Copy, Paste, Pray No More: An AI Translation Workflow Case Study
At Xerox, my creative team took a translated-asset production cycle of 801 exports from three days down to three hours. The secret was not hiring a team of caffeinated interns. It was a validated Excel file, a writer.ai playbook, an export script and a quality-control checklist that refuses to let anything slip.
11/8/20264 min read
The problem: 801 files and one very tired mouse
On the creative design team at Xerox, speed is the job. Campaigns don't wait, and "we're still exporting the Dutch social tiles" is not a status anyone wants to say out loud in a stand-up.
Here's the math that kept us up at night. One campaign meant hero banners, social posts, in-the-box inserts and a landing page. Each came in multiple designs and sizes. Then every one of them had to ship in nine countries.
That adds up to 801 individual exports. Built by hand, that's 801 rounds of open, paste, check, rename, export, repeat. Somewhere around file 240, a designer starts to see kerning in their sleep.
As the AI-centric creative lead, my question wasn't "how do we work harder?" It was "which parts of this are we doing that a machine would happily do for us, without complaining about it in Teams?"
The old way: three days of copy, paste, pray
The old pipeline took three full days per campaign. Nobody was slacking; the work was simply manual at every step that mattered.
Copy went in by hand. Translated strings were pasted into InDesign one text frame at a time. Every paste was a fresh chance to drop a German compound noun into a Spanish layout.
File naming was freestyle. Nine countries times hundreds of files is a lot of opportunities for hero_final_v2_FR_REAL.jpg.
Exports were one-by-one. Each JPEG meant the same dialog box, the same folder pick, the same click. Multiply by 801.
Checking happened at the end. Mistakes surfaced late, when fixing them was most expensive and everyone was most tired.
The real cost wasn't just time. It was creative energy spent on clerical work, which is the most expensive way to rename a file ever invented.
The new way: let the robots do the boring part
The new workflow has four steps. Humans own the first two, where taste and accuracy matter. Automation owns the last two, where speed and consistency matter.
[embed: node/380d0e3c-0f73]
Finish the master in English (EN-US) in InDesign. Designers build and approve every layout once, in one language. This is the only place creative decisions get made, which keeps them where they belong.
Translate and validate the strings in Excel. All copy for every country lives in one structured spreadsheet. Translators fill it in and reviewers sign it off before anything touches a layout. The spreadsheet becomes the single source of truth, so no one is translating out of an email thread from last Tuesday.
Run the writer.ai playbook: "IDML Translation from Excel." The playbook reads the validated Excel and does three jobs in one pass:
Replaces the English copy with the translated strings.
Swaps the country code on the pasteboard, which the export script reads later.
Names each file with its country code, so FR means France every single time.
Run the export script to save out the JPEGs. A script, built with writer.ai, reads those pasteboard country codes, names each file and exports every JPEG to whatever folder you choose. You click once. It handles the other 800.
The quiet hero here is the pasteboard country code. It's a tiny label sitting off-canvas where no customer will ever see it. Yet it's what lets the script know which country each file belongs to without a human squinting at a filename.
The other quiet hero is step 2. Validating copy before layout means the automation is only ever fast at doing the right thing. Automating a broken process just gets you wrong answers sooner.
The results: three days became three hours
The full production cycle for 801 exports dropped from 3 days to 3 hours. Counting three eight-hour workdays, that's about eight times faster. Counting the snacks we no longer stress-eat, it's incalculable.
What changed beyond the clock:
Designers design again. Creative time goes into the EN-US master, not into file housekeeping.
Naming is boringly consistent. The playbook and script name every file the same way, every time. Boring is the goal.
Errors show up early. Copy problems get caught in the Excel validation step, before they're baked into 801 JPEGs.
The process scales. A tenth country means more rows of validated strings, not a fourth day of production.
The Quality Control Playbook: trust, but count
Fast is great until a market launches without its landing page. So every run gets a second check, built on one simple idea: the files that come out should match the files we expected, and every one should be in the right language. Automation does the work; the QC playbook makes sure it did all of it.
The playbook checks the work at a few key moments rather than only at the end. We look at the source copy before anything runs, because a wrong spreadsheet just produces wrong files at impressive speed. We look at the layouts after the translations land, because longer languages and special characters love to cause surprises. And we count the exports at the finish line, because the easiest problem to miss is the file that simply isn't there.
Two habits make it stick. First, a second set of eyes, someone who didn't run the export, signs off before anything ships. Second, when something's wrong, we fix it at the source and re-run, instead of patching a single file by hand and quietly breaking the system for next time.
The best part is that the QC check can be automated too. A robot that counts the robot's work is the most relaxing thing in creative operations.
What I'd tell another creative lead
AI didn't replace anyone on my team at Xerox. It replaced the part of the job nobody would put on their portfolio.
Automate the repeatable, protect the creative. Our designers still make every design decision in the EN-US master. The machine just multiplies it by nine.
Clean data first, speed second. The validated Excel is what makes the automation trustworthy. Skip it and you'll produce 801 mistakes in record time.
Small hacks, big wins. A country code hidden on the pasteboard sounds trivial. It's the hinge the whole export script turns on.
Build your own tools. We used writer.ai to build the playbook and the script ourselves, shaped to how our team actually works.
Always count the robot's homework. Speed without a QC check is just a faster way to find out what's missing on launch day.
Three days to three hours didn't come from one big breakthrough. It came from asking, at every step, "does a human really need to do this?" Most of the time, the honest answer was no. The rest of the time, the answer was "yes, and they'll do it better now that they're not exporting JPEG number 612."