The week we compressed into four hours.

At 10:14 on a Tuesday morning, a designer at a gift-card company in Sydney clicked Approve on a Christmas campaign. Everyone agreed it was good. Everyone was excited to ship it. It went live the following Wednesday. Eight days. Nothing about the creative changed in those eight days — no strategy shifted, no copy improved. The work was finished on Tuesday. The shipping took a week.

Every marketing team on earth has this gap, and almost none of them can see it, because it does not look like waste. It looks like work: exporting assets, rebuilding a design as email HTML, adapting it for each region, QA rounds that catch the same three bugs they caught last time. I started calling it the approval gap — the distance between “this is done” and “a customer can see it”. Nobody budgets for it. Everybody pays it.

So why does the gap exist at all? Not because anyone is slow — because the work keeps changing hands. Designer to producer, producer to developer, developer to QA, QA back to producer — and at every handover the campaign gets re-explained, re-built, and slightly re-broken. The gap is not a speed problem. It is a translation problem. Once you see that, the solution stops being “work faster” and becomes “stop translating”.

So we built a production line that never hands the work to anyone. It reads the approved creative straight out of Figma — layout, copy layers, image frames, the lot. No screenshots, no handover doc, no designer re-explaining their own file in a Slack thread. Then it walks the whole distance itself. Press the button and watch it go.

The step people underestimate is the last one. Most automation stops at “it published successfully”, which is the machine equivalent of a builder saying the house is done without opening the front door. Ours renders what was actually published and diffs it against what the designer approved — a font substitution, a broken image, a padding collapse in one locale, all caught before any human wastes ten minutes reviewing something that was never right.

And one thing we refused to automate: judgement. A human still sits at the gate before every send, because the cost of a bad campaign in front of millions of customers is violently asymmetric to the ten minutes of review it takes to prevent one. The trick is what those ten minutes now contain. Judgement — is this right, is this us, should this go out — instead of rebuilding tables in an email editor.

But the number I actually care about is not on that board. Before the line existed, the team quietly killed campaign ideas because the production cost was not worth the punt — the strange ones, the seasonal ones, the “probably not but imagine if” ones. Now they say yes to those. That is what automation is for. Not doing the same work cheaper. Doing the work you were never doing at all.

If you build one of these, take the one lesson that cost us the most to learn: start from the design tool, not the marketing platform. Everyone automates outward from Braze or Salesforce and hits the same wall, because the sending platform holds a copy of the campaign — the design file holds the truth. Go get it where it lives.