For business reasons, the company ended its contract with the agency that produced every e-commerce asset for Amazon, Walmart, TikTok and Instagram. The transition was planned to take six months. By my second day, roughly 90% of the agency operation had already stopped. Fourteen accounts and around sixty brands were live and selling, waiting on MPIs, carousels, A+ and A+ Premium content and brand video. Every one of those assets sits inside a listing or an ad set that is already spending, so a gap in supply is not a missed deadline, it is spend running on stale creative.

I was hired as Creative Director to build the internal function from zero, and suddenly the six-month runway was gone. Before hiring anyone I set the standard the team would be judged on: not how much we shipped, but what the work returned, read on ROAS, conversion and customer acquisition cost. That decision shaped everything downstream. I activated my professional network, opened accelerated hiring through LinkedIn, and structured a cross-functional five-person team, a Creative Director, an Art Director and three designers including an e-commerce specialist, to replace the twelve people the agency had assigned.

The build ran on a creative operating model of my own: human work and AI together, roughly half and half, Figma plus AI image generation inside a cascade flow, designed for traceability, brand understanding, performance and repeatability. A person approved brand and final quality before anything reached a client. That gate cost throughput and I never moved it, because sixty brands share one pipeline and a single tone breach travels further than any speed gain pays for. Every asset carried a job in the funnel and was read against the previous cycle, so winning formats went back into the template library as the new baseline and losing ones were logged with the reason.

Operational in about fifteen days. Self-sufficient in under two months. Five people covering the scope of twelve, at about half the prior budget. Around 200 assets a cycle, sustained, with per-asset turnaround from three or four days down to four hours. Every one of them read on ROAS, conversion and customer acquisition cost against the previous cycle as control, which is the number that decided what we made next. The mandate was time-boxed from day one and closed on completion, and the function kept running without me.