11 screens designed
3 user flows
5 core features
Speculative PWM project
KYC onboarding at a private wealth management firm is one of the most regulated workflows in finance. At Goldman Sachs, onboarding a new high-net-worth client means satisfying AML and KYC requirements under strict compliance timelines, while still delivering the seamless, white-glove experience clients expect from the brand.
The current system fails all three people involved.
The Client doesn't know where they are in the process. Document requests arrive without context. Rejections come back with vague error messages. They feel like they're being processed, not onboarded.
The Advisor is stuck in the middle. They're the client's primary contact but have no real-time visibility into compliance review. They find out about problems when the client calls to complain.
Compliance is overwhelmed. Cases pile up without prioritization. When AI flags something, reviewers get a result. Not a reason. They have to trust the system blindly or dig through raw documents themselves.
This isn't a communication problem. It's an information architecture problem.
The client, advisor, and compliance officer are all operating with different views of the same process, with no shared source of truth. Each handoff is a potential failure point. Each delay compounds downstream.
For Goldman Sachs's clients, who expect bespoke, seamless service, this creates a gap between brand promise and actual experience.
I started by auditing how the leading KYC platforms actually work. Not marketing pages. The real product.
I ran through live demos of Persona, Jumio's KYX Portal, Alloy, Onfido/Entrust, and Stripe Identity. The goal was to understand what the market had solved and where the structural gaps were.
Persona's flow design is the clearest benchmark. Their consent screen is clean, the form is progressive, and they handle risk routing well. Low-risk cases complete in under 10 minutes. High-risk cases auto-escalate without creating friction for the client.
Jumio's backend is powerful. Their KYX Portal generates risk scores with reason codes, supports 800+ detection rules, and surfaces a relationship graph that shows connections between flagged entities.
Alloy is built for the case management layer. Their workflow is centered around manual review queues with configurable routing rules.
After running through all five platforms, a clear pattern emerged: the tools are built for one user at a time.
Persona optimizes the client flow. Jumio optimizes compliance review. Alloy optimizes case routing. None of them are designed for the coordination problem between all three parties.
No platform surfaces AI confidence scores in plain language for reviewers. No platform gives advisors real-time case visibility without a separate portal login. No platform gives clients actionable rejection context. No platform adapts the onboarding flow based on a client's risk profile before they start.
The gap isn't in verification technology. It's in cross-role coordination design.
The research pointed to three people who are always present in a KYC case, but rarely designed for together.
Robert Chen is a 52-year-old private equity partner initiating onboarding at Goldman Sachs PWM. He was referred in. This is an invitation, not a cold application. He expects a process that reflects that.
Robert's Core Frustration: He submitted everything he was asked for three days ago. He hasn't heard back. He has no idea if there's a problem or if this is just how long it takes.
Robert's breaking point is ambiguity. He doesn't need a fast process. He needs a process that respects his time by being transparent about where things stand.
Design Implications: Proactive status communication. Plain-language explanations when documents are rejected. A progress tracker that tells him what's next, not just where he is.
Michael Torres is Robert's relationship manager. He brought Robert in. His reputation is on the line.
Michael's Core Frustration: He's Robert's point of contact, but he has less information than the compliance team. When Robert calls with a question, Michael has to say 'let me find out,' which undermines confidence in both Michael and the bank.
Michael's structural problem is that he's accountable without visibility.
Design Implications: A real-time case dashboard across his entire client roster. One-click compliance messaging. Instant visibility when a case gets flagged.
Jessica Park manages a queue of 40+ cases. Most come through with AI risk scores. When the AI flags something, all she gets is 'High Risk.' No breakdown, no confidence level, no specific issue.
Jessica's Core Frustration: She can't explain a rejection to a client without digging through raw documents herself. Every override she approves has to be auditable. She needs a reason, not just a result.
Jessica's structural problem is that the AI is a black box.
Design Implications: AI risk breakdowns in plain language. Confidence scores per check. A structured override workflow with an audit trail baked in.
Robert, Michael, and Jessica are all looking at the same onboarding case from completely different angles. The system doesn't reflect that.
The solution had to make all three people feel like they were inside the same process, even when they're playing different roles in it.
With the research pointing to a multi-party coordination failure, I defined five features that would directly address the structural problem.
Advisor Case Dashboard. Gives Michael real-time visibility across his entire client roster. Every case status, every action required, in one place. He can message compliance, request documents, and see where each client is without leaving the dashboard.
AI Explanation Layer. Translates AI risk scores into plain English. Instead of 'High Risk,' Jessica sees exactly which check flagged, what the finding was, and what the confidence signal is. Structured. Auditable. Actionable.
Client Status Portal. Gives Robert a transparent view of his onboarding progress. Diamond progress tracker, document-by-document status, plain-language rejection messages, direct advisor contact. No ambiguity.
Risk-Calibrated Client Flow. Routes Robert's onboarding based on his risk profile before the session begins. Low-risk: 3 steps, 10 minutes. Enhanced due diligence: 5 steps across multiple sessions with clear expectations set upfront. One flow doesn't fit all.
Actionable Rejection Recovery. When a document gets rejected, Robert doesn't get a generic error. He gets the reason, what's wrong, and exactly what to upload to fix it. One clear path forward.
I explicitly scoped out biometric verification UI, document OCR screens, multi-factor authentication flows, and CRM integration. These are real implementation needs, but they're vendor and engineering layers. The coordination design is what needed to be addressed first.
Three flows. Eleven screens. Each designed for a different person, all connected by the same case.
The first thing Robert sees isn't a form. It's context. Two versions, depending on his risk profile.
Low-risk: three steps, ten minutes, start when you're ready. Enhanced: five steps, some sessions may span multiple days, here's what to expect.
This sets honest expectations before the first field is filled in. For a client who was invited in, that framing matters.
If Robert gets here, something needs attention. The screen shows his full document table: every submission, every status. For any rejected document, the reason is inline, with a direct Upload now link in the same row.
He doesn't have to go find the problem. The problem comes to him.
Robert's progress dashboard. Diamond milestone tracker at the top. Below it: an action-needed callout if something's waiting on him, a document summary with color-coded status badges, and Need Help? Message your advisor pinned to the bottom.
He knows exactly where he stands. He knows what to do next. And he knows who to call.
Michael's full client list. Tabs for All / Action Required / In Progress / Completed. Columns for client name, risk tier, status, assigned compliance officer, time in onboarding, and next step.
One view, every client. No digging.
Michael clicks into Robert's case. Left column: case metadata (risk tier, time in queue, assigned compliance officer, current status). Below that: three action buttons. Send Document Request. Message Client. Contact Compliance.
Right column: the full case timeline, document checklist, and compliance notes. Michael can take action from here without touching another system.
Full-width messaging thread between Michael and Jessica, with a thin case header at the top for context. Their conversation lives in the case record. Messages are retained as part of the case record so both parties know this is documented, not informal.
Jessica's full case list. Tabs for All / Full Manual Review / Expedited Review / Auto-Approved. Columns: client name, advisor, risk score, status, time in queue.
The tabs reflect actual routing logic. Full Manual Review cases need her attention. Expedited means a deadline. Auto-Approved is logged and auditable even without her action.
Jessica opens a case. At the top: a three-panel AI summary card (AI Screening Summary, Issues, AI Confidence) with a View Full Breakdown link. Below that: the timeline and a document checklist with color-coded status badges.
She can see at a glance what's been submitted, what's been reviewed, and what's still pending.
This is the thesis screen for the whole project.
Left column: case status (flagged), the specific flags with issue type and source document, a plain-language confidence summary, and three action buttons: Approve, Request Documents, Escalate.
Right column: the full check table. Five KYC checks, each with: Result, Finding, and Confidence Signal. PEP Screening: clear. Adverse Media: clear. Source of Funds: inconclusive. Identity Verification: verified. Address Verification: inconclusive.
Jessica sees exactly what the AI found, what it couldn't verify, and how confident it was in each result. She can make a judgment call with actual information.
Three states depending on Jessica's decision.
Approve Flagged Case: note field, optional file attachment for supporting documentation, Submit Approval Override. The override is documented, not just clicked through.
Request Documents: document type dropdown, reason dropdown, free-text instructions, file upload. The request goes back to Robert with context, not a generic please resubmit.
Escalate to Senior Compliance: note field, assign reviewer dropdown. Internal handoff, no file upload needed.
Three different decisions. Three different forms. Each one structured so the action is auditable.
This project started as a compliance problem and turned into an information design problem.
The KYC tools on the market are technically capable. Persona can verify identity. Jumio can score risk. Alloy can route cases. What none of them do is design for the space between those systems: the handoffs, the status gaps, the moments where Robert doesn't know what's happening, Michael can't answer his client's question, and Jessica has to approve a case she doesn't fully understand.
The five features I designed aren't new technology. They're a redesign of how existing data gets communicated across three people who are all working on the same problem from different angles.
The AI Explanation Layer is the highest-stakes design in this system. In this project I designed the display of AI results, but the real design challenge is the logic behind confidence scoring and how that gets calibrated over time. That's where the collaboration with engineering and compliance leadership would get interesting.
I'd also prioritize usability testing with actual compliance officers. The check table on Screen 15 is designed to be scannable, but whether the column structure actually maps to how a compliance officer processes risk is something I'd want to validate before committing to the pattern.
And the client flow is the part I'd want to test most carefully with real HNW clients. The language around Enhanced Due Diligence on Screen 1 is deliberately plain, but the right level of disclosure and the right framing is something that would need to be tested with the actual population.
Ask me anything about Kevin's work!