Thirty-two exports. Every two weeks.

• Paid social meant a new ad every two weeks for UK, US, French, and German markets, each cut four ways (16×9, 9×16, 1×1, 4×5), then again as clean and burnt-in subtitle versions, plus SRTs for the platforms. That is thirty-two videos per cycle before anyone starts the next brief.
• Editors were burning hours in Premiere and Media Encoder on the same repetitive export and file-organisation grind. The craft suffered while the queue grew.
• A/B testing needed both clean and subtitled masters so Meta and the other networks could pick the winner. Skipping a version was not an option.

I built a custom app with ChatGPT for the video team, wired into how we already worked in Adobe Premiere Pro and Adobe Media Encoder. Editors stayed on the edit. When it was time to export, they handed the job to the app, organised the deliverables, and went on a break or the next cut while it ran.

The product was not a slide about AI. It was a tool sitting next to Premiere: batch the aspect ratios, languages, clean vs burnt-in, and SRT handoff so Media Encoder did the heavy lifting and the folder structure stayed sane. Same campaign cadence (every two weeks, four markets) without the export marathon that used to own half the day.

Four languages × four aspect ratios × clean and subtitled masters. The maths was the pain. The app was the way out, so the team could hand off export and get back to making ads that actually win the A/B.

• Turned an export process that could take hours into one that finished in minutes.
• Freed editors to focus on the cut instead of babysitting thirty-two near-identical renders and file names.
• Proved the AI brief I still use: build for the job the team already has (ads, Premiere, Media Encoder), not a generic chatbot on the side.