Why scaling your Meta budget 25% a day can wreck your performance

We ran the test on a jewellery brand. A DPA campaign that was performing well, ROAS above 4 the week before, so we decided to push. +25% budget per day. Start of the week, all good. Then it collapsed. The DPA's ROAS dropped to 2.66, then 1.16. The overall COS went from 3.68 on a rebound to 1.7 for the week. CAC shot up hard.

This is what happens when you scale too fast. I'll explain why, with the numbers, and most importantly how to avoid making the same mistake.

What breaks when you scale too fast

Meta's algo needs stability to deliver well. When you throw +25% a day at a campaign, it sees a big budget increase and immediately goes hunting for volume. It targets tons of people who aren't necessarily in your audience, just to spend the budget you're asking it to spend. The result: conversion rate drops, and CAC explodes.

That's exactly what we saw on the DPA. We should have scaled up gradually to let the algo rebalance. Instead we turned everything upside down at once. And Meta isn't symmetrical. If you overspend badly for a week, fixing that week of overspend takes you at least a week, sometimes more. You pay for a long time for a mistake you made fast.

Scale in steps. You spend more, but with guardrails, and you watch what it does before you push again. Not +25% every day without thinking.

The ad set breakdown tells a different story

The campaign was structured in ABO, not CBO. So 4 ad sets, 4 audiences, each with its own budget:

• all products
• a recently launched hero product
• the best-sellers
• the low-CAC products

When you look at the campaign's overall performance, you just see it falling apart. But when you break it down by ad set, the truth is a bit different. Only one ad set is actually holding, with a CAC around €30 and a ROAS of 2. The other three are way down compared to before.

Another counterintuitive point. The best-seller ad set has the best CAC, but not the best ROAS, because best-sellers have a low average order value. In other words, you pay little to acquire, but each customer brings in little. Lots of volume, but low revenue.

If you only look at CAC, you'd think this ad set is your best one. But on ROAS alone, you'd miss the volume of new customers it brings in. CAC and ROAS tell you two different things, and neither one alone is enough to make a decision. Hence my rule: never pilot on a single number. On the same logic, if you want to stop lying to yourself about your real ROAS, you need to seal off your acquisition and repeat budgets.

Wait or cut when it collapses

When it collapses, two instincts clash. Either you wait, because a budget increase naturally drops performance in the short term, while the algo rebalances. You tell yourself, I'll let it run a few days, it'll bounce back. Or you cut. You reduce budgets on the ad sets that burned through without converting, and you re-split between ad sets.

That's what we did. We lowered the audiences that were spending without bringing in sales, and we reconcentrated where things were holding.

There's no automatic answer. It depends on how long you've been overspending, how bad the damage is, and your ability to financially survive a week of degraded performance while waiting for it to come back. But my bias, honestly, is that when an ad set has burned through several days without converting, you shouldn't hope it'll fix itself. You cut and you reallocate.

Your day-by-day reporting is lying to you (especially on weekends)

A methodology point that changes everything on short cycles. When you scale a budget fast and look at your spend, your CPA and your ROAS day by day, you're reading it wrong.

Weekends completely distort your reading. Because of the attribution window, you get conversions that come in later, sometimes two or three days afterward. So on Saturday you see a catastrophic ROAS, you panic, you cut, and in reality Saturday's sales will log on Monday or Tuesday. You made a decision on incomplete data.

The dead-simple tip: filter Monday-Friday when you analyse a scale-up. You smooth out the weekend noise and look at a cleaner window. On a very short cycle, raw day-by-day reporting is almost useless. And this ties into a bigger topic, because if part of your conversions happen offline or delayed, it's your whole Meta attribution to your sales you need to review before drawing conclusions.

What this test really taught me: don't bet everything on one campaign

This test taught us something that goes beyond the DPA. The DPA was unstable. As soon as we pushed hard, it fell off. Meanwhile, the Traffic and acquisition campaigns were more stable, even if they looked like they were performing worse.

So you need to pilot at the global level, not campaign by campaign. A low-CAC campaign, like the one on best-sellers, brings in volume and new customers. Its ROAS is average, but it feeds the account. And you can balance your overall ROAS with other campaigns pushing more expensive products with higher order values.

If you put all your budget on the campaign with the best displayed ROAS, the day it becomes unstable, you've got nothing left to cushion the blow. You want a mix. Campaigns that bring cheap volume, campaigns that push margin, and stability so you're not at the mercy of a single lever.

I also always look at ad budget based on margin, not revenue. A ROAS of 2 on a high-margin product isn't worth a ROAS of 2 on a discounted product. If you pilot only on the ROAS Meta shows you, you can scale the hell out of a campaign that's losing you money.

How I'd scale today

If I had to redo the budget scale-up on this account, here's what I'd do:

• Scale in reasonable steps, not +25% a day. You increase, you let it breathe two or three days, you check, you push again if it holds.
• Filter Monday-Friday to read performance during the scale-up. Never a decision made on a single weekend.
• Look at the ad set breakdown, not just the campaign. CAC and ROAS together, never one without the other.
• Cut fast the ad sets that burn through without converting, and reallocate to what's holding.
• Don't bet everything on the campaign with the best ROAS. Keep a mix of volume + margin so you don't depend on a single unstable lever.

The most concrete lesson from this test: the speed at which you scale matters as much as the amount you put in. You can spend much more on Meta, but gradually, with guardrails, and reading the right numbers. +25% a day on a campaign that's working is the best way to break it and spend the following week repairing it.