Designing conversational onboarding to streamline institutional KYC and reduce turnaround time
A conversational onboarding experience powered by Sophia, an AI agent that guides users through the process, validates and cross-checks uploaded documents, and provides personalized support at every step.
Project Type
Conversational Design
Workflow Optimization
My Role
UX Designer,
Rapid Prototyper (Claude)
Team
1 UX Designer, 1 Solution Designer,
3 Domain Experts, 2 Sales People
Duration
1 Month
TL;DR
I designed a conversational onboarding experience that used an AI assistant to replace fragmented, email-driven document collection with a guided, validated self-service flow. I translated a complex institutional onboarding process into clear conversational steps, determining where the AI should guide, validate, and hand off to the user to make the process faster and more reliable.
Predicted Impact
In Development. No recorded impact just yet, but this is what the design, especially what UX aims to achieve.
Customer effort
Reduction of manual entry and document categorization required from customers
First-pass completion
Increased rate of complete, accurate documentation without repeated corrections.
Time to complete
Shorten the time required to gather and submit onboarding documentation
Confidence in AI
How often customers accept AI-generated document matches and validation results
Context
When this company acquires a new institutional client, the client has to share documentation across 20-50 internal teams, each running their own due diligence.
Document collection from the customers happens over email and through a relationship manager, with no shared inventory of what's already been gathered. Requirements surface late as individual teams realize they need more, documents get requested more than once, and every file requires manual validation before any team can act on it, driving up turnaround time and frustrating clients.
The Problem
For clients this means
Unclear requirements
Requirements surfaced gradually across teams
Repetitive requests
Clients were asked for the same documents more than once
High manual effort
Finding, organizing & submitting documents was tedious
No real-time feedback
Errors were discovered only after manual review
Fragmented experience
Onboarding was spread across email threads and RMs
For internal teams this means
Manual validation
Teams repeatedly reviewed and verified the same information
Duplicate work
Lack of shared visibility led to repeated requests and reviews
Late issue detection
Missing or inconsistent information surfaced downstream
High coordination overhead
Teams spent time chasing documents and resolving issues
Longer onboarding cycles
Back-and-forth increased turnaround time across 20-50 teams
The Solution
A Customer Portal: A Client’s Hub for Everything Related to the Organization
The customer portal brings onboarding, account information, and service requests into one place. Its dynamic dashboard surfaces the right information, status, and next actions based on where the customer is in their journey.
I designed it around what customers need to know and act on now, reducing information overload and making key actions easier to find.
Guided Onboarding in Natural Language
Sophia, an AI-powered conversational guide that walks customers through onboarding step by step. Customers can follow her prompts, ask questions, or request clarification at any point without leaving the flow.
I designed the conversation to keep guidance contextual, giving customers help at the point of need without interrupting the onboarding flow.
Bulk Upload Documents
Users are guided through each requirement one at a time, but can also bulk upload documents if they already have everything prepared. Sophia parses and automatically matches each document to the relevant checklist item.
This gives prepared users a faster path while removing the need to manually identify, categorize and upload each document against the right requirement individually.
Agent-led Document Validation
Sophia validates documents as they are uploaded by parsing their contents, cross-checking information across documents and public records, and flagging potential authenticity or consistency issues.
Catching issues upfront reduces customer back-and-forth and manual validation effort for internal teams, preventing problems from surfacing days later.
When an issue is identified, Sophia surfaces it immediately and guides the customer through resolving it before moving to the next requirement. Resolving errors in the moment prevents them from becoming downstream blockers and reduces follow-up work for internal teams.
A Manual Mode to Upload
For customers who prefer a more traditional experience, a manual upload mode presents each checklist requirement individually without the conversational interface. It still uses Sophia’s bulk upload, document matching, and validation capabilities in the background.
This provides a familiar path for customers who may be less comfortable interacting with AI, while retaining the efficiency benefits of AI-assisted document processing.
The Outcome
As the experience is still in development, I’m defining success through a set of measurable UX and operational outcomes rather than claiming final impact. The goal is to make onboarding faster, lower-effort, and more reliable for customers, while reducing repetitive validation and coordination work for internal teams. I’ll use the following measures to evaluate whether the design is achieving that:
Catch issues earlier
Fewer errors surfacing downstream.
Build trust in AI
Customers understand and confidently act on Sophia’s recommendations.
Reduce customer effort
Fewer manual inputs, uploads, and repeated requests.
Increase first-pass completion
More submissions completed without corrections or follow-up.
Reduce processing time
Faster document collection and validation.
Reduce internal effort
Less manual classification and validation.
What I Learned
Designing this experience reinforced that AI is most valuable when it takes on complexity without making the user understand that complexity. My role wasn't to add AI to an existing workflow, but to decide where it could meaningfully reduce effort, where users still needed control, and how to make its actions understandable.
I also learned that flexibility is part of good AI UX. Not every customer wants to interact conversationally, so providing a manual path while retaining AI-assisted capabilities allowed the experience to adapt to different levels of comfort and familiarity.
Most importantly, I learned to evaluate AI experiences beyond whether the agent can perform a task. The better question is whether the experience helps users accomplish their goal with less effort, more confidence, and fewer opportunities for error.
