How I work

I set up the conditions for good design to happen without me in the room.

For most of the last five years I’ve been the only designer on large product suites, so I’ve had to build the systems, rituals and relationships a bigger team would normally carry. Here’s how that works day to day.

9 products
Sole designer across WM’s internal tools suite
2023
Start of principal-level scope at WM
40+
Engineers across the teams I partnered with

Principles

Serve the new hire and the veteran at the same time.

On WM’s dispatch board, a week-one dispatcher and a fifteen-year one used the same screen. Defaults and inline suggestions carried the new hire, the layout kept the patterns veterans already knew, and the shortcuts I recorded in observation sessions came along. Jargon stayed, explained in tooltips that never cost the table any density.

A model’s suggestion is worth nothing if nobody acts on it.

So I design for legibility, confidence and the freedom to say no. Dispatchers’ override rate on WM’s scheduling model fell from 41% to 12% once the recommendations explained themselves.

Match the interface to the thing in people’s hands.

At AT&T the self-install flow followed the physical box, not the task checklist, and support calls fell 28%.

Systems should outlast me.

If you can’t tell when I stepped away, the system worked. Driver Assist launched after I left WM, built on the decision log, data map and tokens the team already had.

How I run a design system

01
Propose
Anyone: eng, PM, design
02
Review
Me + eng lead
03
Build
Contributing engineer
04
Document
Me
05
Adopt
Product teams
06
Measure
Me + PM
Measure feeds the next proposal ↺

Governance

One rule decided what joined the library: the moment a component was about to be reused, in another product or another part of the same one, it became a candidate. One-offs stayed local until that happened.

Contribution

With one designer and nine products, requests had to flow without a meeting. Engineers filed them in Jira; I checked each against what already existed, adapted where requirements genuinely differed, and kept the products aligned.

Rituals

Monthly reviews with engineering leads, more often around releases. Release notes went out through Confluence and Slack, so nobody had to ask what changed.

21%faster delivery across the suite Read the design system case study →

Working with engineering and product

Kickoff
Pin down what the problem costs and which number should move.
Hands over: one-page problem statement
Discovery
Sit with users; map real constraints with engineering.
Hands over: constraint and data map
Concepts
Rough options with engineers in the room, cost flagged early.
Hands over: scoped direction
Build
Stay embedded, answer fast, write every decision down.
Hands over: decision log
Launch
Plan the rollout like a design problem.
Hands over: release notes in users’ language
Follow-up
Check the kickoff number against reality.
Hands over: what to fix next

Engineers come in at concept stage, not at handoff. When a debate is going in circles I’ll prototype in code or with AI tools to settle it faster, and every decision gets written down so nobody has to argue it twice.

A trade-off I negotiated

Driver Assist: truck sensors couldn’t report reliably through patchy telematics and dead zones. Rather than fake a live reading, we labelled every value with when it last updated. We gave up latency and kept the driver’s trust.

Designing with ML and AI

People act on a model when they can see its reasoning and are free to ignore it.

01
Show the reasoning
What the model suggests, why, and how sure it is, in the user’s own terms.
02
Keep people in control
Overriding is one tap, and saying no carries no penalty.
03
Earn trust over time
Start where the model is strongest, and expand as people see it work.
04
Measure the trust
Track override rate, not acceptance. Acceptance can be gamed; overrides tell you the truth.
$3.2M saved in the pilot year of WM’s ML dispatch scheduling. Read the case study →
Now: leading design on an AI agent platform at Koi Studios. Details stay private until it ships.

Mentorship and team health

How I run critique

The presenter sets up the problem and constraints before any screens, so the room starts level. Everyone else asks questions until the decisions make sense. Feedback is direct and kind, and we close on written action items and a genuine “nice work.”

How I’ve mentored

A junior designer joined WM on a one-off B2B customer portal. I introduced her to every PM and engineer she’d need, met weekly to set direction and constraints, and asked where she wanted to grow. She moved onto the projects I recommended her for and started contributing to the design system.

How I work

Focused hours, written decisions, async by default, and no expectation of after-hours replies, from me or from anyone I work with. Good teams don’t run on heroics.

“Matt was in charge of so much, it was crazy that he still made me feel so heard and taken care of at the same time, all the time.”
Designer, WM

If your team needs a designer who builds the systems around the work, I’d love to talk.