./relay

one screen for dispatchers juggling six tools and a phone.

role:

design lead, v2 (contract)

team:

1 pd · 8 eng · 2 pm

timeline:

feb — oct 2024

platform:

desktop web · ops

[ relay — dispatch board, map + queue, full-bleed ]

$ cat ./context

the problem

1 in 12 jobs was going to the wrong driver.

i designed relay v1 while embedded at shiplane (2019—21). it worked — then the company 4×’d, and the board grew into a junk drawer of filters, tabs and modals.

by 2024 dispatchers were running shifts across six tools plus a phone: relay, two spreadsheets, whatsapp, the carrier portal and a wall calendar. shiplane brought me back to lead the v2 redesign.

the brief >

“give dispatchers one screen they never have to leave.”

−57%

mis-assignments within two months

90 s

median time per dispatch (was 8 minutes)

94%

dispatcher adoption in six weeks, 12 depots

6 → 1

tools per shift (the phone survived)

$ cat ./process

research → flows → iterations

(a) research

14 hours of shift shadowing beats any survey.

three depot ride-alongs and full-shift shadowing. the killer finding: dispatchers left relay not because it lacked data, but because switching cost less than scrolling. every context switch was a mis-assignment risk.

[ shift shadowing notes — tool-switch map ]

[ board information architecture — v2 ]

(b) flows

the board became the single source of truth. keyboard first.

map + queue side by side, exceptions promoted to their own lane, and a command palette so a dispatch never needs the mouse. everything else — carrier data, driver chat — pulled in, not linked out.

one depot piloted live. then twelve.

we ran the bilbao depot on v2 for four weeks while the rest stayed on v1 — same jobs, two systems, honest numbers. the command palette’s discoverability failed in week one; a visible “⌘k” hint fixed it.

[ pilot dashboard — v1 vs v2, week by week ]

$ open ./solution

what shipped

[ dispatch board — final ui, map + queue + exception lane ]

[ command palette — ⌘k ]

assign, reroute, message — every action reachable in two keystrokes. power users never touch the mouse.

[ exception lane — late / at-risk jobs ]

at-risk jobs surface themselves. no more hunting through filters while a truck idles.

$ tail ./outcome

after launch

v2 rolled out to all 12 depots by october 2024. the two spreadsheets were formally retired in november — a small internal funeral was held.

“i haven’t opened the old spreadsheet in two months.”

— depot lead, bilbao · pilot retro

// next case — 04/04

./loop →

back to index ↑