Why does scaling break ROAS?
Meta's and Google's delivery algorithms calibrate to your current budget and targets. When you raise the budget sharply in one move (for example, doubling it), you fundamentally change the delivery conditions the algorithm learned, and the system has to rebuild its prediction models. This is the learning phase reset. During this period CPAs are typically 20-50% higher and ROAS swings until it settles. In 2026 Meta is stricter about these boundaries than before.
Vertical or horizontal scaling?
There are two ways to scale, and using both together is usually the most robust.
- Vertical scaling: raising the budget of the same campaign. It is fast but risks the learning phase, so stay within the 20% limit.
- Horizontal scaling: replicating the winning setup to new audiences, placements, geos, or platforms. Slower but more stable; it does not disturb a single campaign's learning.
- Creative scaling: adding new creative variations. It delays ad fatigue, which is often the invisible ceiling on scaling.
The 20% rule and the learning phase
The practical rule is simple: do not raise a campaign's budget by more than 20% in a single step, and wait 3-4 days after each increase. Starting at 100 units per day and raising 20% every 72 hours puts you at about 170 units per day after one week. To exit the learning phase, Meta wants roughly 50 optimization events in a 7-day rolling window; budget, creative, targeting, and bid changes all reset that counter.
When should you scale? 5 signals
Scaling is a timing decision. Do not touch the budget until these five signals line up:
- Profitable, stable ROAS: above your target and consistent for at least 1-2 weeks.
- Out of the learning phase: the campaign is in active delivery, not 'learning limited'.
- Fresh creative pipeline: the next 2-3 variations are ready; scale fatigues creative faster.
- Enough conversion volume: weekly conversions clear the learning threshold comfortably.
- Budget headroom: the current budget is not already capped or impression-share limited.
Stop chasing scaling signals one by one
Ads Sensor reads your Meta, Google, and GA4 data and tells you which campaign is ready to scale and where the learning phase has reset, with the reasoning.
Scaling on Google: set tROAS by economics
On Google, the 2026 preferred approach for accounts with enough conversion data is value-based bidding (tROAS). The most common mistake is setting the target from historical averages: saying '4x tROAS was the past norm' and expecting the algorithm to hold it while discovering new volume. Instead, calculate tROAS from your margin and acceptable customer acquisition cost. After a budget increase, wait 1-2 conversion cycles before judging; a few strongly-signaled campaigns scale better in 2026 than many fragmented ones.
What to watch while scaling
During scaling, ROAS alone is misleading; read several metrics together:
- Incremental return: is the extra budget really bringing new sales, or buying traffic you would have won anyway?
- Frequency and CPM: if they rise, you are exhausting the audience; switch to horizontal scaling.
- Learning status: a 'learning limited' warning says the scaling is too aggressive.
- Cross-platform mix: do not sink one platform's efficiency while scaling another; look at blended ROAS and MER.
Tracking this by hand quickly becomes impossible across multiple platforms and accounts. Ads Sensor unifies your Meta, Google Ads, and GA4 data in one panel; during scaling it automatically catches learning resets, rising frequency, and falling incremental return, and tells you what to do, with the reasoning.