How we replaced a media buyer with an AI that never sleeps
A Dubai-based e-commerce brand was managing three Meta ad accounts manually, spending 15+ hours a week on optimisation with diminishing returns. We built an AI agent that monitors performance every 15 minutes, adjusts bids autonomously, and reallocates budget in real time.
The client had three Meta ad accounts running across fashion, accessories, and homeware. A dedicated media buyer was spending 15 hours per week on manual optimisation.
Despite the effort, ROAS had plateaued at 1.8× for four consecutive months. Creative fatigue was going undetected — ads were running well past their effective lifespan, burning budget on audiences that had already converted.
We started with a two-week data audit, ingesting 18 months of historical performance data. We identified patterns that predicted ROAS degradation 48 hours in advance.
Our architecture: a Python-based monitoring agent connected directly to the Meta Ads API, running on a 15-minute polling cycle with three decision modules: bid optimisation, creative fatigue detector, and budget reallocation engine.
Week 1–2: Data ingestion and model training. Week 3: Shadow mode — predictions without execution. Week 4: Partial live deployment on one account. Week 5: Full live deployment across all three accounts. Week 6+: Continuous operation with weekly reports.
ROAS improved from 1.8× to 8.0× within 90 days. Weekly management time dropped from 15 hours to under 2 hours. The brand scaled monthly ad spend from £12,000 to £38,000 while maintaining ROAS.
BY THE NUMBERS
We were sceptical that AI could outperform an experienced media buyer. After 90 days, the data was undeniable.
Discovery calls are free and no-pitch. We'll tell you honestly if we can deliver results like this for you.