← Nancy Xing

VMware NSX: Feature Flags Management

Enterprise SaaS · Product Design

Role
Sole Product Designer
Timeline
2022
Tools
Figma, UX Research
Client
VMware NSX
NSX Feature Flag Management project poster: overview, key users, user flow, prototypes, and future vision

Overview

The Feature Flags Management (FFM) tool sits inside the VMware NSX portfolio. It lets teams toggle functionality at runtime through code flags scoped to user properties, so features ship, roll out, and roll back without redeploying.

I was the sole product designer, owning the work end to end: research, information architecture, user flows, and final high-fidelity screens.

Goals

What FFM set out to do

At a system level, the product is organized around three jobs, Control, Insight, and Action, that feed management & monitoring and experimentation & delivery.

FFM information architecture: Control, Insight, Action feeding management, monitoring, experimentation and delivery
High-level architecture: Control, Insight, Action.

Research

Understanding the people

The key users for FFM are internal-facing roles at VMware as a service provider. During research I interviewed eight VMware employees, three Product Owners, three Site Reliability Engineers, and two Developers, asking each to share how they used the Feature Flags Management tool and where it hurt.

The three FFM personas: NorthStar Product Owners, NorthStar Developers, and SRE Engineers, with their needs
Three internal personas and the jobs each needs FFM to do.

Solution

Flows built around real jobs

From the research I designed flows for the two roles who live in the tool most, the Product Owner and the SRE, anchored by a flag dashboard, multi-stage creation, in-context analytics, and ML-based recommendations for safe rollouts.

NorthStar Product Owner

A Product Owner runs experiments and reads the results to decide what ships:

  1. Create a flag with three variants going to parallel markets.
  2. Roll out the three versions and compare completion rate, accuracy, and cancellation, including how often users return to modify the policies and where they made changes.
  3. Get an ML-based recommendation on the strongest variation, click into the real statistics behind it, and roll back the others.
Product Owner user flow across Day 0 flag creation, Day 1 in-product analytics, and Day 2 recommendation
PO flow: Day 0 flag creation → Day 1 in-product analytics → Day 2 ML recommendation.

NorthStar SRE

An SRE uses flags as a reliability lever when something goes wrong in production:

  1. Discover that the latest release has a severe latency problem caused by a new networking feature.
  2. Because that feature is behind a flag, remediate instantly by switching it off, with no waiting for developers to fix the underlying code before users are unblocked.
  3. Treat feature flags as a way to support the core goal: maximizing reliability.
SRE user flow across Day 0 flag monitoring and Day 1 troubleshooting
SRE flow: Day 0 monitoring in the NSX SRE portal → Day 1 troubleshooting and rollback.

Prototypes

High-fidelity screens

The flows came to life as a connected set of screens inside VMware Cloud Services, from the at-a-glance dashboard to flag creation, ML-driven insights, and a dedicated SRE portal.

FFM dashboard: flag and insight summary, total flags over time, and a feature dependency chart, plus the insight engagement detail view
The dashboard: summary metrics, flags over time, the feature dependency graph, and per-feature engagement insights.
Feature flag management: the flag overview table, the add-new-flag creation modal, and the deactivate-flag flow
Managing flags: the overview table, create-flag modal, and a guided deactivation with incident notes.
Insight detail: ML recommendation list, recommendation creation, model training setup, and evaluation chart
Insights & ML recommendations: define an outcome, train a model, and read which variation wins.
NSX SRE portal overview with open incidents, uptime, SLA, flags related to incidents, plus the incident detail view
The SRE portal: open incidents, uptime and SLA, flags tied to incidents, and incident detail for fast rollback.

Takeaways

What I'd carry forward

Enterprise design rewards fewer interactions, not more features. The wins came from deeply understanding a small set of expert users, then shaping each flow around the one job they actually came to do, working shoulder to shoulder with engineering and product.