A monthly promoted-giveaway growth program for a retail partner, scaled to roughly 10x its prior budget under pre-agreed kill rules, and operated through my AI agent system with human gates on spend and anything that ships externally.
Paid growth usually scales blind: spend goes up, attribution gets fuzzier, and nobody agreed in advance on when to stop. A budget owner saying yes to 10x needs the opposite, namely certainty about what every dollar bought, stop conditions negotiated before scaling instead of argued about after, and a report that arrives monthly without being chased. The campaign had to be an instrument, not a bet.
A monthly giveaway campaign engine. Promoted YouTube creatives drive entries to a giveaway page whose entry actions generate tracked visits to the partner's site. Every creative carries its own cost-per-visit attribution, spend flows through a live budget tracker, operator hours are logged so output-per-hour is measurable, and the cycle closes with an executive report. Kill rules were agreed before the first scaled dollar was spent. The whole cycle runs through the agent system: agents draft, track, and compile; a human approves spend and sends.
The biggest counterintuitive finding came from research before scaling: for broad-reach paid promotion, hand-tuned narrow targeting backfires badly (a real head-to-head showed a custom segment performing 26x worse than the platform's broad algorithm). So the program trusts broad targeting and runs few creatives long through the learning phase instead of churning many. Second lesson: write ad copy with no dates and no prize counts, so a converting ad never goes stale between cycles. And the meta-lesson that shaped everything: agreeing the kill rules before scaling is what turns a 10x budget ask from a leap of faith into a reversible, two-way-door decision. That is why it got approved.