before the next step. If cost per result holds, step again. If it jumps, do not keep pushing. Hold or step back.
- Use a duplicate for a big jump. If you want to double, duplicate the ad set at the higher budget and leave the original alone. You get the bigger spend without touching the proven one.
- Write down each change and the date. If performance moves, you need to know what you changed.
How do you scale horizontally, and why does it matter more now?
Horizontal scaling means adding more winning material: new creative angles, new offers, new hooks. It matters because a single ad set can only absorb so much spend before frequency climbs and costs rise.
Creative is the lever now
Meta's engineering team described its Andromeda retrieval system in December 2024. At the first step of delivery it cuts an enormous pool of possible ads down to a few thousand that might fit, and it is built to handle a growing volume of creative from Advantage+ and generative AI tools.4 The same post argues that a more varied ad mix is better for viewers and for advertisers.4
My reading, which is an opinion and not Meta's guidance: ten near-identical variants of one idea probably give the system less to work with than three genuinely different ideas. Different promise. Different proof. Different opening. When you scale horizontally, add distinct concepts, not cosmetic edits.
Advantage+ is part of the picture
In February 2025 Meta said it was testing a streamlined setup that would make Advantage+ optimization available across all campaign elements, and renamed Advantage+ shopping campaigns to Advantage+ sales campaigns.5 I covered the details in Advantage+ Sales Campaigns. Whether you use those settings or manual control, the scaling principle holds: feed the system good inputs and avoid abrupt changes.
How do you read Learning Limited?
Meta shows Learning Limited when an ad set is unlikely to get around 50 optimization events in the week after its last significant edit.3 It is not a punishment. It tells you the structure is starving the system of data.
Common causes and fixes
| What starves the ad set | What usually helps |
|---|---|
| Budget too small for the cost per result | Raise budget on fewer ad sets |
| Too many ad sets splitting the same budget | Consolidate into fewer, bigger ad sets |
| Optimizing a rare, deep event | Optimize a more frequent event, then qualify later |
| Very narrow audience | Broaden targeting and let creative do the qualifying |
One warning: moving to a more frequent event can fill the funnel with lower quality leads. Check lead quality downstream before you celebrate a lower cost per lead.
When should you not scale?
Do not scale on a cost per lead alone. If your show rate, close rate and cash collected are unproven, more spend just makes the problem more expensive. Fix the offer, the page and the follow up first, and check your capacity ceiling, because ads can buy leads faster than you can serve them.
Budget is a volume knob, not a repair tool. If the funnel leaks at a hundred dollars a day, it leaks faster at a thousand.
Want a scale plan audited before you push the budget? Start the conversation here.
Keep building.Ekai
Questions people ask
How much can I raise my Meta ads budget without resetting learning?
Meta says bid or budget changes may count as significant edits depending on their size, and does not publish a single safe percentage. The practical approach is small, spaced increases and a duplicate campaign for big jumps.
What is the Meta learning phase?
It is the period after you create or significantly edit an ad set when Meta's delivery system is still learning how to deliver. Meta's guidance is that an ad set typically needs around 50 optimization events after its last significant edit to exit learning.
What does Learning Limited mean?
It means the ad set is unlikely to get about 50 optimization events in the week after its last significant edit. It is not a penalty, but delivery tends to be less stable and costs less predictable.
What is the difference between vertical and horizontal scaling?
Vertical scaling raises the budget on what already works. Horizontal scaling adds new creative, offers or audiences so the account has more winning material to spend on.
Sources
- 1
- 2
- 3
- 4
- 5
- 6
