Case study 05 · Leadership · Banking

Running a 24‑hour design team

How I run the design team behind the AI-native finance rebuild for a US digital bank: requirements settled with AI prototypes, juniors paired with seniors by strength, and a handoff between time zones that keeps design moving around the clock.

Role
UX Manager, design lead on the program
Team
up to 5 designers (1 Principal, 2 Senior, 2 Junior) across PST and IST
Timeline
2024–present
Program
100+ developers and BAs, about 5 development pods and a BA pod
Client
a US digital bank
Outcome
On time, at high accuracy, while development schedules slipped

My part in this

Decision I influenced
Settling each module’s requirements with an AI prototype before the team spent any design time.
What I pushed back on
Leadership wanted each designer siloed in one module, the way development worked. I set up pods that met regularly to share insight and components, keeping the work consistent.
Trade-off I made
Pairing juniors with seniors slowed the seniors a little but sped up the juniors’ growth, and became a skill exchange across the whole team.
What I noticed
The team moved faster when I worked beside them as a hands-on contributor, showing them how to push past their limits and get comfortable at a rapid pace.
What changed because I was there
Two interns grew to own the design system and critical modules, and my team delivered on time with high quality while other teams’ schedules slipped.

The problem

The finance rebuild is a long program made of many modules: every page across the Functional, Functional 2, Workflows and Banking areas, plus Payables, a very large module of its own. Each one needs requirements, direction and design before development can start, and the bank wanted all of it reimagined, not reskinned.

I lead UX for the whole program: I work with the pod leads to define and shape the product from the client’s requirements, coach the pods on writing more efficient stories and setting development goals, and I’m leading the move to Claude-based workflows for design and for the development and BA teams downstream.

On this program, my designers’ hours and leave run through my sign-off, I delegate the work within the project, and their directors come to me for input on their reviews. The team is split between the US and India. The challenge was to keep that team moving fast without me becoming the bottleneck, and to grow two junior designers along the way. Leadership wanted each designer siloed in one module, the way development worked; I set up pods that meet regularly to share insight and components, so the work stays consistent.

What I did

Key insight A design team moves fastest when the ask is clear before design starts, and when each person’s work plays to what they’re best at.

  1. 01

    Settle the ask with an AI prototype

    For each module, I set requirements and direction with stakeholders, then build an AI prototype: a napkin sketch that clarifies the ask before my team spends design time. With Figma Make and Figma agents, we defined requirements for 10 modules in 2 months, a clear path for the year ahead.

    Trade-off: A working prototype can look finished and anchor everyone on a first idea. I hand it over as a sketch, not a spec, and the team owns the solution.

  2. 02

    A 24-hour design cycle

    I gather direction and relay the client’s answers during my day; the team designs during theirs. BA and dev teams review the work in the hours between, and their feedback shapes the next day’s work for me and the team. My Principal designs alongside everyone, clears blockers, and fields dev and BA questions so the designers stay focused.

    Trade-off: A late-day question can wait overnight. Daily syncs, Figma comments and a Principal who can make the technical call keep that wait short.

  3. 03

    Pair by strength

    Each junior works alongside a senior, and modules go to whoever’s strengths fit: technical work to a former developer, research-heavy work to a junior who thinks like a researcher, the component library to our strongest UI designer.

    Trade-off: Matching work to strengths can typecast people. Pairing spreads skills across the team instead of keeping them in one lane.

  4. 04

    Fix the design system’s accessibility, fast

    The design system had an accessibility gap the bank’s UX team and executives had already approved. Using Figma agent skills, we proved the fix in days and reworked about 80% of the system in roughly 8 hours, where checking a few button options used to take 16.

    Trade-off: Reopening an approved system meant raising a problem leadership thought was solved. A working fix and a plan made it an easy yes.

The accessibility fix, before and after

We audited the original library against WCAG 2.2 on every light and dark surface it sits on, including gradients: buttons, tags, status chips, alerts and avatars, with 52 specialty colors tested for tags alone. Two problems ran through it. Soft, tone-on-tone fills looked calm but disappeared into the page, so components had no visible edge. And the primary buttons put white text on orange, which no one could read reliably.

Before · Light
ActiveOverdueIn progress
Chip edge vs surface 1.30–1.37:1 · Fail
After · Light
ActiveOverdueIn progress
Chip edge vs surface 3.15–5.61:1 · Pass
Before · Dark
ActiveOverdueIn progress
Chip edge vs surface 1.08–1.45:1 · Fail
After · Dark
ActiveOverdueIn progress
Chip edge vs surface 5.92–8.11:1 · Pass
Status chips on the hardest surfaces in each mode. The colors stay the same; a 1px stroke in each chip’s accent color gives it an edge that meets the 3:1 minimum for non-text graphics. After the change, all 6 status types pass on all 12 surfaces.
Before · Primary button
White text 2.37:1 · FailButton vs surface 1.51:1 · Fail
After · Primary button
White text 11.73:1 · PassButton vs surface 7.46:1 · Pass
The primary action on the darkest light surface. Blue replaced orange as the primary color. In the audit, all 20 button states (4 styles × 5 states) pass on the hardest light surface and on the dark radial gradient.

The original error alert, light mode

  • Alert vs the page1.05:1 · Fail
  • Icon circle vs the alert1.41:1 · Fail
  • Error glyph vs its circle2.94:1 · Fail
  • Left border vs the darkest surface3.07:1 · Borderline

Text inside the alert passed easily (16.35:1). Everything that tells you an alert is there did not.

Results

  • The most reliable team on the program. While development timelines slipped, design delivered on time and with high accuracy, and the team has scaled down because its work landed.
  • A year of work, planned in 2 months. Requirements for 10 modules, set with AI prototypes.
  • Interns to owners. Both junior designers started as interns. One now owns the design system, working directly with UI developers; the other owns critical modules like Payables and Data Migration.
  • 85% less rework. Rework dropped 85% once modules started from AI prototypes, across 32 modules delivered so far.
  • Prototypes the whole program uses. Beyond my design team, the AI prototypes are used by every pod lead, some of their developers, 5 BAs and 3 members of the client’s product team.
  • An AI-forward practice. My team was the first at Ascendion to put Figma Make and Figma agents to work on client delivery. I’m defining agent skills and agentic practices with the UX Circle directors. Browse the skills library →

What’s next: closing the gap between canvas and code

I’m pushing for Claude team seats so designers and UI developers can work together in the space between Figma and code. The workflow I’m setting up:

  1. Build prototypes in Claude to refine requirements and the experience.
  2. Export the screens, step by step, to the Figma canvas.
  3. Refine details by hand in Figma, and direct Claude, through the Figma MCP, to make the larger, slower updates.
  4. Send the refined workflows and screens back to the prototype to resolve them.
  5. Deliver to the UI developers a foundation where the translation from design to code is already solved.

What I’d do differently

I’d push for more overlap between the client’s hours and my team’s. The time difference made that hard, and while I’m willing to work odd hours, I don’t want to ask that of my team. They would have benefited from more time with the client directly.

See the product this team is building →