Work / Case Study 01
Ernst & Young · GenAI · Risk & Audit

Redesigning GenIAus — an AI auditing tool that engineers built and auditors couldn't use.

A live, beta-stage GenAI tool inside EY's Virtual Internal Auditor platform was failing adoption — not because the AI was wrong, but because the interface couldn't communicate what the AI could do. I led the redesign as a systems thinking exercise, not a UI cleanup.

ClientErnst & Young
RoleUX Designer · Lead
TeamPM · PO · 8 Developers
Year2023
−40%Task completion time
+35%Adoption rate post-iteration
+84%Time spent on platform
01 — The Problem

GenIAus was built by engineers. It showed.

When my team inherited GenIAus, it was already live in beta — a GenAI chat tool embedded inside EY's internal audit platform (VIA). On paper, it could answer auditor questions, analyze documents, and automate parts of the audit workflow. In practice, it was a collection of disconnected features stitched together without any design rationale.

The interface had no hierarchy. Navigation was misplaced — a global feature buried in a local sidebar. There were no suggested prompts, no onboarding, no visibility into what the system could actually do. Auditors were expected to figure it out by trial and error in a domain where mistakes carry regulatory weight.

The business problem was clear: EY had invested in building an AI tool for audit teams, but adoption was stalling because the people it was built for couldn't use it effectively. The gap wasn't in the AI's capabilities — it was in how those capabilities were surfaced, sequenced, and communicated.

The user problem ran deeper: Financial audit managers juggle hundreds of documents across multiple standards, phases, and review cycles. They needed a tool that reduced cognitive load, not one that added to it. The beta did the opposite — it required users to already understand what the AI could do before they could get any value from it.

02 — What I Walked Into

A rescue operation, not a greenfield project.

Eight states from the engineer-built beta — annotated during contextual analysis. Yellow sticky notes mark every usability issue: misplaced navigation, missing prompts, broken scroll behavior, no system status, no chat history, no pagination, no copy. Every Nielsen heuristic violated, and the violations compounded each other. Step through each frame to see the failure mode.

GenIAus wasn't a greenfield project. It was a rescue operation with constraints on every side: a pre-built engineering-first UI, beta-stage instability, no AI design system to inherit from, a complex regulated domain with diverse users, and scalability pressure to integrate with VIA's broader ecosystem. Every design decision had to be defensible against what was already shipped to real users.

03 — My Approach

Systems thinking, not screen design.

I made a deliberate choice early: this would not be a UI cleanup. The problems were structural, not cosmetic. I re-framed GenIAus from a collection of AI features into a risk-aware human–AI decision system — built on reusable interaction patterns, contextual AI behaviors, and embedded transparency.

01 Layer 01 End-to-end workflow redesign 02 Layer 02 Context-aware AI interactions 03 Layer 03 Human–AI collaboration model 04 Layer 04 Scalable IA & design patterns 05 Layer 05 Ethics & accessibility as system rules Each layer reinforces every other
Fig 09The systems thinking framework I used to approach GenIAus — five interconnected layers that ensured every design decision reinforced the whole, not just individual screens.
01
End-to-end workflow redesign

Mapping the full auditor journey, not just the chat interface. Designing entry points that matched real task flows.

02
Context-aware AI interactions

AI responses, prompts, and capabilities adapt based on what the user is doing and which audit phase they're in.

03
Scalable IA & design patterns

A modular architecture that accommodates new features (analytics, issue tracking) without restructuring the whole interface.

04
Human–AI collaboration model

When the AI leads, when the human leads, how confidence is communicated, how user agency is maintained.

05
Ethics & accessibility as system rules

Transparency disclaimers, AI limitation disclosures, and accessibility — embedded as constraints, not afterthoughts.

In regulated domains, users won't adopt tools they can't explain to a regulator. Every design decision had to answer the question: "Could an auditor justify this to their oversight board?"

Input New workflows, prioritized features, & design system. Translated through systems- level redesign into five measurable outcome pillars. Outcomes · 5 pillars 01 User-friendly interface Approachable for non-technical users; built around how audit and risk teams actually work. 02 Time-saving efficiency Smart Quickstart and prompt scaffolding cut analysis time from hours to minutes. 03 Highest standard of data security & privacy Internal-only deployment; persona-scoped access; transparent data lineage. 04 Contextual understanding Persona, prompt, and document grounding so outputs reflect real audit context. 05 AI-powered document analysis Inline summaries, evidence extraction, and cross-document synthesis at scale.
Fig 10Transforming GenIAus for Success — mapping new workflows, prioritized features, and the design system to five outcome pillars: user-friendly interface, time-saving efficiency, data security compliance, contextual understanding, and AI-powered document analysis.
04 — The Design Sprint

Five days from analysis to clickable prototype.

A structured 5-day design sprint moved the work from analysis to validated structure — divergent before convergent, lo-fi before hi-fi, structure before surface. The carousel below cycles through the sprint's working materials; it advances on its own, but pause to study a frame any time.

Day 1–2 · Understand & Define. Contextual analysis of the beta. Stakeholder interviews to map pain points, workflows, and expectations. Defined the core problem statement: "How might we redesign GenIAus so that its AI capabilities are intuitively understood and seamlessly integrated into the auditor's existing workflow?"

Day 3 · Ideate. Generated multiple flow variations on Miro — exploring different approaches to onboarding, file selection, prompt scaffolding, and capability communication. The key tension: how much structure to impose on a conversational interface without killing the flexibility that makes AI tools useful.

Day 4 · Decide. Evaluated iterations against three criteria: does it reduce time-to-first-value, does it scale to future features, can it be built incrementally without a full rewrite. Selected the flow that balanced all three.

Day 5 · Prototype. Built a clickable prototype in Figma — lo-fi first to validate structure, then hi-fi to test visual hierarchy and interaction patterns.

05 — Research That Shaped Decisions

Four insights that changed the design.

After the sprint, I led usability testing with audit professionals — the actual end users — using a moderated think-aloud protocol focused on Document Intelligence and Prompt Scaffolding. Insights were synthesized through thematic analysis, empathy mapping, and affinity mapping.

Labeling mattered more than we expected. Users consistently misunderstood feature names inherited from the engineering team. "Find in Files" didn't communicate what the feature actually did. Renaming to "Document Intelligence" with clear sub-descriptions immediately improved comprehension.

Users needed to see AI capabilities before they could use them. The empty chat state was a dead end. Adding structured entry points — Ask me anything, Document Intelligence, Q&A in Issue Tracker, Q&A in Analytics — gave users a mental model of the system's scope before they typed a single prompt.

File selection needed progressive disclosure. Showing 80+ audit files in a flat list overwhelmed users. Introducing an audit name dropdown first, then revealing the filtered file list with pagination and search, dramatically reduced decision fatigue.

Trust required transparency. Users in a regulated domain needed to know the AI's limitations upfront. Adding the disclaimer — that insights should serve as reference points, not definitive answers — wasn't just legal compliance. It was a trust-building mechanism that made users more willing to engage with the tool.

06 — Key Design Decisions

Eight features. Eight specific problems.

Each redesigned feature was a deliberate response to a specific insight — the problem, the design decision, and the reasoning behind it.

01
Smart Quickstart — getting users productive without confusion.

The Problem

The beta dropped users into an empty chat with no guidance. Users with low AI literacy didn't know where to start. Users with high AI literacy didn't know what this specific AI could do.

The Decision

A landing state that doubles as onboarding and navigation. Four capability entry points — Ask me anything, Document Intelligence, Q&A in Issue Tracker, Q&A in Analytics — each with a clear icon and label. Below: a transparency disclaimer that sets expectations about AI limitations.

Why It Works

It answers the three questions every new user has — what is this, what can it do, what should I do first — without requiring them to read documentation or watch a tutorial.

Smart Quickstart — landing state with four capability entry points
Fig 17Smart Quickstart — the redesigned landing state. Four clear capability entry points replace the blank chat, and the disclaimer below sets expectations about AI limitations upfront.
02
Prompt Scaffolding — making AI interactions purposeful.

The Problem

Users didn't know how to prompt the AI effectively. Enterprise users aren't ChatGPT power users — they needed guidance on what to ask and how to ask it.

The Decision

Persistent contextual prompt suggestions surfaced at the bottom of the chat — tailored to the active feature mode and updated based on context. Tap a suggestion to start, then refine.

Why It Works

It transforms the chat from a blank text field into a guided interaction. The barrier to first interaction drops from "compose a prompt from scratch" to "select and customize."

Prompt scaffolding — contextual suggestion chips beneath the input
Fig 18Prompt scaffolding in action. Contextual suggestion chips beneath the input field reduce the blank-page problem and teach users the system's language through use.
03
Conversation History — continuity and confidence.

The Problem

The beta had no memory. Every session started fresh. Auditors working across multi-week engagements couldn't reference previous AI interactions, forcing them to repeat queries and lose context.

The Decision

A collapsible sidebar showing recent chats (last 10 conversations) with preview text and the ability to resume any previous thread. New Chat button prominently placed for starting fresh.

Why It Works

It gives auditors the continuity they expect from any modern tool — and builds confidence. Users can see that their work persists, which increases willingness to invest time in detailed queries.

Conversation history sidebar with recent chats and New Chat button
Fig 19Conversation history panel — recent chats are always accessible, giving auditors continuity across sessions and confidence that their work persists.
04
Document Intelligence — working with your files.

The Problem

The beta's file selection was a flat, unsearchable list of every file in the audit. No filtering, no pagination, no feedback on selection.

The Decision

A two-tab approach — "Use existing files" and "Upload a new file" — within a dedicated Document Intelligence section. Existing files use progressive disclosure (audit name first, then paginated list). Upload supports drag-and-drop with clear constraints. Step through the carousel to see all three states.

Why It Works

It separates two fundamentally different user intents (work with what's already in the system vs. bring in something new) and optimizes each path independently.

05
Smarter data tables — focused review.

The Problem

The beta showed all files in a single scrolling list with no way to filter, search, or paginate. Users doing focused analysis had to visually scan every row.

The Decision

Redesigned tables with search, pagination, column headers (Name, Location, Source, Action), and inline actions (Generate Summary). A "Continue to Q&A" button provides a clear exit point once files are selected.

Why It Works

It respects the user's time. An auditor reviewing a specific standards-mapping document shouldn't need to scroll past 60 irrelevant files to find it.

Smarter data tables with search, pagination, and inline actions
Fig 20Redesigned data tables with search, pagination, and inline Generate Summary actions — turning a wall of files into a focused review experience.
06
Visibility of system status — no more guessing.

The Problem

The beta gave zero feedback during processing. Users clicked "Proceed" and stared at an unchanged screen, unsure if the system was working, broken, or finished.

The Decision

A full-screen loading state with progress indicator and clear messaging. Background content visually dimmed to prevent interaction during processing. Toast notification confirms completion.

Why It Works

It's Nielsen's #1 heuristic for a reason. In an enterprise context where processing can take 30+ seconds, silence feels like failure. Visible progress maintains user trust and prevents duplicate submissions.

Visibility of system status — loading state with progress bar and toast notification
Fig 21System status visibility — loading state with progress bar over a dimmed background, toast confirming generation. Replaces the beta's silent processing and maintains trust during longer operations.
07
Persona selection — tailored AI perspectives.

The Problem

Different stakeholders — audit managers, compliance officers, CROs, IT risk specialists — need different types of analysis from the same data. A one-size-fits-all AI response style served no one well.

The Decision

A persona modal accessible from the chat. Four personas with clear focus-area descriptions; selection changes the AI's response style and analytical lens mid-conversation. Persona prefix ("AM:") confirms which lens is active.

Why It Works

It's a UX solution to a prompt engineering problem. Instead of expecting users to frame queries differently for different perspectives, the interface handles the context-switching.

Persona selection modal with four analytical lenses
Fig 22Persona selection — users choose an analytical lens (Audit Manager, Compliance Officer, CRO, IT/Cyber Risk) that tailors GenIAus's response style. The persona prefix in the input field confirms which lens is active.
08
Prompt Library — institutional knowledge at your fingertips.

The Problem

Even with prompt scaffolding, advanced users wanted access to proven, reusable prompts. The organization wanted to standardize how teams interacted with the AI to ensure consistency.

The Decision

A searchable Prompt Collection modal with two tabs: "My Prompts" (personal, 7/25 used) and "System Prompts" (organization-wide, 25 available). Domain-specific, ready to run.

Why It Works

It bridges individual productivity and organizational standardization. Users get a head start on complex queries; the org gets consistency in how AI-assisted analysis is conducted across teams.

Prompt library — My Prompts and System Prompts tabs with searchable list
Fig 23Prompt library — personal and system-level prompts, searchable, domain-specific. Bridges individual productivity with organizational consistency.
07 — Collaborating With Engineers

Not a handoff. A shared problem.

Redesigning a live product with 8 developers required a different kind of collaboration. Here's what worked.

Designing within technical reality. I sat with the engineering team to understand API constraints, processing limitations, and backend architecture before proposing interface changes. Every design decision was pressure-tested against what was actually buildable within the current sprint cycle.

Shared problem framing. Instead of presenting solutions, I brought the usability findings to engineering as shared problems. "Users can't find their files" is a problem everyone owns. "Add pagination to the file table" is a prescription only design owns. The former gets buy-in; the latter gets resistance.

Incremental refactoring strategy. We couldn't rebuild everything at once. I prioritized changes by impact-to-effort ratio: prompt scaffolding and onboarding shipped in Sprint 1 with minimal backend changes; Document Intelligence restructuring required API changes and shipped in Sprint 2; persona selection needed backend persona configuration and shipped in Sprint 3.

Design artifacts as alignment tools. The Miro boards, annotated screenshots, and lo-fi wireframes weren't just design process — they were communication tools. Engineers could see the rationale behind every decision, which reduced rework and misinterpretation during implementation.

08 — Impact

From a beta auditors avoided to a tool they requested.

−40%Task completion time
across the redesigned flows
+35%Adoption rate
after iteration ship
+84%Time spent on platform
(deeper engagement, not friction)

Beyond the numbers: GenIAus moved from a beta experiment that audit teams avoided to a tool they actively requested access to. The design patterns established in this project — prompt scaffolding, persona-based AI interactions, progressive file disclosure — became the foundation for EY's broader AI product design standards.

09 — Reflection

What I'd do differently. What it taught me.

What I'd do differently

More upfront co-design with auditors. The 5-day sprint was effective for speed, but I'd have benefited from embedding with an audit team for a full engagement cycle before designing. Some workflow nuances only surface when you watch someone do the actual work over days, not hours.

Earlier investment in a design system. I defined patterns as I went, but a dedicated sprint for establishing the AI interaction design system upfront would have reduced inconsistencies and accelerated later feature design.

Push harder on analytics integration. The Q&A in Analytics feature was scoped but not fully designed in my tenure. If I were continuing, that's where I'd focus — audit analytics is where the highest-value AI use cases live.

What it taught me

Transparency and guidance aren't nice-to-haves in AI products — they're structural requirements. In regulated domains, users won't adopt tools they can't explain to a regulator. Every design decision had to answer the question: "Could an auditor justify this to their oversight board?"

Context retention improves efficiency. Conversation history, file selection persistence, and persona memory aren't features — they're respect for the user's time.

The best AI interface design is invisible. When GenIAus works well, auditors don't think about the interface. They think about the audit. That's the goal.

Next Case Study Scaling a Wealth Platform Without Breaking Trust →
Let's Connect

Have a project in mind?

I'm open to senior product design roles, advisory work, and selective collaborations. Whether you have a defined brief or a fuzzy problem space, let's talk it through.

damleaalvee@gmail.com
Let's
create.