Signage

I automated it only after running it by hand long enough to know what the rules actually were.

A signage library had grown to roughly 297 items across 69 folders, with version history kept through duplicated directories instead of a live source. I rebuilt it into eight category files, then took personal ownership of the recurring monthly setup so I could see where the work really broke. Automation came after that, not instead of it. Two other designers now extend the live source.

Sur La Table · Signage

~297items

69folders

8category sources

Sur La Table signage source inventory
Explanatory reconstruction

Explanatory reconstruction from the recorded source inventory: approximately 297 items across 69 folders became eight category sources. Items and folders are distinct units. Source architecture: Jin Mi.

Version history had become the filing system.

Roughly 297 items sat across 69 folders — five annual cycles, retired directories, several versions of the same source material. SharePoint already had version history. The team had built its own out of folder copies anyway.

Nothing was lost. It just was not possible to tell which file was current.

I consolidated it into eight category-based Illustrator sources between January and March 2025, with shared layout and typography rules, and let SharePoint do the versioning it was already doing.

Then I took the monthly setup myself.

Fixing the files did not fix the recurring work, so I took direct ownership of the monthly artboard setup. Deliberately, and at a real price: a workload surge for me every month.

That was the trade I wanted. Doing the setup by hand surfaced what the folder structure had hidden — which assets moved together, where the same exceptions kept reappearing, which rules were stable enough to encode, and which decisions still needed a person in front of them.

Automation came fourth.

Only then were scripts worth writing. Illustrator scripts and AI-assisted helpers took over setup, auditing, organization, and handoff. The floor-set workflow placed 65 of 65, turning a full day of setup into 30 to 45 minutes of touch-up, calibrated against 340 real placements sampled over five months and consolidated from four scripts into one.

It did not go smoothly, which is most of why I trust it. Before any code, I ran a stress test with two reviewers given deliberately opposite goals; it caught two load-bearing risks. The first June test then placed about 12 of 65 across 18 pages. My initial diagnosis was wrong. The actual cause was a regex reading six digits where the data had seven — a one-character fix. Five later improvements were built, run, and reverted.