Case study · Acme Robotics · Industrial / manufacturing · 2025
A shift-handoff system the floor would actually use.
Acme's three shifts were losing 12 minutes per handoff to verbal-update churn and missed exceptions. The plant manager wanted one system every operator would actually open — without ripping out the MES they'd just signed a five-year contract for.
- 12 min → 3 minHandoff time
- + 410%Daily exception capture
- 93%Operator adoption (day 14)
- 0 in 90 daysOn-call pages
- →Two weeks shadowing two shifts before writing a line of code.
- →Phone-first interface, keyboard-friendly on the panel PCs already in place.
- →Wrote against the MES read-only API; never bought into a migration.
- →Audit trail built on Postgres + logical replication — survives a node failure.
“We expected six months and a half-dead pilot. We got nine weeks and a tool the floor uses without being asked.”
Acme is a 220-person robotics fabricator with three rolling shifts. Handoffs ran on a paper clipboard, a Google Sheet for exceptions, a second Sheet for safety incidents, and Slack DMs for everything else. Twelve minutes per handoff, four handoffs per day, three shifts — 144 minutes of skilled labor evaporating daily.
We started with two weeks of shadowing. We watched two full handoffs per shift, and we watched the moments where the existing tools failed: the operator who couldn't find their cursor on the panel PC, the supervisor who wrote ‘see Mike about #4’ on a sticky note because the Sheet wouldn't load, the night-shift lead who maintained a private exception log on her phone because she didn't trust the central one.
We built phone-first. The panel PCs work too — but the design pivot was treating the phone as the canonical surface. Everything had to work with one thumb in a glove with safety glasses fogging up. Big tap targets, no modals, signed-off entries that don't ask twice.
We never tried to replace the MES. The plant had just signed a five-year contract, and the contract was the contract. We wrote against the MES's read-only API and kept the rest of the system independent. When the MES misbehaved, the portal kept running. When the portal misbehaved, the MES didn't notice.
The audit trail lives on Postgres with logical replication to a hot standby. We've been called twice in 90 days for vague things, never for the portal itself. The plant manager's quote is on the wall now — the one at the top of this page.
Working on something like this?
We're booking Q3. 30 minutes, no pitch.