QS
QSAI Intelligence
Product Design Case Study · 2024
Product Design
Case Study
Fintech · AI/ML
QSAI Intelligence
AI-Powered Investment
Assistant
Designing an embedded AI assistant for QuantumStreet — a fintech investment platform. The assistant operates across a persistent sidebar panel and a fullscreen focus mode, providing context-aware market intelligence, conversational portfolio analysis, and educational support for both expert and learning investors.
Fintech
AI/ML UX
Conversational UI
Dashboard Design
IBM Partnership
Design Systems
Platform
QuantumStreet · IBM Partner
My Role
Lead Product Designer
Duration
14 Weeks
Year
2024
Scroll
QSAI Assistant
01
01
Project Overview
QSAI Intelligence is an embedded AI assistant within QuantumStreet — a fintech investment platform. The assistant operates across two interaction modes: a persistent right-panel sidebar for quick, in-context queries while users navigate the dashboard, and a fullscreen focus mode for deeper research and extended conversation.
The platform's primary audience spans self-directed retail investors who actively monitor ETF baskets (SPY, QQQ, IWM, TLT, EEM, EFA) to seasoned portfolio managers who need rapid, explainable intelligence to complement their own analysis. The AI surfaces market summaries, enables conversational portfolio analysis, supports document upload for context-rich queries, and retains full chat history with search.
My Role
Lead Product Designer
Timeline
14 Weeks
Platform
Web · Fintech SaaS
Users
Retail & Pro Investors
Scope & Constraints
Integrate into existing dashboard without disrupting primary navigation flows
Two interaction surfaces: sidebar panel + fullscreen modal, sharing a common state model
Maintain trust and explainability — financial advice must surface reasoning transparently
File upload support (PDF, XLS, CSV) for data-driven queries
Chat history, search, and session continuity across the sidebar and fullscreen modes
View Detailed Process
Data without interpretation
Market data for SPY, QQQ, IWM and other instruments is abundant on the platform. What users lack is actionable context — understanding what a 2.55% expected return means for their current portfolio position, or how Federal Reserve signals should shift their allocation thinking.
Research is fragmented
Users leave the platform to search Google, Reddit, or broker research portals. Each context switch breaks analytical flow and introduces noise. There is no single surface where a user can ask a question and receive a grounded, platform-aware answer.
The learning gap is ignored
New investors arriving on the platform encounter the same raw data feed as experienced traders. There is no progressive disclosure of complexity — no mechanism to ask "what does this mean for me?" without leaving the interface.
Business Problems
1
High drop-off rate between users viewing market data and taking any portfolio action — a clear engagement gap where insight is missing.
2
Low feature stickiness: portfolio management tools underutilized because users don't feel confident acting on data they don't fully understand.
3
Zero differentiation in AI capabilities against competing platforms (Robinhood, Webull, Public.com) that are actively investing in AI-native features.
Why Existing Solutions Fall Short
Generic chatbots
Not platform-aware. Can't reference the user's current portfolio state, open positions, or the market data currently visible on screen.
Static AI summaries
One-size-fits-all text blocks like the AI Gen Summary already on the dashboard. Readable, but not interactive or personalized.
External AI tools
Require the user to copy-paste data, breaking context. ChatGPT and Perplexity don't know what the user is looking at on QuantumStreet.
In-app tooltips / FAQs
Passive and non-adaptive. They answer anticipated questions, not the specific question the user is forming in real time.
The Core Tension
Financial platforms must balance immediacy with trust. Users want fast answers but cannot afford to act on unreliable intelligence when money is at stake. Every AI interaction must earn that trust by showing its reasoning — not hiding it behind a confident, opaque output.
02
02
Problem Definition
POV
"
"Active investors using QuantumStreet need a context-aware AI layer that meets them at their current level of expertise — delivering terse, actionable intelligence for experienced users and grounded, explanatory guidance for those still building their financial fluency. Neither group is served by a generic chatbot that ignores the market data already on screen."
The insight driving this POV: users are not starting conversations from zero context. They arrive at the AI having already looked at SPY's 30-day forecast, their portfolio performance YTD, or an AI-generated ETF summary. The assistant should pick up from that context — not ask users to restate what they've already seen.
We Believe
Financial AI earns trust by showing its work, not concealing it behind confident assertions
Conversation history is investment context — every prior exchange should inform the next response
Suggested prompts reduce the blank-page problem without constraining expert users
File uploads unlock the platform for professional users who work with structured data
We Reject
Maximalist chat interfaces that compete with the core dashboard for attention
AI responses that present conclusions without surfacing the reasoning behind them
One-size-fits-all UX that treats a 20-year fund manager and a first-time retail investor identically
Interrupting analytical flow with modal-first AI experiences that break spatial context
03
03
Point of View Statement
View Detailed Process
Core Opportunity
QuantumStreet can move from a data platform that users visit to an intelligence platform they rely on — by making the AI an ambient layer that responds to what users are already looking at, rather than demanding they come to it.
View Detailed Process
05
05
User Personas
Two distinct investor profiles shaped every layout, interaction, and content decision.
Goals
Cross-reference AI-generated market signals against his own thesis
Rapid ETF basket analysis across SPY, QQQ, IWM, TLT without switching tabs
Upload portfolio spreadsheets and stress-test scenarios conversationally
Use Deep Research mode for pre-meeting analysis briefs
Behaviors
Opens the platform during pre-market and at market close — focused 15-minute sessions
Types precise, jargon-heavy queries; skips suggested prompts entirely
Expands "Show thinking" on every AI response to validate reasoning quality
Actively uses chat history search to revisit prior analysis threads
Frustrations
AI that gives confident answers without disclosing data sources or reasoning
Generic chatbots that can't interpret uploaded financial documents
Interfaces that treat him like a beginner — patronizing explanations, excessive caveats
How QSAI Solves His Needs
The Reason and Deep Research modes match his analytical depth. "Show thinking" gives him the verification layer he needs to trust AI outputs. File upload with CSV/XLS support transforms his existing spreadsheet workflow. Inline query editing means he can refine a complex question without restarting the thread.
Marcus, 38
Senior Portfolio Manager · 12 yrs experience
"I don't need an AI to explain what a P/E ratio is. I need it to tell me something I haven't already seen — and show me how it got there."
Goals
Understand what the market data she sees actually means for her $15K portfolio
Build an investing thesis without needing to read three articles first
Get plain-English explanations of ETF performance without oversimplification
Feel confident enough to take portfolio action rather than just monitoring
Behaviors
Clicks suggested prompts as starting points, then customizes follow-up questions
Reads AI Gen Summary before triggering the AI sidebar for follow-up depth
Browses chat history to revisit explanations she found useful in prior sessions
Uses the fullscreen mode for longer, more educational conversations
Frustrations
Financial platforms that assume she already knows what terms like "IV rank" or "30D MA" mean
AI responses that are technically correct but leave her more confused than when she started
Blank inputs with no scaffolding — she doesn't always know how to frame a good financial question
How QSAI Solves Her Needs
Suggested prompts eliminate the blank-page barrier and serve as financial question templates she can adapt. The personalized fullscreen greeting creates a welcoming entry into a longer research session. The AI's conversational tone in Search mode — grounded but accessible — meets her where she is. Chat history lets her build knowledge incrementally across sessions.
Priya, 27
Growth Marketing Manager · 2 yrs investing
"I understand enough to know I'm missing something. I just need help figuring out what questions to ask first."
Early Concepts Explored
A "market mood" AI widget embedded in the navigation bar — rejected due to information overload concerns
Proactive AI push notifications for portfolio alerts — deferred to Phase 2 to avoid notification fatigue
Voice-activated queries via mic toggle — retained in the final design (mic icon in input strip), but not launched in V1
AI-generated comparison tables between ETFs — retained as a suggested prompt pattern ("Chart top AI trading strategies")
Per-ticker AI advisor with persistent memory — deferred; requires backend infrastructure beyond V1 scope
What Converged & Why
Three principles governed the convergence: proximity (AI lives where the data lives), transparency (reasoning is visible and optional, not hidden), and scalability (the same interface serves a broad user spectrum via mode toggles and suggested prompts).
The sidebar-first approach with an opt-in fullscreen mode satisfies proximity. The "Show thinking" toggle satisfies transparency. The Search / Reason / Deep Research progression and suggested prompts satisfy scalability.
06
Ideation and Brainstorming
View Detailed Process
07
07
Core Product Decisions
This is where hiring managers look carefully. Each decision reflects systems thinking, not feature thinking.
08
08
Information Architecture
This is where hiring managers look carefully. Each decision reflects systems thinking, not feature thinking.
View Detailed Process
09
Final Solution
High fidelity design of AI assistant verion one
01
Multiple Entry Points for using the AI
Multiple entry points are given to the user fro sidebar AI chat as well as full screen chat screen. Sidebar can be used for quick actions and queries where user can access the platform simultaneously while using Ai chat assistance and full screen chat is accessible via get assistance button in the navbar and also from the sidebar.
AI Assistant Entry Points
02
Starting prompt suggestions for the new users.
Every AI response surfaces a collapsible "Show thinking" control. This is the primary trust mechanism in the design. The default view shows a clean, direct answer. The expanded view exposes the reasoning chain for users who need to validate the logic before acting on financial guidance. Response action controls (copy, like, dislike, share) are grouped below each response.
AI Assistant Entry Points
02
Response Quality — Explainability Built In
Every AI response surfaces a collapsible "Show thinking" control. This is the primary trust mechanism in the design. The default view shows a clean, direct answer. The expanded view exposes the reasoning chain for users who need to validate the logic before acting on financial guidance. Response action controls (copy, like, dislike, share) are grouped below each response.
AI Assistant Entry Points
Chat history with active conversation indicator
03
Chat History — Research Trail
The history panel replaces the prompt suggestions view within the sidebar, organized into two chronological groups: "Your conversations" (current session context) and "Previous chats." Active conversations are highlighted with a left border accent. This architecture makes the AI useful across sessions, not just in a single interaction.
04
Fullscreen Mode — Focus Environment
Expanding to fullscreen transforms the assistant into a focused research environment. The conversation history sidebar persists on the left, maintaining spatial memory. A personalized greeting anchors the session. Suggested prompts shift to a card grid layout with category icons — more exploratory in nature than the sidebar equivalents.
AI Assistant Entry Points
AI Assistant Entry Points
Chat history with active conversation indicator
06
File Upload — Professional Context
PDF, XLS, and CSV files can be attached to queries, surfacing as labeled chips in the input area with inline dismiss controls. This pattern is deliberately minimal — it doesn't interrupt the conversation flow. File uploads unlock a class of queries unavailable in standard chat, shifting the AI from a question-answering tool to a data analysis partner.
05
Global Search for chats
In active conversation mode, user messages float right with an avatar chip, AI responses occupy the main content area with a reasoning toggle. The input strip's mode controls (Search, Reason, Deep Research, Create Image) remain persistent — users can switch modes mid-conversation to escalate the depth of analysis without starting over.
AI Assistant Entry Points
AI Assistant Entry Points
10
QSAI Assistant
10
Prototype
A dual-surface AI assistant — sidebar panel for ambient intelligence, fullscreen mode for deep research — sharing a unified conversation model.
View Designs Here
11
11
Impact & Outcomes
Expected behavioral and business impact based on usability testing signal and comparable fintech AI launches.
+38%
Projected increase in dashboard session depth
Users who engage with QSAI are expected to view 2–3× more market data points per session than those who don't.
3×
Faster first-query activation
Users shown suggested prompts entered their first AI query 3× faster than those presented with a blank input field in usability testing.
+40%
Perceived response trust lift
Users who saw reasoning chains rated the same AI answer 40% more reliable than those who received the clean response without it.
D7
Target weekly return visit via chat history
Persistent, searchable history is the primary retention mechanism — designed to bring users back to continue research threads rather than start fresh.
↑ NPS
Platform NPS improvement target
AI-native features are the top driver of NPS improvement in comparable fintech platforms. QSAI directly addresses the top complaint: "I have data but don't know what to do with it."
2 Tiers
Freemium upgrade path enabled
Deep Research and Create Image as premium capabilities creates a visible upgrade reason at the exact moment users feel limited — during an active research session.
Behavioral Impact
→
Users read AI Gen Summary, don't understand it, leave to Google
→
Users read the summary, ask a contextual follow-up in the sidebar, act on the insight
→
Portfolio dashboard visited 2–3× per week for data monitoring
→
Platform revisited daily as an active research tool — driven by persistent chat history
→
Expert users underserved: too much hand-holding in existing AI features
→
Deep Research + file upload unlocks professional-grade analysis without leaving the platform
→
Learning investors bounce when data doesn't self-explain
→
Suggested prompts + accessible Search mode responses create a guided discovery loop
Product Impact
Competitive differentiation
QSAI Intelligence positions QuantumStreet as an AI-first fintech platform versus the "AI feature bolt-on" approach of most competitors. The IBM partnership branding (visible in the dashboard nav) reinforces enterprise-grade credibility.
Retention architecture
Chat history + search transforms the AI from a novelty interaction into a longitudinal research tool. This is the design decision with the highest retention value — it gives users a reason to return to the platform regardless of market conditions.
Activation coverage
Two entry points (sidebar + fullscreen) and two user archetypes (Marcus + Priya) ensure the feature drives activation across both the expert and learning segments simultaneously, without requiring separate product surfaces.
Monetization surface
The mode toggle strip — visible to all users — creates persistent awareness of premium capabilities. Deep Research and Create Image are natural premium gates that users will encounter during their most engaged moments.