From Break-Even to 4.98x: Rebuilding a Meta Account Around One Ad Set

NextUp PH runs paid creative workshops in Metro Manila, online and in person. The class list changes every month, and that turnover was quietly destroying their Meta performance. Two structural changes took the account from break-even to a 4.98x blended return.

The challenge

The account ran fifteen separate campaigns, roughly one per class. Every month, as the slate turned over, most were torn down and rebuilt.

Three problems followed:

Nothing ever finished learning. Meta's delivery system needs a run of conversions before it stops guessing. A campaign built for one class, run for a few weeks, then deleted, never gets there. The account was permanently paying exploration prices.

Campaigns competed with each other. Fifteen campaigns chasing overlapping Metro Manila audiences meant bidding against themselves.

About half lost money. Blended return sat at 1.17x: above break-even on paper, but not a business you can scale.

Underneath it was a false assumption: because the classes changed monthly, the team believed the campaigns had to change monthly too.

The approach

The idea that reframed everything: the ad set is the brain, not the campaign. Learning accumulates at ad set level. Keep the ad set alive and the learning survives, however often the creative changes.

So a new class stopped being a new campaign and became a creative swap inside a permanent ad set.

Fifteen campaigns became two. One for online classes, one for onsite. One ad set each. Same total budget.

Interest targeting came out entirely. This is the change people resist hardest. Stacking interests feels precise, but in a market the size of Metro Manila it starves the algorithm of the volume it needs. Broad geographic targeting beat manual interest selection in testing, so broad is what stayed.

Budget moved to the ad set and campaign budget optimisation was switched off. With CBO on, a performing ad set can have its budget raided by a sibling. Holding budget at ad set level protects the thing that is accumulating learning.

Daily budgets, never lifetime. Lifetime budgets front-load and run dry, starving classes late in the month.

The ad set is never paused or rebuilt. When a class fills, we pause the creative or drop the daily budget. Pausing an ad set for more than about a week resets its learning, which is exactly what the old structure did every month.

The result

Blended return went from 1.17x to 4.98x in two months, on a budget that never moved.

Confirmed paid enrolments against ad spend, same basis each month

1.17x

2.46x

4.98x

May15 campaigns

Junerebuilt: 2 ad sets

Julyplus creative refresh

Budget was held flat across all three months. The gain came from structure and creative, not from spending more.

The creative swap that did it

Because the ad sets were permanent, we could change one variable and read the result cleanly. In mid-July we replaced a set of single-class ads with one consolidated multi-class creative, inside the same ad set, on the same budget.

Indexed to the pre-swap period. Same ad set, same budget, same city.

100

↓ 61%

39

1 to 15 Julysingle-class ads · 83 enquiries

16 July to 1 Augconsolidated creative · 301 enquiries

Cost per enquiry fell 61% and volume rose 3.6x, same month, same city, same budget, same ad set. Only the creative changed.

That is the payoff from the structural work. Because the ad set had accumulated learning and was never rebuilt, a creative change produced a clean, attributable result instead of disappearing into a campaign restarting from zero.

The pattern held across the account: the consolidated creative carried the month, while the leftover single-class ads cost several times more per enquiry for a fraction of the volume.

When to add retargeting and lookalikes

The sequencing matters more than the tactics, and most accounts get it backwards.

Retargeting and lookalike audiences are usually the first thing a new account reaches for. They should be close to the last. Both are built from data, so running them before you have any means building an audience out of noise. A lookalike seeded from fifty mixed-quality conversions models the wrong person, and then you scale that mistake.

The order that worked here:

1. Fix the structure first. Permanent ad sets, broad targeting, budget held where the learning accumulates. Nothing else is measurable until this is stable.
2. Let it gather real conversion data. Roughly 50 conversions per ad set, which takes weeks rather than days.
3. Then build retargeting, once there is a genuine pool of people who have engaged rather than a handful.
4. Then look at lookalikes, seeded from the buyers you now know are real.

There is a second reason to wait, and it is the one people miss. Retargeting and lookalikes are also how you manage frequency. A tight geography saturates: the same people keep seeing the ads, frequency climbs, and every additional peso buys less attention. In this account, frequency across the local radius reached 2.23 over a five-week window, which is the point where widening matters.

At that stage you have two levers. Widen the geography to reach fresh people, and add properly-seeded lookalikes to find new audiences that resemble your actual buyers. Both need the data from step 2 to work. Reach for them on day one and you are just spending more to show the same ads to the same people.

What we would tell another business in this position

Do not confuse targeting precision with targeting quality. The instinct to add interests is strong and it usually hurts. Let the platform find buyers inside a sensible geography.

Know what your market can actually pay. These are affordably priced workshops, and that price band decides which audiences are worth reaching at all. No amount of clever targeting fixes a mismatch between offer and affordability.

Give it about three months. Meta needs roughly 50 conversions per ad set to stabilise. Most accounts that "don't work" were never allowed to finish learning.

Change creative, not structure. If your offering changes monthly, that is an argument for permanent ad sets and rotating ads, not monthly rebuilds.

Consolidated creative beats one ad per product. The single biggest in-month gain came from replacing separate per-class ads with one creative covering several. Splitting spend across many ads splits the learning with it.

Earn your retargeting and lookalikes. They are the reward for a working structure, not the shortcut to one. Build them once you have real conversion data, and use them to relieve frequency when your core geography starts to saturate.

How these numbers are counted

Return is calculated as confirmed paid enrolments in the reporting window, priced at actual class rates, measured against ad spend for the same window. Refunded and cancelled enrolments are excluded. Enquiry and cost figures come directly from the platform's day-level export.

This is a blended return: it measures total confirmed revenue against total ad spend and does not separate ad-driven enrolments from organic or referral ones. The same basis is used for every month shown, so the trend is like for like.

Absolute budget and revenue figures are the client's own and are not published here.