Case Study · Meta Business Blog · 3/28/2026

Performance Marketing Team Improves ROAS by 41% With an AI Optimization Agent

效果营销团队借助 AI 优化 Agent 将广告投资回报率提升 41%

# marketing⚡ data-analysis⚡ automation⚡ decision-supportGPT-4🔴 Dev needed
Why it matters
Ad platforms optimize within campaigns. Humans optimize across campaigns. Agents can do both, continuously, at a speed no human team can match. The 41% ROAS lift came entirely from timing — acting on signals within hours instead of days.

The Problem

A DTC brand spending $800K/month on paid social and search was reviewing campaign performance weekly. The problem: ad performance can degrade within hours (creative fatigue, audience saturation, competitor activity) and peak windows can be captured only if budget is shifted immediately. Weekly reviews meant leaving money on the table and burning budget on degrading campaigns for days.

The Agent Solution

They built a performance optimization agent connected to Meta, Google, and TikTok ad APIs. The agent runs every 2 hours and:

  1. Checks ROAS, CPA, CTR, and frequency for every active ad set
  2. Identifies statistically significant performance changes (not noise)
  3. Pauses ad sets where ROAS has dropped below target for 3 consecutive checks
  4. Scales budget on ad sets with ROAS 20% above target
  5. Flags creative fatigue (frequency > 4) for human creative refresh
  6. Sends a daily summary to the marketing team with decisions made and rationale

Hard budget limits and approval thresholds are set by humans — the agent can't exceed them.

Results

The Guardrails Design

The agent has three permission levels: (1) pause ad sets under $500/day spend — autonomous, (2) reallocate budget within campaign — autonomous up to 20% change, (3) pause campaigns or make changes over $2,000/day — requires human approval within 2 hours before executing. This tiered system captured 94% of the value while keeping humans in the loop for significant decisions.

Related Cases