Why Target ROAS Campaigns Can Struggle to Scale

This article identifies four common reasons Target ROAS campaigns stall: setting overly aggressive ROAS targets that demand immediate payback, cutting budgets during the learning phase, using overly short data windows that miss downstream value, and making frequent structural adjustments before stabilization. These issues limit the algorithm’s exploration capacity and skew optimization toward short-term efficiency. To achieve sustainable scaling, the article proposes three pillars: (1) Budget size—larger budgets enable broader exploration of user segments, actions, and placements, allowing the model to identify high-value patterns.

Small or frequently adjusted budgets fragment learning. (2) ROAS targets—setting flexible targets early allows the system to bid on a larger user pool; targets should be tightened incrementally after stabilization to avoid premature inventory restrictions. (3) Data windows—sufficiently long windows let the model observe how early behaviors translate into long-term ROAS, improving predictions.

Key actionable takeaways include: avoid micromanaging the algorithm during learning, ensure stable campaign structures, and prioritize exploration capacity over strict short-term efficiency.