A few years ago I was staring at a delivery that could not be done in time. Big launch, going out across a bunch of platforms, about 60 different assets that all had to carry the same branding and the same offer. Everything matched. It was ready.
Then my boss wanted one change. Small. Nitpicky. The kind of thing nobody scrolling past the ad would ever notice. And to make it, I'd have to hand-edit all 60 assets, which was about a day and a half of work I did not have before the deadline.
So I got on a call and told him the truth. My coworkers were listening in, which made what happened next a little more memorable. I said, this isn't going to go out on time if we make this change, and honestly the change won't move the performance of these ads at all. Because it wouldn't. It had nothing to do with whether the ads worked.
He said, "You're an editor. Figure it out. I want it." And that was the end of the conversation.
So I made the change. And the ads never went out. Later he asked me why they never went out, and I told him the truth again: because you wanted the change. He looked at me with a completely blank face, and then he walked away. It just did not land that he was the bottleneck. He wasn't willing to hear it.
That was the moment. I couldn't win that argument with words. So I decided to win it by deleting the bottleneck entirely.
I couldn't code. So I sat down with ChatGPT, and this was back on GPT-3.5, when the thing was genuinely dumb, and I vibecoded my first app. That was the first version of Split-Test-O-Matic.
What it did was simple and boring and it saved my life. You gave it your offer graphics in each size, and it took your videos, cut every one of them to the right size, dropped the right offer graphic on top, and slapped a watermark on the ones that needed a watermark. As many assets as I wanted, in one run.
The day-and-a-half job became about an hour. Two, if I was being slow.
Here's the part that still makes me laugh. When I left that job, I offered to leave the tool behind so they could keep using it. They didn't want it. They passed. So I took my little spite-machine with me, and I've been using it ever since.
The graphics thing was step one. The real prize was video.
I wanted to split test ads properly. Not one video against one other video. I wanted to test different hooks against different stories against different offers, every combination, so I could actually find what worked instead of guessing. And when I went looking for something that would build all those variations for me, there was nothing. Plenty of tools will test a video you already made. Plenty will generate a generic AI ad. Not one of them takes your hooks, your stories, and your offers and assembles every version for you.
Doing it by hand took days. I'd make mistakes. The file naming alone was a nightmare, and I had to do all of it manually. And I think that right there is why almost nobody runs real video split tests. It's expensive and it's slow. Even wanting to test 144 versions of something sounds insane to most people, because for most people it would take weeks. The only folks I hear about running tests at that scale are people like Alex Hormozi.
So I rebuilt Split-Test-O-Matic to kill that problem for good.
Now it's dead simple. You make your sections. Your hooks, your stories, your offers, and an end card if you want one. You drop them in as compound clips. Split-Test-O-Matic cuts every combination for you, in vertical, square, or horizontal, names everything correctly, adds the captions, and exports the whole batch ready to post.
It runs inside DaVinci Resolve, which turns out to be the best place on earth to automate this kind of thing. I got the whole pipeline automated end to end. The only parts left for a human are the parts that should stay human: actually editing your sections and choosing your music. Everything after that is push-button. And now that Claude Code exists, you can have it help you set the thing up, and then let Split-Test-O-Matic do the cutting even faster.
For one client I accidentally made over a thousand video assets. Not on purpose. I just ran it and it kept going, and then we had to whittle it back down. The part where I generated more than a thousand ads took me about as long as walking to the fridge for a drink.
For another, I delivered 500 video ads off a single day of shooting. That took a week or two, and it only took that long because I was doing it around a full-time job at the same time. On its own it would have been faster.
It does my captions too. The whole thing is a little bit wild to watch run.
I've kept this quiet for a long time. I'm done keeping it quiet.
I built Split-Test-O-Matic. I built the first ugly version out of spite on a dumb model years ago, and I built the version that now turns one shoot into hundreds of tested ads. I checked, and there's still nothing else that does what it does. I might turn it into something you can subscribe to down the road, and there's a tutorial coming. But before any of that, I wanted my name on it, in public, with a date. So here it is.
It went from "you're an editor, figure it out," to me figuring it out, and now I edit about 30 times faster than most people who do this for a living. Same guy. I just stopped arguing with the bottleneck and built my way around it.
I'll send the Split-Test-O-Matic walkthrough, plus what I'm building next. Free, no fluff.
Free. Unsubscribe anytime.I'm a marketer who builds his own tools. More at adamgarceau.com.