Last month a client called me genuinely upset. They'd been running a supplement ad for about three weeks, the thing was pulling north of 4x ROAS, daily spend around $180. So they did what I probably would've done before I learned how to properly scale Meta ads — they went into the account on a Tuesday night and moved the budget to $600.
Woke up Wednesday to a 1.4x. They cut the budget back, but they were already tanking the campaign. The ad never recovered. They paused it. We had to start from scratch.
This happens more than I'd like to admit, and honestly, the logic behind doing it makes total sense. You've got something working. Of course you want to pour fuel on it. The problem is that Meta's algorithm doesn't see "more fuel" — it sees a new campaign, basically. And it reacts accordingly.
1. The budget jump that kills campaigns
There's a threshold — not an official one Meta will tell you about, but one you'll feel — somewhere around a 20-30% daily budget increase where things start to wobble. Go past 50% in a single move and you're almost certainly triggering a full re-learning cycle. The campaign that was printing money is now burning through your daily cap trying to relearn who to show your ad to.
What actually works: duplicate it. Run the new one broad. Leave the original alone entirely. I know that feels weird, like you're splitting your own attention, but the original stays stable and the duplicate gets a fresh shot at audiences the algorithm hadn't found yet. Nine times out of ten the duplicate starts performing within 48 hours and you've effectively scaled without touching anything that was working.
It's boring. It works.
2. Ad fatigue hits faster than you expect at scale
At $200/day in a niche market, your frequency stays manageable. At $800/day, you can blow through your warm audience inside a week. I've watched accounts where the creative hadn't changed in six weeks, ROAS was eroding every day, and the client kept blaming the algorithm. We swapped in two new hooks — same exact video underneath, just different first three seconds — and the numbers came back the next day. Not gradually. The next day.
The hook is almost everything. Most people are watching Reels with the sound off, deciding in under two seconds whether to stop scrolling. If your opening frame is a logo or a product sitting on a table, you've already lost them. What stops the scroll is a face, a problem statement, something slightly unexpected. The rest of the video can do the selling — the hook just has to earn you those extra seconds.
A few things we keep coming back to:
- Shoot five hooks, keep the body: Takes an afternoon. Gets you essentially five new ads. The production cost is near zero and the testing data you get back is genuinely useful — you start to learn what your audience responds to, which makes the next round even better.
- Don't overthink the production quality: We have a client who shoots everything on an iPhone in her kitchen. Converts better than the studio content we helped them make last year. There's something about rough-around-the-edges UGC that people trust more, especially on Reels. It doesn't look like an ad.
- Dynamic Creative for new accounts: If you're not sure what's going to land, throw your variants into a DCO test and let the platform figure it out. You'll usually have a clear winner within $200-300 of spend, and you didn't have to guess.
3. The audience settings you keep fiddling with are probably hurting you
I spent a long time building detailed audiences. Layering interests, adding behavioral qualifiers, excluding people who'd already purchased. It felt like I was being precise. What I was actually doing was starving the algorithm of data and making its job harder.
Meta's targeting has changed a lot in the last few years. The signals they're working with now — purchase behavior, app activity, cross-site data — are genuinely sophisticated in ways that interest targeting from a dropdown menu just isn't. When you tell it "women, 28-45, interested in yoga, organic food, and sustainable fashion," you're filtering out a huge chunk of people who would absolutely buy your product but don't fit that tidy profile.
Go broad. Seriously. Some of our best-performing setups right now are just age range, location, and a strong creative. The video finds the buyer. That's not a platitude — it's mechanically how it works. The creative acts as a targeting signal, and the algorithm uses engagement patterns to identify who else looks like the people responding to it. The more you constrain it up front, the less it can do that job.
That said — broad doesn't mean lazy. The creative still has to be good. The offer still has to be clear. You're not outsourcing the thinking to Meta, you're just not adding friction that doesn't need to be there.