UX designer · Minneapolis, MN · Nine years · product and service · in-house and contract

Every portfolio says “we”.

Four projects, eight disciplines, and a cell for each one saying whether I led it, did it, helped with it, or had nothing to do with it — with the people who did the rest named underneath. Then the outcomes, including the two that did not move and the two nobody measured.

Who did what Outcomes, including the flat ones

4 projects · 8 disciplines · 32 cells

9 of 32 say I led it. 5 say it was not me at all.

Four words, and the fourth one is the reason the table exists. On the freight console I did not touch visual design and it is the best-looking thing on this page. On the benefits project I ran four of nineteen research sessions and would have been within convention to write “led discovery”. Nobody would have checked.

Led

I set the approach and other people worked to it.

Did

I did the work, alone or nearly.

Helped

Somebody else owned it and I put in real hours.

Not me

Somebody else did it. They are named under the project.

Every cell says a word — the colour is only a shortcut for a reader who already knows it
Project	User research	IA and flows	Interaction design	Visual design	Content design	Prototyping	Usability testing	Front-end build
Patient intake, regional clinic group2025 · Team of six · eleven weeks	Led	Led	Did	Helped	Did	Did	Led	Not me
Dispatch console, regional freight2024 · Team of four · nine months, part time	Led	Led	Led	Not me	Helped	Did	Did	Helped
Catalogue search, county library system2024 · Team of three · six weeks	Did	Led	Did	Did	Led	Did	Helped	Did
Benefits enrolment, 900-person employer2023 · Team of nine · fourteen weeks	Helped	Did	Did	Not me	Led	Helped	Not me	Not me

The four, and who else was there

Named, not implied.

Patient intake, regional clinic group

2025 · Team of six · eleven weeks

The brief

Replace a paper intake form and a kiosk nobody used with something a person can finish in a waiting room on their own phone.

What shipped

A seven-screen flow, in production at nine sites. Median completion 4 min 20 s, down from a paper form that took 9 minutes and was 31% incomplete at the desk.

Front-end was Halvard Nyquist, two engineers. The visual language was already a design system I did not build; I chose from it and argued about two components.

Dispatch console, regional freight

2024 · Team of four · nine months, part time

The brief

One screen that eleven dispatchers look at for eight hours a day. The old one had 340 rows visible at once and everybody had built their own colour-coded system in a spreadsheet beside it.

What shipped

A density-first console with saved views. Still in use. The spreadsheets did not disappear and I have stopped thinking that was a failure.

Visual design was entirely Ines Bergqvist, who is the reason the thing is legible at 340 rows. I wrote CSS for the table layout and nothing else.

Catalogue search, county library system

2024 · Team of three · six weeks

The brief

Search that returned 900 results for “toni morrison” and put the large-print edition of a study guide first.

What shipped

Reworked ranking and a facet rail. The interesting part was not the interface — it was a metadata cleanup nobody wanted to fund and I did not do.

The ranking change was Marguerite Tolliver, a librarian who understood the catalogue better than any of us. I built the front end because there was nobody else, and it shows in two places.

Benefits enrolment, 900-person employer

2023 · Team of nine · fourteen weeks

The brief

An annual enrolment where 40% of staff picked the default because reading the alternative took an hour.

What shipped

A comparison step and plain-language summaries. Default-picking fell to 26%, and I do not know how much of that was the design and how much was HR emailing everybody twice.

Research was led by Anneke Prud’homme; I ran four of nineteen sessions. Visual design and build were an agency. Usability testing was run by their team and I watched.

Patient intake, regional clinic group

The prototype was wrong by week three.

A case study normally runs challenge, process, solution, which is the order the work is written up in and never the order it happens in. This is the order it happened in.

What I believed in week one That the kiosk failed because it was ugly and slow, and that the same flow on a phone would fix it. I had a prototype by day four. It tested well with eight people in a room in an office.
What the research actually found Eleven of nineteen patients were filling the form for somebody else — a parent, a partner, an adult child. The kiosk was not slow for them, it was wrong: every field said “you”.
What that killed The prototype, and three weeks. Also the idea that completion time was the metric, since a proxy filling a form carefully takes longer and that is the good outcome.
What we changed One question at the top: who is this for. It rewrites the pronouns in every subsequent field. It is four lines of logic and it is the entire project.
What it cost Three weeks of my work and a difficult conversation with a product manager who had already told a board the date. We shipped two weeks late. The board did not notice; the desk staff did.
What I would do differently Run the nineteen sessions before building the prototype rather than beside it. I knew that. I built it anyway because a prototype is how I think, and the honest fix is to build it and then not show anybody for a week.

Open the Research Plan →

Eight metrics · 2 flat · 2 never measured

Including the ones that did nothing.

Most results are flat. A portfolio that shows only the ones that moved is reporting a selection, and everybody reading it knows that, which is why the ones that did move get discounted too.

Every row carries the caveat that would normally be left off the slide
Metric	Before	After	Verdict	What the number does not tell you
Completion at the deskPatient intake	69%	94%	Moved	Nine sites, six months either side, same season. The most defensible number on this page and I still would not put it on a slide without the window.
Median time to completePatient intake	9 min	4 min 20 s	Moved	Genuine, and less meaningful than it looks: a proxy filling the form for somebody else takes longer, and that is the outcome we wanted.
Time to reassign a loadDispatch console	41 s	38 s	Flat	Three seconds is noise at n=11. I called this a win internally for about a week and I was wrong to.
Spreadsheets still in useDispatch console	11 of 11	9 of 11	Flat	The thing I actually set out to remove. Two dispatchers stopped; nine did not, and their spreadsheets do something the console genuinely does not.
Zero-result searchesCatalogue search	12.4%	3.1%	Moved	Almost entirely the metadata cleanup, which I did not do. Putting it in my portfolio at all is a stretch and leaving it out would be a different kind of dishonest.
Whether people found the right bookCatalogue search	—	—	Never measured	Never measured. We had no way to see whether a search ended in the book somebody wanted, and every number above is a proxy for that question.
Staff taking the defaultBenefits enrolment	40%	26%	Moved	HR also emailed everybody twice that year. I have no way to separate the two and neither does anybody else who has ever reported a number like this.
Whether people chose better plansBenefits enrolment	—	—	Never measured	Never measured, and it is the only question that mattered. Nobody wanted to fund the claims-data analysis that would have answered it.

Four things, and one of them is the whole job

What I need from you.

Five of your users, for an hour each: Not five stakeholders. Not a survey. Five people who use the thing, on a call, in the first two weeks. If that is genuinely impossible the engagement is still doable and everything I produce is a hypothesis, and I will label it as one throughout.
One person who can decide: In the room, or at least reachable in a day. A project with three decision-makers and no tiebreak does not produce a worse design; it produces the same design four times.
Read access to whatever you already measure: Analytics, support tickets, the spreadsheet somebody keeps. I would rather spend a day in your existing data than a week reproducing it badly.
Permission to ship something smaller: The single highest-value thing I can be given. Almost every project on this page got better when somebody agreed to cut a third of it, and none of them were scoped that way at the start.

Three weeks before anything is drawn.

Week 0 — Nothing is designed Reading. Support tickets, existing analytics, whatever research already exists and got shelved. Most organisations have already been told the answer once.
Weeks 1–2 — Still nothing is designed The five conversations. Recorded with permission, transcribed, and quoted rather than summarised — a summary is where a research finding goes to become whatever the reader already believed.
Week 3 — One page What I think is true, what I am unsure of, and what I would need to find out. Reviewed with the decision-maker before anything is drawn. This is the meeting that decides whether the project works.
Weeks 4–7 — Design, out loud Flows, then screens, in public — a shared file with everything in it including the discarded versions. No reveals. A reveal is a way of avoiding a conversation.
Weeks 6–9 — Testing, in rounds of five Overlapping the design, not after it. Five people, three rounds, and the third round exists to check the fixes from the first two rather than to find new problems.
Handover — A file and two hours Annotated states including the empty, error and loading ones, plus two hours with whoever builds it. The two hours are worth more than the annotations.

Common questions

Mostly about the matrix.

Why publish who did what? Nobody else does.

Because the first twenty minutes of every interview I have ever had were spent establishing it, politely and slowly. Putting it on the page gets that out of the way, and it means the things I did claim can be taken at face value. The cost is that the best-looking project on this site has “Not me” under visual design.

Is “Led” different from “Did”?

Led means I set the approach and other people worked to it. Did means I did the work, alone or nearly. Helped means somebody else owned it and I contributed real hours — not that I was in the meetings. If I was only in the meetings, the cell says “Not me”.

Why show flat results?

Because most results are flat, and a portfolio showing only the ones that moved is reporting a selection. The dispatch console reassignment time went from 41 seconds to 38 at n=11, which is noise, and I called it a win internally for a week before I looked at it properly.

What does “never measured” mean here?

That the question nobody answered was usually the important one. On the library project we never found out whether people got the right book; on benefits enrolment we never found out whether people chose better plans. Both would have been expensive to measure and both were the point.

Do you do visual design?

Competently, not excellently, and the matrix says so in two places. On a project where the interface has to be beautiful I would rather work alongside somebody who is genuinely good at it, and I have twice, and both times the work was better.

What if we cannot give you access to users?

Then we do the work anyway and everything I produce is labelled as a hypothesis rather than as a finding, throughout, including in the final deck. It is a real constraint and it is not a dealbreaker. What is a dealbreaker is being told there is access and then finding out in week four that there is not.

Work with me

Tell me who decides.

Available from March · one engagement at a time, which is a constraint rather than a policy. Fields marked with an asterisk are required.