I owned this one end to end — framing the problem, running the research, synthesizing the evidence, then designing every screen from information architecture through high-fidelity prototypes, usability validation, and developer handoff. A firmwide platform consolidating NBIA and PSCM into one standardized risk-assessment and approval experience.
Client
JPMorganChase (via Cella, a Randstad company)
My Role
Senior User Experience Designer — end-to-end design & research
What I Owned
Problem framing · discovery · synthesis · IA · wireframes · interaction & visual design · prototypes · usability testing · handoff · design QA
Stakeholders
Product, Risk & Compliance teams across Lines of Business
Duration
May 2025 – Jul 2026
Tools
Figma · Figma AI · Claude AI · Microsoft Copilot · UserTesting
Overview
The New Product Assessment (NPA) platform was launched at JPMorganChase as a firmwide solution to streamline the approval process for new products and product changes. Two long-standing workflows — NBIA (New Business Initiative Approval) and PSCM (Product/Service Change Management) — were being run as separate, disconnected processes across multiple Lines of Business. NPA unified them.
The platform consolidates both workflows into a single experience, enables end-to-end tracking, risk assessment, and governance, and improves efficiency, transparency, and compliance across LOBs. I was the Senior User Experience Designer on it, and I owned the whole arc: framing the ambiguous problem, planning and running the research, synthesizing the evidence, then designing the risk-assessment flow, task-management surface, and approval-tracking views myself — information architecture through high-fidelity Figma, interactive prototypes, usability validation, developer handoff, and implementation review.
No handoff to another designer at any stage. I designed and defended these interfaces under direct scrutiny from cross-functional reviewers, holding up interaction and information-density decisions in a highly regulated, low-tolerance-for-error environment — and stayed on through build to review what shipped against what was specified.
The Challenge
Before NPA, getting a new product or a change to an existing product approved at JPMorganChase meant navigating a maze of disconnected systems. Different LOBs had built their own tools and processes over years of incremental changes — and the cost was being absorbed silently across the firm in the form of slower time-to-market and weaker governance visibility.
Multiple disconnected systems across LOBs — Each Line of Business had its own approval tooling, with no shared data layer. Risk teams couldn't see a firmwide picture without manual reconciliation.
Manual and inconsistent workflows — Approval paths varied by LOB and even by product type. The same kind of decision was being made differently depending on who initiated it.
Redundant questionnaires and approvals — Initiators were answering the same risk questions multiple times across different systems. Reviewers were reviewing the same content multiple times in different formats.
Limited visibility into risk, tasks, and approvals — No single view existed for an initiator, a reviewer, or a governance lead to see "what is in flight, where is it stuck, and what risk does it carry."
No unified standard for risk assessment — Without a standardized rubric for evaluating risk across LOBs, decisions were inconsistent and compliance reporting was a recurring fire drill.
Research
A firmwide platform only succeeds if it absorbs the reality of how the work is actually done — not the idealized version in a process document. Our team ran a structured discovery program in the first two months, focused on understanding every role that touches an NPA submission.
20+ moderated interviews with initiators, reviewers, risk SMEs, and governance leads across multiple Lines of Business to capture how each role experiences the current process.
Cross-LOB current-state and future-state journey maps for both NBIA and PSCM, identifying every handoff, delay, and redundancy across the end-to-end approval lifecycle.
Heuristic audit of the existing LOB tools — documenting the duplicated fields, conflicting terminology, and broken handoffs that the consolidated platform needed to resolve.
Collaborative workshops with risk and compliance stakeholders to align on a single, firmwide risk taxonomy that the platform could enforce by design.
Moderated UserTesting sessions on mid-fidelity prototypes with real initiators and reviewers — three rounds of iteration before engineering committed to the build.
Embedded risk and compliance partners directly in design critiques so the platform's risk-assessment surfaces were validated continuously, not retrofitted at the end.
Key Insights
Insight 01
"I answer the same five questions in three different systems." Redundant data entry was the single most-cited frustration. Every duplicate field was eroding trust in the platform's value.
Insight 02
"I don't know where my submission is or who's blocking it." Visibility into approval status — not the approval process itself — was where the platform could deliver immediate, felt value to initiators.
Insight 03
"Risk reviewers and product owners speak different languages." The same field meant different things to different roles. The platform had to translate without dumbing down — terminology was a design problem, not a copywriting problem.
Insight 04
"Governance needs an audit trail, not a static report." Compliance leads weren't asking for prettier dashboards — they were asking for an evidentiary record of every decision and who made it.
Design Process
The design work was structured around three principles that came directly out of research: ask once, reuse everywhere, show status by default, and standardize risk without flattening it. Every interaction had to defend one of those.
Role-based personas for initiators, reviewers, risk SMEs, and governance leads — grounded in real workflow observations across LOBs.
Prioritized problem statements, design principles, and a shared firmwide vocabulary that risk and product partners signed off on before we sketched a single screen.
Cross-functional design sprints with product, risk, and engineering — exploring multiple workflow patterns before converging on a unified information architecture.
High-fidelity Figma prototypes, three rounds of moderated usability testing, and continuous risk-partner critique throughout the build phase.
Design Decisions
Three design decisions had the greatest impact on how the platform feels to use day-to-day:
Unified intake with smart reuse — One questionnaire intelligently scopes to NBIA or PSCM based on the answers given. Previously-submitted data auto-populates wherever the firm already has it, so initiators are never asked the same question twice.
Live approval-tracking surface — Initiators, reviewers, and governance leads each see a tailored view of "what is in flight, where it is, and who is blocking it." Status is communicated by default, not by request.
Standardized risk assessment, role-aware presentation — A single firmwide risk taxonomy is enforced under the hood, but the way risk is presented adapts to the role — initiators see plain-language prompts, risk SMEs see the underlying rubric and evidence trail.
Hands-On Deliverables
Everything below came out of my own Figma file. This was not a direction-setting role with execution handed off — I did the drawing, the prototyping, the testing, and the spec writing.
Current- and future-state maps for initiators, reviewers, risk SMEs, and governance leads — spanning both NBIA and PSCM, with every handoff, delay, and duplication marked.
A unified IA and navigation model that absorbed two legacy taxonomies into one structure, plus the task flows and user flows for every path through submission, review, and approval.
Low- and mid-fidelity wireframes for the intake questionnaire, risk-assessment flow, task-management surface, and approval-tracking views — annotated with interaction logic and conditional branching.
Production-ready screens in Figma using auto-layout, variables, and design tokens — built as reusable components against the firm's design system, covering default, hover, focus, loading, empty, and error states.
Clickable prototypes covering the key scenarios and edge cases — partial submissions, rejected reviews, reassigned approvers, multi-LOB routing — used for both stakeholder alignment and moderated testing.
Interaction specs and design documentation for engineering, followed by implementation review against the spec — logging and prioritizing the gaps between designed and built.
The Screens
Each screen below exists to settle a specific question raised earlier in this case study. They are ordered the way a compliance officer moves through the platform — from what needs attention this morning, through assessment and remediation, to the report an examiner actually reads.
These are dense enterprise surfaces. Select any screen to open it full-size, then scroll, pinch, or use the zoom controls to read the detail — the numbers, control IDs, and status language are the point, and they do not survive being shrunk.
The landing surface answers “what needs my attention today” before the user has to ask. Counts, due dates, and severity all resolve without a click — the direct answer to the most frequent and most emotional question in the research.
One firmwide taxonomy, four categories, every risk carrying an owner and a review cadence. This is the artifact that settled the product-versus-risk terminology conflict — agreeing the structure here is what unblocked screen design everywhere else.
The intake path. Scope, category and ownership are captured once and reused downstream, so nothing the firm already knows gets asked a second time — the “ask once, reuse everywhere” principle in its most literal form.
The assessment itself, staged as inherent → controls → residual so the reasoning is legible rather than reduced to a final number. Each control carries its type, effectiveness and last test result, so a reviewer can see why the residual score landed where it did.
Designing the assistant, not just using one. It answers in specifics, cites the control IDs behind the answer, and routes its output to a reviewable draft instead of asserting a conclusion — because in a regulated workspace an unsourced AI answer is worse than no answer at all.
Provenance made visible. The issue shows which regulatory change raised it, which policy and control it touches, who owns it and how far past due it is. The activity rail is the evidentiary record governance asked for — a live trail, not a static report.
The output an examiner actually reads — and the proof the flow holds together. Every figure here is drawn from records seen on the earlier screens: the same residual score from the RCSA, the same ineffective control, the same overdue issue. Nothing is re-keyed, so there is nothing to reconcile.
AI in the Workflow
I used Claude AI, Figma AI, and Microsoft Copilot throughout this project. Not as a novelty, and not as a replacement for judgment — as a way to compress the slow, mechanical parts of the process so more of my time went to the decisions that actually needed a designer.
Research synthesis — Claude AI. Twenty-plus stakeholder interviews across multiple Lines of Business generated a large volume of transcript. I used Claude to produce first-pass thematic clusters and surface contradictions between what product partners and risk partners were describing — then read every transcript myself to confirm, correct, and weight the themes. The AI accelerated the sorting; the interpretation stayed mine.
Concept exploration — Figma AI. For high-branch-count surfaces like the risk-assessment flow, I used Figma AI to generate layout and component variations quickly, so I could evaluate five directions in the time it would have taken to draw two. Every variation was then rebuilt by hand against the design system — the generated output was a thinking tool, never the shipped artifact.
Workflow documentation — Microsoft Copilot. The terminology conflict between product and risk teams meant a lot of decisions needed to be written down precisely and circulated. Copilot drafted the first version of workflow documentation and decision records from my notes, which I then edited for accuracy before they went to stakeholders.
Prototype iteration. Between usability rounds, AI-assisted iteration let me turn feedback into a revised, testable prototype faster — shortening the loop between what a participant said and the next version they could react to.
Where I kept AI out. Accessibility decisions, risk-taxonomy language, and anything a compliance reviewer would need to defend. In a regulated environment, "the tool suggested it" is not a rationale that survives scrutiny — those calls were made deliberately, documented, and owned.
Impact
Early post-launch feedback and platform usage data showed meaningful improvements on the dimensions that mattered most to the program — speed, completeness, and visibility.
↓ ~40%
Reduction in overall cycle time — validated through usage data on the redesigned submission workflow
↓ ~55%
Reduction in redundant data entry — initiators no longer re-enter information the firm already holds
↓ Marked
Drop in status-check support requests as live approval-tracking made "where is my submission?" answerable at a glance
Key Learnings
Consolidation is a UX problem before it's a tech problem. The technical lift of replacing legacy systems was significant — but the harder work was aligning four LOBs on a single vocabulary, a single risk taxonomy, and a single mental model of "what counts as an approval."
Standardization succeeds when it stays role-aware. A firmwide standard fails the moment it ignores the fact that initiators, reviewers, and risk SMEs experience the same data differently. The win was a single underlying model with role-tailored surfaces.
Visibility is the highest-leverage feature in any workflow tool. "Where is my submission?" was a more frequent and emotional question than any approval logic. Investing in clear status communication paid back across every role on the platform.
Owning the full lifecycle is what made the research stick. Because I carried the work from interview transcript to shipped screen, nothing was lost in translation between a finding and the design that answered it. The insights that usually die in a handoff deck stayed alive all the way into the build.
AI is leverage on the mechanical work, not the judgment. It compressed synthesis, exploration, and documentation — which bought me more time on the decisions that genuinely needed a designer. Knowing which half was which turned out to be the actual skill.