JPMorgan Chase
VP CRM Platform LLM Mobile Platform Salt Design System Creative Technology
UBS
Neo Platform Delta Analytics Asset Management & AI
HSBC
Digital Transformation Contact
Portfolio JPMorgan Chase LLM Mobile Platform
02 · Enterprise AI · 2021 — 2025

LLM Mobile
Platform

Directing end-to-end UX strategy for JPMorgan's firmwide AI assistant on mobile — translating complex enterprise AI into a secure, compliant, and intuitive experience for 200,000+ employees on the go.

UX Strategy
Mobile First
AI Interaction
Enterprise
Salt Design System
Competitive Research
User Interviews
200k+
Employees reached
#1
American Banker's 2025
VP
Digital UX Designer
2 forms
Phone & tablet
01 Problem Statement

Enterprise AI needed to
work in your pocket

JPMorgan's LLM Suite was scaling rapidly on desktop — but mobile remained an open question. For a global workforce of 200,000+ employees, many of whom operate primarily from phone and tablet while travelling, attending client meetings, or working across time zones, the absence of a well-considered mobile experience was a significant adoption gap.

The challenge wasn't simply porting a desktop product to a smaller screen. It meant rethinking navigation architecture, conversation flow, security constraints, and compliance patterns in a form factor where context-switching is constant, sessions are shorter, and cognitive overhead must be minimal.

Core Problem
How do you bring the full power of an enterprise LLM assistant to mobile — without compromising the security, compliance, and usability standards that a regulated financial institution demands — while ensuring it feels native to the way bankers actually work?
Constraints
Enterprise Guardrails
Every interaction had to operate within JPMorgan's strict data governance and compliance framework. Sensitive client information, regulatory obligations, and audit requirements shaped every design decision — from conversation storage to session management.
Compliance-first Data governance Audit trails Session security
Opportunity
First-mover Advantage
No major competitor had cracked enterprise-grade AI on mobile for financial services at this scale. Designing the interaction model well — and getting adoption right — would define the firmwide standard and set a benchmark the wider industry would follow.
Firmwide standard 200k users Industry benchmark
User Tension 01
Speed vs. Depth
Mobile users need fast, decisive answers — but complex enterprise queries often demand nuanced, multi-step responses. Designing conversation flow that felt appropriately concise without losing fidelity.
User Tension 02
Familiar vs. Novel
Consumer AI apps (ChatGPT, Gemini) had shaped expectations. Employees would arrive with mental models from personal AI use, but enterprise context required distinct trust signals and interaction patterns.
User Tension 03
Desktop Parity vs. Mobile-native
Product stakeholders wanted feature parity with desktop. Design advocacy required demonstrating where parity was the wrong goal — and where mobile-specific affordances would drive better outcomes.
02 Discovery & Research

From competitive signal
to design principle

Discovery ran in parallel across four workstreams: competitive analysis of the consumer AI landscape, an internal UX audit of existing JPMorgan patterns and the Salt Design System, user interviews with employees across Lines of Business, and a shared library review to identify reusable components versus gaps requiring new design investment.

Phase 01
🔍
Competitive Analysis
Audited ChatGPT, Gemini, Copilot, Claude, and Perplexity — cataloguing navigation patterns, conversation UX, mobile affordances, and trust signals.
Phase 02
📐
Internal Audit
Reviewed existing JPMorgan AI patterns, the Salt Design System shared library, and prior mobile work to map reuse opportunities and gaps.
Phase 03
🎤
User Interviews
Conducted interviews across key Lines of Business — understanding real mobile usage contexts, trust concerns, and unmet productivity needs.
Phase 04
🧩
Synthesis
Translated findings into design principles and a component strategy, aligning mobile patterns with the desktop LLM Suite for a coherent firmwide system.
LLM Mobile — Discovery Research.jam
✦ Move 💬 Comment 🔗 Connect 75%
ACompetitive Analysis — Consumer AI Landscape
🤖
ChatGPT Mobile
Persistent conversation history
Clear session management
Strong voice input affordance
Bottom input bar — thumb-native
No enterprise/compliance signals
No data residency transparency
Custom GPTs — model for tools
Gemini (Google)
Deep Workspace integration
Context-aware suggestions
Multimodal — images, files
Inline rendering of outputs
Consumer-grade trust framing
Cluttered information hierarchy
Extension model → relevant for tooling
Microsoft Copilot
Enterprise auth integration
Strong compliance positioning
Document context (M365)
Commercial data protection
Mobile UX lags desktop
Navigation inconsistency
Complex feature surface — overwhelming
🔶
Claude (Anthropic)
Clean, minimal conversation UI
Strong mobile reading experience
Projects — relevant mental model
Thoughtful long-response handling
Limited tool integrations
No enterprise trust signals
Perplexity
Source citations inline
Research-oriented UX — audit relevant
Speed as a core value
Progressive disclosure of detail
No enterprise security posture
Consumer-only positioning

Key signal: All consumer products optimise for speed & delight. None address enterprise trust, compliance framing, or regulated context signals — a significant white space for JPMC.

Navigation pattern consensus → bottom tab + persistent input bar. Thumb-reachable. Adopt.

BUser Interview Findings — 14 Participants Across LOBs
Investment Banking"I use ChatGPT on my phone for drafts on the train. If JPMC had this, I'd use it — but only if I knew data wasn't leaving the firm."
Asset Management"My biggest mobile pain: I start a thought on desktop and can't easily continue. Cross-device continuity is critical."
Markets / Trading"I need answers fast. 5 seconds on mobile feels like forever. Speed is trust."
Compliance Officer"I need to know what the AI can and can't access. That's not a nice-to-have — it's a requirement before I'll endorse it."
Tech & Ops"The current desktop AI tool is powerful but I can't use it when I'm away from my desk. Half my day is away from my desk."
Private Banking"Client conversations happen everywhere. I need to look something up quickly without it feeling clunky or showing something I shouldn't."
Research"If you show me a wall of text on my phone, I'll close it. Summaries first, detail on demand."
HR / Ops"I don't need all the features — I need the three things I use daily to work perfectly."

Themes: (1) Data trust as prerequisite — security framing must be visible, not assumed. (2) Cross-device continuity expected. (3) Speed as a signal of quality. (4) Progressive disclosure essential for mobile. (5) High-value use cases narrow and consistent.

CInternal UX Audit — Salt Design System & Existing AI Patterns
Salt DS — Component Status
Navigation Patterns
Bottom Tab Bar ✓ Side Sheet ✓ Contextual Menu ~ AI Mode Toggle ✗
Salt DS — Component Status
Conversation UI
Input Field (mobile) ~ Chat Bubble ✗ Typing Indicator ✗ Badge / Status ✓
Salt DS — Component Status
Trust & Compliance UI
AI Disclosure Banner ✗ Data Context Label ✗ Alert / Warning ✓ Consent Pattern ~
Existing AI Patterns — Desktop
LLM Suite Desktop
Prompt Input ✓ Model Selector ✓ Tool Calls (complex) ~ Mobile adaptation ✗
Existing AI Patterns — Desktop
History & Sessions
Session List ✓ Mobile sidebar ✗ Search in history ✓ Pin / Favourite ✗
Gap Analysis
New Components Required
AI Loader animation Response rating Context chip strip Secure data badge

Salt DS covers ~60% of mobile needs. Conversation primitives (bubbles, typing, AI disclosure) require new component designs to be contributed back to the system. This work will benefit all future AI surfaces across the firm.

DSynthesis — Design Principles
01 · Trust FirstSecurity and compliance signals must be visible and legible — never hidden in settings. Users need ambient confidence that data is protected.
02 · Speed is UXResponse latency perception shapes trust in AI. Every animation and progressive-reveal state must make the model feel fast and attentive.
03 · Progressive DepthLead with summary. Expand on demand. Never front-load cognitive load in a short-session, high-interruption mobile context.
04 · Desktop CoherenceNavigation and interaction vocabulary must feel recognisably related to the desktop LLM Suite. Not identical — but conversant.
05 · Thumb-native ArchitecturePrimary actions live in the bottom third of the screen. No critical interactions require a stretch to the top. Designed for one-handed use.

Research artefact — FigJam board: competitive analysis, user interviews & internal audit / Salt Design System gap mapping

03 Concept Designs

Conversational AI,
built for enterprise mobile

The concept designs explored a light-mode conversational interface — clean, readable, and clearly enterprise in character. The phone form factor prioritised single-focus conversation with thumb-native input. The tablet design leveraged the additional screen real estate with a persistent sidebar for conversation history and context management, while keeping the main conversation area open and uncluttered.

Both surfaces shared a unified design language aligned with Salt Design System tokens — ensuring this work could be contributed back as reusable components for future AI surfaces across the firm.

9:41

JPMC AI Assistant

Firmwide · Secure session

JP
Good morning. I'm your JPMC AI Assistant. How can I help you today?
9:38 AM
Summarise the key points from the Q3 markets outlook report
9:39 AM
JP
Here are the key takeaways from the Q3 Markets Outlook:

1. Rate environment — Fed expected to hold through Q3 with possible cut in Q4.

2. Equities — Cautious bias; defensives outperforming growth.

3. Credit — IG spreads tight; HY more vulnerable to refinancing risk.
9:39 AM
JP
Full report ↗
Compare to Q2
Email summary
🎤
Ask anything…

iPhone — Light mode chat
375 × 812 · Single-focus conversation

JPMC AI Assistant
🔍
EA
Conversations
+ New chat
Q3 Markets Outlook
Summarise the key points from…
Today, 9:39
EMEA Credit Risk Brief
What are the main credit concerns…
Yesterday
Rates strategy — Q4
Help me draft a summary for…
Mon
Equity sector rotation
Compare defensive vs growth…
Last week
Client memo draft
Can you help me write a 1-page…
Last week
Q3 Markets Outlook 🔒 Secure · Firmwide data
JP
Good morning, Elwira. I'm ready to help. What would you like to work on today?
Summarise the key points from the Q3 markets outlook report
EA
JP
Based on the Q3 Markets Outlook (Internal — Research, July 2025), here are the headline findings:

Macro & Rates
The Fed is expected to hold rates through Q3, with a 25bps cut priced in for November. Inflation is trending toward target but stickiness in services remains a key risk.

Equities
Our strategists maintain a cautious stance. Defensives (Healthcare, Utilities, Consumer Staples) are outperforming. AI-related tech remains volatile but selectively attractive.

Credit
Investment-grade spreads are near multi-year tights. High-yield faces headwinds from a wave of Q4 refinancing. Monitor energy and consumer discretionary closely.
📄 Full report
🔄 Compare Q2
✉️ Email this
💾 Save
📎
Ask a follow-up, or start a new topic…
🎤

iPad — Light mode split view
834 × 1194 · Sidebar + conversation

JPMC AI Assistant
🔒
JP
Good morning. How can I help?
Summarise Q3 markets outlook
EA
JP
Q3 Markets:
Fed holding through Q3. Defensives outperforming. Credit spreads tight — monitor HY refinancing.
Full report ↗
Compare Q2
Email
Ask anything…

iPhone — Simplified preview
Single-focus conversation

Phone Design Decisions
Single-focus, Thumb-native
Full-width conversation view removes distraction. Bottom input bar and suggested chips sit within natural thumb reach. Session security label in the header — always-on, never intrusive. Progressive response rendering prevents cognitive overload on small viewports.
Bottom input Suggested chips Security label Full-bleed chat
Tablet Design Decisions
Context-rich, Spatially Efficient
Persistent sidebar for conversation history prevents context loss during longer sessions. Wider message bubbles support richer formatting. Action row beneath AI responses — copy, draft, source — surfaces utility without cluttering the chat flow. Context chips above the input for quick follow-up navigation.
Persistent sidebar Action row Context chips Rich formatting
Impact & Outcomes

Designed, shipped,
and recognised

200k+
Employees with access to AI-assisted productivity on mobile
#1
American Banker's Innovation of the Year 2025 — Grand Prize
2
Form factors designed — phone & tablet — within a unified design system
Broader Impact
New conversation UI components contributed back to the Salt Design System, becoming reusable across all future AI surfaces at the firm — extending the value of this project beyond the LLM Mobile Platform itself. The design established firmwide standards for enterprise AI on mobile.
← Back to JPMorgan Chase
CRM Platform ↗ UBS →