Persona memory. Market intelligence. Portfolio context. Risk. Evidence. Experts. Reports. One continuous decision thread.
The analyst’s problem was not access to information. It was the distance between information and judgement: separate conversations, separate research, separate charts, separate risk tools, separate people — all asking the user to reconstruct context by hand.
Every switch creates a small loss of context. Every manual reconstruction slows the next decision.
One continuous decision thread can pull in tools, evidence, memory and people without resetting the user.
Instead of asking the analyst to decide which tool to open next, the environment decides which capability, evidence, person or visualization belongs next to the question — while keeping the analyst in control.
Across the supplied product behaviour, one pattern kept appearing: context existed, but it lived in different places. The user had to carry the client objective from a conversation into research, remember the mandate while checking the market, rebuild the story for a specialist, then rebuild it again for a report. We treated those transitions as the real product surface.
The declining bars are a conceptual representation of context loss across handoffs, not measured client telemetry.
We combined domain research with the behaviour implied by the product screens to turn a large feature set into five concrete design problems.
The problem is not missing data; it is reconstructing the same context across tools.
7→1conceptual workflow compressionFinance requires evidence, counter-evidence, source visibility and human accountability.
Why?before BUY / HOLD / SELLAnalysts need depth; RMs need client relevance; CXOs need consequence; managers need exceptions.
6distinct role modesThe right design does not remove people. It knows when to bring one in without losing context.
1 tapcontext-preserving escalationPins, previous research, reports and unresolved questions should remain part of the decision thread.
∞continuing decision memoryRepresentative JTBD statements based on the product intent you supplied. These are design synthesis, not verbatim interview quotes.
“When a stock looks attractive, show me what agrees, what disagrees and what could break the thesis before I recommend it.”
EVIDENCE → JUDGEMENT“Before I call the client, tell me what changed, why it matters to *their* portfolio and which explanation will be easiest to defend.”
PORTFOLIO → TALKING POINT“Don’t show me every analysis. Show me where confidence is low, risk is rising or a human review is overdue.”
TEAM → INTERVENTION“Tell me the three shifts that can change revenue, client risk or strategic exposure — and let me drill down only when I need to.”
SIGNAL → CONSEQUENCE“I need to know not only what looks good, but whether it is good *for this client* under this mandate.”
OPPORTUNITY → FIT“Remember the report, stocks and questions I was working on so I can continue instead of rebuilding the session.”
MEMORY → MOMENTUMThe evaluation matrix below makes the product direction explicit: a chatbot maximised simplicity but hid capabilities; a mega-dashboard exposed everything but broke role relevance; the environment model let the interface assemble itself around the active decision.
Role, client, mandate and open questions survive every transition.
01Start with the next move; expand into evidence only when needed.
02Plugins are discoverable without turning the screen into a toolbox.
03Specialist escalation is part of the product architecture, not an escape hatch.
04The journey moved from “AI chat” toward a role-aware intelligence operating layer. We tested each idea against one question: does this reduce the distance between evidence and a responsible next move?
It made the interface simple, but made the analyst remember what the system could do. Capability discoverability collapsed.
Dense information was useful to analysts but wrong for RMs, managers and CXOs. One screen could not represent six different jobs.
needs evidence depth + comparison
needs client context + talking points
needs consequence + exceptions only
The role, client, mandate, permissions and risk appetite should travel with the user, changing the tools and output without asking them to configure every session.
Charts, quant, prediction, risk and voice remain visible as capabilities, but the analyst can let the system orchestrate them or intervene manually.
An answer can become a graph, table, pinned security, report, risk check, specialist conversation or future monitoring rule. Conversation becomes the beginning of work, not the end.
This is the dedicated operating screen you asked for: multiple conversations with recognizable job icons, a rich live output in the center, and connected plugin/system intelligence on the right. Click chats and systems to change state.
Synthesising market, company, portfolio and client context
Auto leadership is strengthening, but the portfolio already carries concentration risk elsewhere. Tata Motors becomes interesting only if it improves diversification while staying inside the client’s 8% single-stock boundary.
The visual language changes with the analytical job. No repeated chart grammar. Every graph answers a different decision question.
X = client fit. Y = expected opportunity. Bubble size = portfolio consequence. Border colour = dominant risk state. Hover a candidate to compare the trade-off.
Raw evidence does not jump directly to BUY. It passes through three gates: company truth, market confirmation and client suitability. The braid makes convergence and rejection visible.
The centre line is the current conviction path. The fan shows uncertainty widening through future catalysts. Decision thresholds make it obvious when the thesis becomes HOLD or exits the client mandate.
Instead of only explaining the current answer, the system lets the analyst stress the recommendation. Move the inputs and the score, state and explanation update.
The strongest product opportunity is not “more AI.” It is coordinating AI, people and evidence around the decisions where capacity and judgement matter most.
Microsoft’s 2025 Work Trend Index describes a workforce where demand is outpacing human capacity.
The market evidence reinforced our direction: the opportunity is not another isolated AI tool. It is a governed workflow where human judgement, agents, evidence and role context can work together.
McKinsey’s 2025 global AI survey found most organizations still early in enterprise scaling even as agent experimentation grows. Microsoft’s 2026 research shows advanced users are getting more high-value work from AI, but organizational alignment remains uneven.
CFA Institute’s 2025 research repeatedly points toward human judgement, explainability, governance and workflow redesign as core requirements for AI in high-stakes financial decision-making.
A CFA Institute report citing NVIDIA’s 2025 financial-services survey notes 57% of respondents were using or considering AI for data analytics.
The same cited survey reported generative AI usage at 52%, up from 40% in 2023 — increasing the need to make AI contribution visible and governed.
CFA Institute’s explainable-AI guidance stresses stakeholder-specific explanations, real-time transparency and human–AI collaboration for high-stakes financial decisions.
We treated the flow as an operating system, not a chat response. Every step can become a chart, table, specialist plugin, report, saved decision or human conversation.
Role, client, mandate, risk appetite, permissions.
01Research, compare, explain, report, monitor, decide.
02Voice, charts, quant, prediction, risk, custom tools.
03Sources, calculations, market context, internal notes.
04Expert, adviser, RM, manager or reviewer when needed.
05Pins, reports, next actions and continuing context.
06Each widget has a clear analytical job, detailed axes, labels, time context and drill-down affordances. No decorative graphs.
We designed the handoff so an analyst can call, chat or invite a specialist without rewriting the problem. The brief travels with the conversation.
Availability, expertise match, relationship ownership and previous coverage are all used to rank who should join.
Generated from active thread · no context re-entry required
Each behavior gets a distinct motion language so analysts understand what the system is doing without reading an instruction.
The product should prove whether analysts spend less effort reconstructing context, reach evidence faster, involve the right human sooner and reuse prior intelligence more effectively.
Decision cycle time
Question → evidence inspected → next action. Segment by analyst, RM, manager and CXO.
Context reuse
% of decisions that reuse persona memory, pins, prior reports or existing specialist context instead of restarting.
Depth of useful action
Answer → source / chart / compare / report / alert / human handoff. This reveals whether intelligence actually moves work forward.
The product connects people, evidence, models and workflows so employees move from “where do I look?” to “what do I know, who should I involve, and what should happen next?”
The decision brief, active persona, evidence graph and unresolved questions would travel with this handoff so the next person does not start from zero.