Linc
4 week rapid design and build of GenAI knowledge tool MVP
CLIENT
Lincoln International, a banking advisory firm that helps various organizations with selling or buying a business, securing financing solutions and valuing their organization or portfolio.
IMPACT
0-1 design and build of MVP in 4 weeks
Front-end designs for roadmap features for future releases
Foundational visual language and navigational architecture for client with no prior history with product development
Beta users returned high NPS and satisfaction scores, with a 72% user retention rate
Enabled client to quickly prove value proposition of product
Secured CEO approval for funding of product org
Established foundational user research and design assets for the team to build on
TEAM/ROLE
UX+UI Lead (myself) + Product Owner, AI/ML Lead, Offshore Engineering team
RESPONSIBILITIES
As UX+UI Lead, the scope of my work covered:
Workstream scoping, planning and management
Defining user flows & wireframes for MVP features
Leading concept and usability testing with target users
Collaborating with client SMEs to refine MVP feature set and usability
Working with AI/ML and Engineering to refine MVP UI+UX
Design documentation (redlines, prototypes, annotated files) & build support
Context
Lincoln’s Mergers & Acquisition teams are a core function of the organization, and they were looking to improve their speed and quality of client service by addressing how deal documentation and information was stored, accessed and disseminated.
Problem
When a Lincoln client initiates the process of looking for a suitable buyer or company to acquire, it kicks off a multi-year effort that includes due diligence, email exchanges and a myriad of presentations and financial paperwork - which can result in thousands of individual documents being collected on the client’s shared folder.
This makes it challenging and time-consuming for deal teams to keep track of past exchanges, information or knowledge as the process wears on, as they typically have to rely on team memory or manually searching the for whatever information they need.
Brief
Lincoln leadership viewed this as the perfect opportunity to build a custom tool that would allow deal teams to quickly access the deal information they need through chat-based prompting.
To derisk investment in the project, the team was directed to leverage off the shelf architecture as much as possible - with the goal of releasing a functional MVP in 1+2 sprints, to test feasibility and identify opportunities for further improvement.
Rapid development & refinement of use case, UX and front-end design
With the high-level use case defined, I set out the priorities for design and codependencies for M/L and Engineering workstreams:
Identify target users and their individual requirements
Map out MVP features, user flows and key screens
Understand implications of feature requirements on tech architecture and limitations that we would need to design for in MVP
Audit off the shelf patterns and components and identify gaps
Rapid 80-20 ideation and solutioning for new interface elements
Align with branding and marketing teams on visual styling guidelines, and visual branding development
Validate core assumptions with quick and dirty user testing sessions throughout
Achieving all this in 2 sprints, required colocating with Product, AI/ML leads and Business SMEs 10 hours a day, 5 days a week to facilitate rapid discussion and testing of ideas.
Key features are highlighted below.
Prompting UI
I chose to heavily leverage interaction patterns and visual layout commonly found in the prompting interfaces of existing GenAI tools like ChatGPT, Bard and Perplexity - reducing the learning curve needed for users. Tweaks, however, were needed to accommodate features unique to Linc; simultaneously displaying chat history for the various knowledge modes, for example.
Switching seamlessly between multiple knowledge sources
One of the key features of Linc is the ability for users to toggle between multiple sources of data from which they could query:
Lincoln Library - data and information that is visible to everyone in the organization
Deal workspace - Deal and client-specific data that is only visible to members of the Deal team
SafeGPT - Information publicly available from ChatGPT, without having usage information stored by OpenAI
The combination of a data toggle and mode-specific flyout control allows for the various mode and sub-modes to be displayed cleanly.
Client workspace (RAG) creation and setup
The Deal workspace is effectively a custom RAG database set up for each client, pulling information solely from the uploaded documents that exist within a client workspace. Early wireframes helped define the key elements needed in the flow, including user provisioning and permissioning, document upload/download and general metadata capture.
Workspace management
The workspace management flow allows admins to quickly and easily maintain and update the users and data within a client workspace.
PPT Deck and slide retrieval
User testing and feedback sessions helped to prioritize the key types of data users would look to get from Linc. Users can quickly prompt for and retrieve specific Powerpoint decks and slides, with the ability to preview content and metadata before downloading it to their desktop.
RAGAs validation
A key differentiator that Linc has over publicly available LLM tools is its ability to provide a “confidence rating” for the text-based responses it provides. Using a RAGAs framework, the rating acts as a guardrail against LLM hallucinations and synthetic artifacts. To ensure that users remember to use the provided information responsibly, all ratings come with reminders of their professional responsibilities when using GenAI tools in their process.
