12 design awards & 7 years in B2B, AI, and Web3.

001

AI Chatbot

What

An enterprise AI assistant that brings connected apps, project conversations and device access into one desktop workspace.

The design challenge was to make delegation feel clear: what the assistant can use, what it is doing, and when it needs a decision from you.

The user types What can you do? The Assistant explains that it can find information, prepare meeting briefs and draft follow-ups, then shows app access and a concrete example. Replies arrive in order; the last includes feedback and a timestamp. Projects and Settings hover labels sit four pixels from their icons. Design notes opens from project history, followed by Settings and Connections.

1. A useful head start

A blank prompt asks people to decide both what to do and how to ask. Prepared suggestions give them a concrete starting point.

Starting a suggestion brings it into the conversation. Short progress updates explain what happens next.

The camera makes one continuous close-up of a suggestion and its kickoff hover state. It pulls back before the prompt appears in chat and the card leaves the panel. Three replies arrive in sequence with a typing cue between them. The completed response ends with feedback controls and a timestamp.

2. Context, under your control

Before delegating work, people need to know what the assistant can access. App choices sit beside the request, and enabled apps stay visible.

Access is scoped to the current chat, so the available context stays explicit.

The view moves closer to the app menu and composer. Enabling Notion adds its icon last. The menu closes and the camera pulls back. The located-sources message stays in place while the questions and final draft confirmation append below it. Typing ends when the response is complete; feedback and a timestamp appear together, and the composer returns to its ready state.

3. Finding the right balance

The composer needs to show two things clearly: what the assistant can use, and what it is doing. I compared three ways to place controls and progress around the same input.

Three composer explorations appear together at the same width on a transparent frame, without labels or divider lines. App logos and More share a chip in the first option. The indicators spin while details change from Reading context to Checking sources to Drafting a reply. Each icon and label form one compact pair. The shared shelf and full strip align the entire pair to the same right edge for all three messages. Each update fades through a clear frame over 200 milliseconds. The layouts and camera stay still.

4. Controls within reach

The composer holds both the request and the choices that shape it: sources, project, computer and model.

Choices stay visible after each menu closes, so people can see which apps, project and model will shape the work.

The view moves closer for compact menus and widens for the taller app and suggestion panels. Add context offers Files and folders, Paste a link, Sketch and Connected apps. Suggestions follow. Figma starts disabled and absent from the composer. Switching it on adds its logo after Slack. Project 1 changes to Project 2. Two computers have green availability dots; Home Mac mini is gray without a dot. Studio MacBook is selected, then Auto changes to 5.6 Sol. Each selection persists after its menu closes. Rounded hover surfaces use primary black labels on pale gray, and checkmarks identify selected items.

5. Your turn

Sometimes the assistant needs a decision before it can continue. The question appears at the composer, where the next action already happens.

A dark question shelf temporarily replaces the controls. Sending an answer restores them and work continues.

A working bubble remains in the conversation. The app selectors fade out and a black shelf with a white question appears above the stationary input. The camera moves closer as the user types an answer. Sending appends the answer to the conversation, restores the selectors and resumes the working state. The question never appears as a chat bubble. Your turn uses a gently pulsing blue icon. Working returns at the same right edge.

6. A place to return to

Long conversations should not make earlier work hard to find. A conversation map offers a short overview of each moment, a direct jump to the exchange, and a way to pin it.

The map provides orientation while the conversation remains the main workspace.

Hovering a conversation marker reveals a short overview. Selecting a marker scrolls to the earlier exchange. On the next hover, the view moves closer to the map and preview card before pinning. This close-up stays through the return to the latest messages and the end of the demo.

002

AI Native Workspace

What

A mobile AI experience for StubHub brokers to finish urgent work away from the desk, delegate repeatable tasks and stay in control.

I designed a task feed that brings evidence and actions together, with agents that operate within rules the broker reviews. This vision was evaluated in a completed usability study with 16 brokers.

Impact

36%less time to complete tasks

0critical errors, down from 3

+2.8points in task ease, reaching 9.1/10

Design for mobile first

I started with the moments when brokers need to act away from their desks. A phone offers less room for evidence and input, while short sessions and frequent interruptions make it harder to finish a job.

That shaped three priorities: bring enough context to decide, make the next action clear and preserve progress when the session stops. The mobile experience should let brokers finish urgent work without retracing a desktop workflow on a smaller screen.

The navigation cost of modern apps

The constraint goes beyond screen size. Modern enterprise apps organize information by product domain, but a broker’s job rarely stays within one domain. Metrics, worklists, rules and record details each hold a piece of the answer.

I mapped that structure to question the work it puts on the user. Before taking action, the broker has to find the issue, gather its context and navigate to the right record. On mobile, every change of view makes that context harder to keep in mind.

Current model

Proposed model

To reduce navigation, I brought relevant records, evidence and the next action into one task. It explains what needs attention and why, while a finite mobile queue presents one task at a time. Brokers can review the evidence, complete or skip the task, and move forward with a clear sense of progress.

Proposal 1 Bring context and action together

The task card brings the evidence and next action into one place: the action note, relevant dataset, proposed solution and controls. The same structure supports different broker roles.

V1: Infinite scroll Feed

A continuous feed made several jobs visible, but encouraged ongoing browsing. It offered less definition around the decision in front of the broker.

V2: Prioritizes focused completion

The focused view keeps the suggested price, market comparison, inventory and governing rules in sight. Approve and Skip provide clear next steps without asking the broker to navigate elsewhere.

Proposal 2 Automate the routine. Surface the exceptions.

Bringing context and action together reduces navigation. The next step reduces manual execution: agents handle routine work in the background within agreed limits, while exceptions and approvals return to the broker with the context needed to decide and act.

  1. ManualAutomation 0%
  2. AI assistedAutomation 50%
  3. AutomatedAutomation 100%

The workspace makes that relationship visible. Agents and their activity occupy the darker space; the light task surface brings forward work that needs the broker’s attention. The goal is to let brokers oversee the work and step in when needed, while agents handle the routine execution.

See the agents at work

The workspace opens with agents and today’s agenda, making their activity part of the starting view. As the broker scrolls, the task surface rises to bring the next decision forward.

Interaction details

The opening brand mark resolves into the daily workspace. Agent activity and the date appear above the agenda, then the light task surface comes forward as the broker scrolls.

Prioritize urgent decisions on mobile

I prioritized mobile features around decisions that cannot wait. The onsale calendar keeps time sensitive buying opportunities close, so brokers can check what is coming without losing the current task.

Interaction details

The broker opens the calendar from the daily workspace, selects a date to review its events and returns to the agenda.

Step in where attention is needed

Later visits open on the unfinished task, keeping the broker’s next decision close. A deliberate pull reveals the agents and calendar, so the wider operation remains accessible while the task stays in focus.

Interaction details

The workspace starts with the unfinished repricing task. A deliberate pull reveals the agents and agenda; releasing settles the surface into place.

Know when your part is done

The completion state marks the end of the broker’s queue. A message appears before the session numbers count up, making it clear that the work needing their attention is complete.

Interaction details

After the final task is approved, a completion message appears. The session summary counts up after the message, then settles into a completed state.

Proposal 3 Teach the system to evolve

Automation should evolve with the broker’s way of working. I designed three ways for brokers to teach the system, turning their intent into repeatable workflows:

Each route leads to a review of the dataset, trigger, rules and allowed actions before the workflow can run.

That review gives brokers a place to refine the workflow and shape what the system does next. They can adjust the scope and rules, with a clear boundary between what repeats automatically and what still needs approval.

Start from a template

Existing workflows offer a concrete starting point. Brokers can choose a familiar task, then adjust its scope to their own operation.

Describe the workflow

Natural language captures the intent. The assistant gathers missing details, translates the request into a structured rule and asks the broker to confirm it.

State the intent

The broker describes the inventory to watch and the outcome they want. The assistant shows the sources and rules it is considering while drafting the workflow.

Resolve missing details

A focused question turns ambiguity into an explicit choice. The broker can choose an option or write a different answer before the rule is completed.

Review the rule

The trigger, dataset and rules become a structured proposal. The broker can inspect the scope and approval requirement before confirming.

Confirm what will run

The enabled workflow remains visible as a manageable item. Pricing suggestions return to the task feed for review within the agreed scope.

Show it

For work that is easier to demonstrate, recording turns an example into a draft workflow. The broker controls the recording and reviews the inferred steps before they become repeatable.

Choose to record

The workspace explains what will be captured and asks the broker to start the recording. The example begins from an ordinary task.

Demonstrate the task

A visible recording state and Stop control keep capture explicit while the broker performs the work.

Review before repeating

When the draft is ready, the review action returns control to the broker. Observed actions become a proposal to confirm.

Keep AI work visible and controllable

Copilot tasks averaged more than a minute, making waiting part of the interaction. I designed status to explain what is happening, who needs to act next and whether the outcome has actually been confirmed.

Questions and approvals share “Needs your input.” Queued, Working and Completed describe ordinary progress; Retrying, Blocked, Stopping and Stopped explain recovery and control. Brokers can steer a request while it runs and return to work that takes longer.

Define the transitions

The lifecycle connects normal completion, human input, dependency recovery and stop or resume behavior. Completion is reserved for a confirmed outcome.

Keep the handoff in context

The assistant acknowledges the request and makes ongoing work visible in the same workspace. The broker can follow the handoff without losing the original intent.

Interaction details

The broker enters a request in the composer. The assistant acknowledges it, shows its working state and moves into a focused response while preserving the conversation context.

Validate the mobile experience

I compared the current mobile experience with the new task feed in a completed study with 16 brokers. The tasks covered urgent repricing, finding a buying opportunity, completing a delivery and investigating a payment discrepancy.

All 16 brokers preferred the new experience. Task completion time fell from 4 minutes 13 seconds to 2 minutes 42 seconds, a 36% reduction. Unassisted completion increased from 13 of 16 to 15 of 16.

Critical errors fell from 3 of 16 attempts to none. Perceived task ease rose from 6.3 to 9.1 out of 10. These results describe the usability study; they do not represent a production rollout.

003

POS

What
Redesigned StubHub’s Point of Sale to facilitate the full ticket broker lifecycle.
Impact

29%reduction in task completion time

95%drop in batch upload errors

35%increase in mobile session engagement

Event inventory

High volume brokers need to compare listing details with venue location while pricing tickets. A synchronized split view keeps both in sight, reducing the need to navigate between isolated views.

The three pane layout combines a scalable data table, an interactive map with pricing heatmaps, and a listing sidebar for quick pricing adjustments.

Event inventory: table, pricing map, and listing sidebar in light mode

Add purchase

Enable brokers to create complex purchases while keeping critical business context visible. Separating reference data from the task helps users stay oriented during high pressure buying moments.

The left panel holds purchase facts, account information, and payment summaries. The right panel adapts to single or bulk purchase entry, supporting cross checking across Puller and Finance workflows.

Purchase setup and persistent reference panel
Single purchase entry
Bulk configuration
Ticket group configuration
Review

Batch upload 1
File upload

A guided import wizard replaces manual entry for high volume inventory onboarding. The goal is to catch formatting problems before tickets enter the system.

  1. Template & upload. Event specific templates standardize the input.
  2. Smart field mapping. Match spreadsheet columns to system attributes, with manual remapping when needed.
  3. Validation review. Flag invalid prices, duplicate seats, and other row level errors before submission.
Batch upload 1
File upload
Batch upload 2
Field mapping
Batch upload 3
Review
Batch upload 4
Review success

Map ↔ table

Help brokers evaluate and edit large ticket groups without leaving the current view. A map supports section based seat comparison and selection, while inline editing makes bulk updates possible directly in the table.

The bidirectional interaction connects spatial context to inventory data and reduces repetitive page transitions.

Inline editing for ticket groups
Seat map selection synchronized with the table

Earnings & commissions

Help accountants reconcile earnings and payments with clarity. A dual column table presents earnings and payments side by side so users can compare totals and trace each relationship.

Each row shows linkage status and highlights unpaired entries. A summary provides an immediate view of the overall balance.

Earnings and payment reconciliation overview
Reconciliation

Reflection

Scaling through incremental redesign

This was a full scale redesign of core broker workflows. I partnered with product managers and engineers on a phased rollout that supported continuous delivery with minimal disruption.

  1. Retain the existing drill down flow to keep the platform stable during the transition.
  2. Roll out changes tab by tab, starting with high impact areas such as Purchase Detail and Sales Detail.
  3. After the interface stabilizes, rework Allocation and Fulfillment around the new interaction model.

The approach kept design consistent while giving engineering a clear path to scale.

004

AI Support

What
ReachPro brokers run millions in GMS. When something breaks, generic support isn't good enough. A resolution center built for scale, AI triage, gated human escalation, dual surface design for both broker and specialist, and a five lever content framework that shapes every AI response automatically.
Impact
67% faster response · 38% → 71% first contact resolution · 65% of tickets resolved by AI

Support where the work happens

AI support stays available throughout the product, while human support begins with a structured ticket.

Human support is gated

Human support is high cost and high value. It should only start once we have a minimum information package. How a broker accesses a human agent isn't a small decision, it's a core design problem.

Keep tickets and conversations connected

A persistent ticket list keeps support close to the work. Brokers can start a request, open ticket details and continue the conversation without losing context. AI stays available alongside the work, with the context needed to help before escalation.

Open tickets always visible

Enterprise brokers don't manage one issue at a time. The ticket list sits persistently alongside every surface, so they can triage what's open, pick up where they left off, and escalate when needed, without breaking their workflow.

Exploration: Search first entry

An earlier direction that led with self service: search, runbooks, and AI before any human contact. Reaching an agent required an explicit choice.

Final design: Structured intake, immediate access

For enterprise brokers, time is revenue. When something breaks, they need a direct path to the right person not a search bar.

What does a broker need inside a ticket?

Once a ticket is open, the broker needs more than a chat window. They need context, status, history, and the ability to act, all without leaving the thread. The question is how much to surface, and when.

Exploration

The action bar floats on top of the thread, close and escalate sit at the top, but nothing else is there to support the decision. Actions exist, but context doesn't. The broker has to leave the thread to understand what they're dealing with, which breaking focus every time.

Final design

The side panel brings context to the conversation. Account summary, related IDs, financial status, and history, all there without leaving the thread.

The core flow

The happy path is the easiest to design: AI engages, resolves, ticket closes. No human needed. The hard work is in the edges: proactive escalation, reactive escalation, structured intake, human handoff, and the feedback loop that closes the system. Getting every one of these paths right is what makes the resolution center trustworthy at scale.

Happy path

The AI resolves the issue directly, no human needed. Ticket closes with a resolution summary. Fast, low cost, no friction.

Escalation flow

Escalation Flow Proactive + Reactive

Proactive escalation

AI confidence peaks → resolution check surfaced → broker confirms unresolved → escalation begins.

When the AI reaches high confidence, it surfaces a check in, not because it's failing, but because it's done its best work. If the broker confirms the issue isn't resolved, that signal opens the escalation path

Reactive escalation

Broker flags issue → feedback triggers escalation → intake begins.

The broker signals the AI isn't helping, through feedback or frustration. That signal triggers the escalation path, not a dead end.

Structured intake captures context

Before a human picks up the ticket, the broker completes a structured intake. This packages the minimum information a specialist needs to act immediately.

Specialist receives full context → picks up ticket

A specialist receives the ticket with AI summary, intent, and account context already in place. No reexplanation. No cold start. The handoff carries full context forward.

Issue resolved → feedback collected → system learns.

The issue is resolved and the broker is notified with a clear resolution summary. A feedback prompt closes the loop, informing how the AI and intake improve over time.

The other side of the ticket @ System Admin view

The broker's experience of getting help is only as good as the specialist's experience of giving it. When a ticket lands on the agent side, they need a complete context package, not a blank chat window.

This section covers what data we surface for the human agent before they ever type a word: the AI summary, intent classification, account history, and related IDs that turn a cold handoff into a warm one.

What I'd do differently

The entry design went through multiple explorations before landing on the form first approach. In hindsight, I'd have pressure tested the gated model earlier with actual enterprise brokers, not assumed the friction would be too high. The decision to deprioritize it was right, but it should have been validated, not just reasoned through.

What I learned

Designing AI responses is a content design problem, not an engineering one. The five lever framework only existed because someone decided to own it from the design side. Without that, the AI would have defaulted to generic, unhelpful answers that erode trust faster than no AI at all.

005

Design System 1

What
A design system for high density broker work. Tokens, components, and interaction patterns built ground up
Impact
Adopted by 6 designers and 50+ engineers

Overview

Led the design and rollout of StubHub’s Supply design system across design and engineering.

Scope: Tokens, components, interaction patterns, contribution guidelines, and the audit/governance process that keeps it healthy.

Opinionated Design System

A design system without principles is a UI kit.

Four product heuristics turned the system from a visual library into an opinionated expression of what the product believes about its users.

Findability

Accuracy = Speed

Readability

Intelligence

Design principle · Findability

I made search, filter, and navigation first class citizens. Most systems treat filters as isolated patterns. For brokers who live in search all day, they had to become the platform's primary surface.

Design principle · Accuracy = Speed

State, feedback, and bulk action became foundational, not optional. In a volatile market, every component had to acknowledge action instantly and let brokers move at scale without sacrificing precision.

Design principle · Readability

The table system became the heart of the design system. Brokers scan thousands of listings a day, at this scale, readability isn't aesthetic, it's operational.

Design principle · Intelligence

Intelligence became a contextual layer, not a separate destination. AI is most useful when it understands user intent and the context the broker is already working in, so the broker is always the decision maker.

Designed for how brokers actually work.

78% work across 2+ monitors. 84% default to dark mode. The platform exists on densely populated desks, not in pristine demo screenshots.

Research spanned four broker archetypes, Low Volume Broker, High Volume Broker, Accountant, and Pricer, each with different scanning patterns, error tolerances, and monitor configurations. The system was tuned to the highest volume, lowest tolerance workflow first.

Tokens

Primitives → semantic tokens → components. Built bottom up, so theming, density, and contrast modes are remapping, not rebuilding.

Before
Inventory Table View

After
Inventory Table View

Boosted Sales, Total Sales, Margin, Sell Through Rate, and Average Ticket Price live as a single horizontal status row. The inventory table is the page, not a section. Tokens, table system, and Findability principles applied end to end.

Before
Inventory Tile View

After
Inventory Tile View

Expansion now exposes editable seating, pricing, and listing data in line. Insights view, edit filters, and bulk actions sit in a consistent top right action cluster, governed by the Accuracy = Speed heuristic.

Before
Event Detail Page

After
Event Detail Page

Market listings, sales, purchases, and notes share consistent table treatments. The seat map is integrated as a system surface, not a visual exception. Comparing your inventory against the market is now a glance, not a workflow.

Before
Purchase Flow

After
Purchase Flow

Summary, Fulfillment, Earnings & Commissions, and History sit as system tabs inside the purchase. State language is unified across rows and detail views. The same flow handles single and resale tickets without branching.

Before
Bulk Action

After
Bulk Action

Upload File → Field Mapping → Validation Review → Preview & Confirm. Each step is a system component with its own state language. Errors are caught and resolved before they touch real inventory. Built on the Accuracy = Speed heuristic.

Audit & Governance

Storybook as the source of truth, not as documentation of one.

Governance

Every component lives in Storybook with usage notes, props, examples, and a live light/dark toggle. Quarterly audits surface drift between Figma, Storybook, and production. The system stays opinionated by being maintained, not by being written.

006

Mobile First

Crypto Copy Trading

Pulse @ Intelligence attention builder

Portal @ CRA Mobile first for architecture onsite

007

Collaboration SaaS

What
The ACE industry faces scattered information and complex project coordination, leading to inefficiency and disrupted workflows. Archi streamlines the architecture office workflow.
Impact
18% retention increase post launch

Final Design

Internal Community

Internal Library Resources

Project Detail

Notes

Users can capture their thoughts effortlessly with Notes, enabling informal, note taking to foster creativity and management. Start from scratch or utilize pre saved templates for quick access to structured formats.

Notes: Edit & Collaboration

My Tasks

Task Detail

Project timeline

Reports

Research

Coordinate long term architecture projects involved 5+ stakeholders

Coordinating long term architecture projects is challenging, especially for high rise buildings that can take up to seven years and involve frequent management changes. Keeping project timelines aligned and stakeholders, including architects, project managers, engineers, and consultants, synchronized is complex and crucial.

Current workflow metric: Architect Daily

Existing Tool Failed

Insights

Existing workflows and tools fall short in supporting both internal and external collaboration.

External collaborators treated as guests

Tools like Asana and Monday.com are built around internal teams, classifying people outside the company as limited “guests.” In AEC, external collaborators, including consultants, engineers, and client representatives, are essential to the project but are treated as outsiders.

Information loss

Project knowledge is scattered across emails, WhatsApp, and calls, making responsibilities difficult to track over long projects. When team members leave, their knowledge and records are lost.

Misalignment

Each design stage involves 5+ stakeholders. Keeping everyone aligned is difficult, and misalignment leads to project delays and increased costs.

Address challenges

Challenges in coordinating projects among 5+ stakeholders.

Solutions: Project Task Board

Prototype Study

First Round Design

MVP Post Launch User Testing

We conducted a controlled MVP trial with a mid sized architecture firm consisting of six team members (1 PM, 1 PA, and 4 architects). We extended the trial to include external stakeholders, 2 structural engineers, 1 architecture record, and 1 landscape.

Two weeks of usage activity logging revealed:

  • Users left the subtask empty.
  • Users duplicated the task card for internal use and external updates.

Refocus on High Value Users Position the product as a coordination tool for cross functional teams

Iteration

Second Round Iteration

Final HiFi

Retrospective

Testable Assumption

Several of our initial key assumptions didn’t meet expectations. This experience taught us that establishing a flexible, testable product lifecycle is essential. It allows us to validate ideas quickly, learn from them, and pivot as necessary.

Product Thinking

Learn to understand how to design an experience that would lead to good business outcome, help user to understand the value of the product, which requires understanding of the user motivation and comprehension.

008

Workflow SaaS

What
Creative production faces inefficiencies due to fragmented communication and difficult coordination, leading to delays and lower client satisfaction. Portal streamlines collaboration between agencies, clients, and external stakeholders.
Impact
18% retention increase post launch

AI Copilot + Project Dashboard

Project Overview

Project Milestone

Stakeholder Related Highlight

Add Action

Send Action to Stakeholders

Define Content

Stakeholder Management

Digital Asset Management

Delivery Markup

The Delivery Markup interface enhance client interaction with design deliveries. This interface allows clients to directly annotate and provide feedback on the designs through a freestyle markup tool, rather than just leaving textual comments. This approach enables clients to communicate their thoughts and modifications more effectively and creatively.

Effortless Feedback Anytime, Anywhere

Stay in sync with your projects effortlessly. Our mobile app allows clients to check status updates, track progress, and provide voice feedback directly from their devices. Make collaboration seamless, efficient, and intuitive, ensuring every stakeholder is aligned and every project stays on track.

Retrospective

Uncovering the Human and System Connection Through UX

In the Portal project, a key takeaway was that UX design goes beyond just visuals and interactions, it’s about deeply understanding user behavior, motivation, and how the system responds. Through multiple iterations, we uncovered the real issues behind the data. That’s the beauty of UX: uncovering the nuanced relationship between people and the systems they engage with.

009

Design System 2

What

Unity is a centralized design system for Portal, built to improve consistency, efficiency, and inclusivity. Our federated model had created isolated components and slowed development, so I brought reusable components, shared design tokens, and guidelines into one system for our single product startup.

I dedicated 50% of my time to Unity alongside other feature development, with the goal of improving collaboration and reducing time to market.

Impact
Unity V3 adoption grew from 26% to 60.83%

Workflow

Implementation Process

The setup of Unity begins with a team survey and component audit to identify gaps. Collaboration with product and engineering ensures alignment, followed by polishing and testing core components. After training the team and documenting the system, a technical audit finalizes integration before launching the design system.

Highlight Components

The flexibility of component

While reviewing the card components in our previous design system, I noticed that the federated designer governance model had led to the creation of numerous isolated components, increasing complexity. To address this, I analyzed the component usage data and identified the most common use cases for card components. Based on these insights, I designed a universal card component that could be consistently implemented across our platform, simplifying the system and enhancing scalability.

Documentation, configuration & accessibility

For the accessibility documentation, I detailed keyboard navigation patterns to ensure seamless usability for all users, including clear tab, focus, and action states. Additionally, I implemented proper labeling using ARIA roles and attributes, such as accessible names and heading references, to enhance screen reader compatibility.

Buttons: Configuration

Buttons: Accessibility

Input: Configuration

Input: Accessibility

Design System Impact

Global Adoption

To measure the impact of the design system, I implemented an Usage Tracker to regularly monitor and audit its performance. This tracker uncovered key insights, such as which components were most frequently used, which broke often, and which were legacy. Initially just for auditing, it evolved into a data driven decision making tool, guiding me in prioritizing fixes, optimizing commonly used components, and phasing out unnecessary elements. As a result, the latest version Unity V3 design system’s adoption rate rose from 26% to 60.83%, ensuring its efficiency and scalability over time.

Component usage

We display this data with a chart on the dashboard, where each component’s square is sized according to its import frequency in our application.

Metrics Drive Better Iterations

Armed with this visibility, we can confidently phase out outdated elements and focus on standardizing around our most effective components. This data driven approach ensures that every iteration of Unity moves us closer to a consistently cohesive and efficient design system.