Product UX — 2024
Northwind Lending
A twenty-three step mortgage application rebuilt as a resumable, plain-language flow that borrowers finish on a phone.
Mortgage application drop-off fell from 61% to 34% in a single quarter.
- Client
- Northwind Credit Union
- Discipline
- End-to-end product design
- Engagement
- 12 weeks, Jan to Apr 2024
- Team
- Two engineers, one compliance officer, one mortgage loan officer
Overview
Northwind is a regional credit union with eleven branches and a mortgage product that had quietly become its worst digital experience. The online application had been extended one field at a time for nine years. By 2023 it ran to twenty-three screens, asked for the applicant's address four separate times, and could not be paused and resumed.
I was brought in for twelve weeks to redesign the borrower-facing application without changing the underwriting system behind it. Every field that existed had a compliance reason to exist. The work was not about removing questions. It was about asking them in an order a human being could follow.
Role
Lead product designer, contract
12 weeks, Jan to Apr 2024 · Two engineers, one compliance officer, one mortgage loan officer
- Ran the discovery interviews and the branch-side shadowing sessions
- Mapped the existing twenty-three screen flow against the underwriting field requirements
- Designed the full application flow, save-and-resume model and document upload
- Wrote every piece of interface copy, then took it through legal review
- Built the prototype used for eight moderated usability sessions
- Paired with engineering through build and shipped in three staged releases
Problem
The analytics said 61% of applicants abandoned, but not why. The heaviest drop was on screen nine, the employment history screen, which asked for two years of history in a single unscrollable table with no way to save.
Shadowing loan officers at two branches turned up the real behaviour. Applicants were starting online, hitting a question they needed a document to answer, and then phoning a branch to finish. Staff were re-keying entire applications by hand. What looked like abandonment in the funnel was actually a very expensive human fallback.
The constraint was firm: the underwriting system needed the same fields in the same shape. I could change the sequence, the grouping, the wording and the timing, but not the data contract.
Process
Field audit before anything else
I built a spreadsheet of all 118 fields the application collected and, with the compliance officer, tagged each one with who needs this, at what point, and what happens if it arrives late. Thirty-one fields turned out to be needed only at closing, not at application. That single finding shaped the whole redesign.
Resequencing around effort, not around the database
The old flow followed the order of the underwriting record. I regrouped it around what the applicant could answer from memory, what needed a document, and what needed a co-borrower present. Everything answerable from memory came first, so people reached a save point before they hit a wall.
Save and resume as a first-class state
Every section became independently completable, with a magic-link resume that reopened the application on the exact question the borrower left. I designed the partially-complete state as carefully as the empty and finished ones, because it is the state most applicants actually live in.
Copy rewritten, then defended
I rewrote 118 field labels and 40 helper strings into plain language, then sat with the compliance officer line by line. Roughly a fifth needed the regulated wording. For those I kept the legal phrasing as the label and added a plain-language explanation underneath, rather than fighting a battle I would lose.
Eight sessions, three of them at a branch
We tested with real prospective borrowers, including three sessions run in a branch lobby on the applicant's own phone. Branch testing surfaced two problems no lab session would have: the document camera step failed in low light, and people would not enter salary details where a queue could see the screen.
Outcomes
- 34%
- Application drop-off, down from 61%
- 19 min
- Median completion time, down from 41 minutes
- 3x
- Applications completed entirely on mobile
- −72%
- Applications re-keyed by branch staff
Measured against the ninety days before launch, on comparable application volume. The re-keying figure came from the loan officers' own queue, and it is the number the team was proudest of, because it gave four people their afternoons back.
Gallery
Tools
- Figma
- FigJam
- Maze
- Dovetail
- Notion
- Linear