Designing an AI-Assisted MSP Configuration Experience
Context
Pre-Graduation project with MAGNIT GLOBAL
UX Researcher Product Designer
My Role
Product Focus : Configuration Section
Magnit is a leading contingent workforce management platform that helps large enterprises manage contractors, vendors, and temporary workers at massive scale (thousands of workers, suppliers, and rules simultaneously).
The Vendor Management System (VMS) includes a Configuration section where internal teams (especially Managed Service Providers — MSPs) define critical rules, approvals, workflows, and thresholds that control how the system behaves for clients.
This project is focused solely on the MSP user's view of the Configuration section.
Complexity in tools like Magnit's VMS isn't a flaw - it's a feature for scale. The real challenge is bridging the gap between powerful capabilities and everyday usability. This project is about that gap.
research
Understanding The Challenge
Magnit’s VMS handles high-volume contingent workforce operations. MSP users must configure complex settings across deeply nested menus, fragmented data fields, and repetitive manual tasks.
Navigation feels like an “Easter egg hunt,” leading to high cognitive load, frequent errors, and long task times (12–15 minutes per common config).
_Before I could design anything, I needed to understand the full scale of what MSP users were dealing with. I spent time inside the live system mapping every screen, every nested menu, every setting. The goal was to find patterns, where things were fragmented, where users were likely to get lost, and what was silently eating their time.
The Configuration section varies by user role (Client, Supplier, User, MSP), with each containing multiple nested sub-sections and fragmented settings.
To capture the full scale and deep nesting of the MSP Configuration landscape, I mapped the complete information architecture in FigJam followed by a deep analysis of each section of config to find patterns.
Full Information Architecture of the MSP Configuration section (click the image for zoomable view)
What this revealed: the MSP configuration section had over 6 levels of nesting in some areas, with related settings scattered across completely separate parts of the IA. No single user had a full mental map of it - they all relied on memory and each other.
Synthesized User Insights: Mental Models, Personas, and Prioritized Pains
With the IA mapped and its chaos documented, I turned to the harder question: who actually has to navigate this every day?
I combed through interview transcripts, meeting notes, and SME discussions - not to collect data points, but to find patterns in how different people think about the system.
Four personas emerged - CS/VMS, Client Services Admin, Finance/Compliance, and PSO - each with their own mental model of how the system should work. And each quietly frustrated when it didn't.
Each persona had a completely different relationship with the configuration system - the CS/VMS user navigated it daily by memory, the Finance persona avoided it entirely out of fear of breaking something. That gap in confidence became a key design constraint.
These were the top recurring pains and opportunities across all sources as a prioritized list.
focus
Problem Statement :
MSP users in Magnit's Vendor Management System face significant friction in the Configuration section due to deeply nested navigation, poor discoverability, fragmented information architecture, lack of contextual guidance, passive change history, and heavy reliance on manual processes and tribal knowledge. This results in high cognitive load, time-consuming tasks (12–15 minutes on average), frequent errors, risk aversion, unnecessary escalations to L2 support, and overall inefficiency in managing complex, cross-silo configurations.
How might we Statement :
How might we provide intuitive, contextual guidance and reduce manual effort in deeply nested MSP configuration workflows, so users can discover, understand, and safely execute settings quickly and confidently without relying on tribal knowledge or escalating to support?
Problems
Navigation Complexity
Config pages are long, fragmented, and lack reliable search; finding the right setting is time-consuming
Often mentioned as ester egg hunt
Repetitive Manual Tasks
Users spend excessive time on manual, repetitive tasks that are highly time-consuming and could be automated.
Often mentioned by users
Data Visibility Gaps
Essential data like original start dates, visa information, and audit logs aren't visible in the UI, forcing users to rely on external systems and manual spreadsheets for tracking.
Often mentioned by MSP PSO Team
Statements
The configuration area in Magnit feels less organized, often with ad-hoc placement of items, making it harder to understand and navigate.
‘‘
MSP Client Services (VMS only)
‘‘
If LOS were automatic,
a 10-minute extension becomes 2 minutes. Multiplied by 15–20k per month, the savings are huge.
PSO Team (Pushpendra & Suraj)
‘‘
Clients ask why a setting exists or works a certain way, product documentation doesn’t always provide rationale.
MSP Client Services (VMS only)
WORK
Solution & Design Decisions :
From the synthesized pains and HMW opportunity, I explored multiple ways to address discoverability, cognitive load, and manual effort in the complex MSP config space.
The HMW pointed to a need for on-demand, contextual support in a deeply nested system.
I ideated an AI Config Agent - a lightweight, overlay chatbot that activates on hover/click within config screens. It explains settings, guides tasks, flags risks, and suggests actions in natural language, while keeping the user fully in control.
Alternatives considered (before finalizing AI Config Agent):
Full IA redesign (disruptive, requires backend changes).I ruled it out because it would've taken 6 months and broken everything.
Better search + static tooltips (insufficient for deep context and dynamic guidance).
Templates/bulk edits (helps repetition but not discoverability or understanding).
Persona-Based Opportunity Mapping:
Task Journey Comparison: Manual vs. AI-Assisted
To make the gap tangible, I mapped out task journies side by side : on one hand - through the current system, on the other- with the AI Config Agent. The difference speaks for itself.
A task that should take seconds stretches into 12–15 minutes of second-guessing.
With the AI Config Agent? They just ask. The agent locates the right setting, flags any risks, confirms the change - no menu-hunting, no tribal knowledge required.
Same task. A fraction of the time.
Outcome
Low-Fidelity Screens
Before jumping into polish, I roughed out the layout and placement of AI elements - just enough to test whether the flow made sense. No colors, no final copy. Just structure, interaction, and intention.
This kept feedback focused on what actually mattered: does it solve the pain?
High-Fidelity Prototypes
Once the structure was validated, I brought the concepts to life. The high-fidelity screens focused on the details that build trust - clear typography, intentional color cues (red for risk, green for success), and smooth interaction states.
Prototype Demo Video
This short screen recording demonstrates the final clickable prototype of the AI Config Agent in action.
Thank you for reading this far.
This project taught me that the hardest UX problems aren't about making things simpler - they're about making complexity navigable. The goal was never to dumb down Magnit's VMS. It was to make its power accessible to the people who depend on it every day.