The only designer on the system that inventoried Google’s campus
Google couldn’t tell you what furniture they owned across 238 Bay Area buildings. I was the only designer at the vendor hired to fix it.
Ten thousand items landed in the first month, more than the platform had taken in over its entire history. That came from the data model, not the screens.
Nobody knew, including me
The people who knew didn’t work for Google, they worked for vendor partners. The systems didn’t talk.
The same chair had three names depending on which of five organizations was describing it.
Nobody was wrong. Everybody just had a different word for the same chair, which turns out to be harder than nobody knowing.
So the data model was going to be an agreement between five organizations before it was ever a diagram.
Rheaply was seed-stage and nobody in the building knew this domain, me included. I’d never managed a warehouse or moved a pallet.
I think that’s the only reason I went and asked instead of guessing.
My role
Only designer. I ran every research session and workshop myself, designed the interface, and proposed the data model.
Not mine: the line that reframed the project. That came from Stephanie, our PM. My contribution also ends at the first version that launched.
Forty-six sessions
- 16 stakeholder interviews, to find who actually knows anything. Not the same as who has the title.
- 11 people on a clickable version, before anything was built.
- 9 watching people use the real thing, then 10 more after we fixed it.
- 6 workshops covering 8 activities. That’s how five organizations agreed on one model.
Then time onsite, because you can’t design an inventory system from a conference room.
Watching things get moved and relabeled and lost is the only way I was going to find out the paperwork was the real problem.
One finding changed the build: a made-up industry term takes on its own meaning the second it’s inside an application.
People trusted our label over their own vocabulary. I thought the interface was reflecting their language. It was quietly overwriting it. So I defined the words explicitly on screen.
Introducing product sets
My first pass let people group objects into each other. It sounded flexible.
What it produced was items inside items inside items. You can’t search it, can’t report condition on it, and can’t count it, which is the entire job.
A set describes a type, and items belong to it. So 400 identical tables become one set with 400 items, each carrying its own condition.
One line I still use
From Stephanie, out of the research: successful inventory systems avoid creating pain off-platform as much as they create delight on-platform.
Every hour our interface saved was worth nothing if using it meant a warehouse team had to go re-label a pallet. That reframed what counted as a win.
“Knowing how many items are actually on the floor, and how many are available in our attic stock, is exactly what we need.”
— Peter G., Furniture Services Lead, Google
“It’s fantastic. I’m super, super, super happy. Great job.”
— Krista B., Director of Space Management, Google
What I got wrong
I found the nesting problem once engineers were already building, instead of in something clickable and cheap.
It was discoverable in a day of sitting next to an engineer. Instead it cost weeks.
A drawing of how information fits together can look sensible and still be wrong underneath.
I’d spent six weeks in research specifically so I wouldn’t make a mistake like that, which is the part that stings.
I also scoped down late. We shipped on time, but the cuts happened under pressure at the end instead of deliberately at the start.