Overview
ARitize3D is an enterprise platform used by large retailers to convert product imagery into production-ready 3D models at scale. As demand from Amazon grew, the existing enterprise intake and production workflow became a bottleneck; manual, fragmented, and costly to operate.
This case study focuses specifically on streamlining the Amazon enterprise workflow, which had unique constraints around scale, pricing, and turnaround time that differed from other enterprise clients. I led the redesign of this workflow to reduce production hours, prevent duplicate model creation, and enable faster, more reliable pricing and job setup using AI-assisted systems with human validation.
Project Duration
2 Weeks
Platform
Web
Role
Sole designer, design lead, and product partner
Project Scope
Amazon Enterprise Intake & Production Workflow
My Role & The Team
My Role: Lead Product Designer
I led the end-to-end redesign of the Amazon enterprise intake and production workflow within ARitize3D. This was a highly time-boxed, high-impact project completed over two weeks, where I was responsible for both design execution and product-level decision-making.
My responsibilities included:
-
Owning the end-to-end redesign of the enterprise intake and production workflow
-
Designing and validating AI-assisted model matching and complexity assignment flows
-
Translating operational constraints into scalable, productized solutions
-
Leading discovery, design, and delivery in a lean, fast-moving environment
-
Managing parallel priorities while leading two senior designers
This work required balancing speed, cost, and accuracy in a live production system used daily by operations teams.
Cross-functional Partners
Project Manager
(Primary User & Partner)
-
Daily workflow owner for Amazon enterprise intake
-
Ongoing validation partner throughout design and rollout
Scrum Master
-
Delivery coordination and stakeholder facilitation
-
Helped unblock works and maintain momentum during two-week timeline
VP of Operations & Engineering
-
Technical direction, feasibility checks, and AI strategy
-
Primary engineering partner (no broader engineering team involved)
Data Scientist (Supporting)
-
Built an early technical prototype for model matching
-
Provided input on AI capabilities and constraints
This work was completed within a live production system already supporting active Amazon operations, requiring fast decisions and close cross-functional alignment.
What ARitize3D Enables
The End Experience
ARitize3D transformed standard product images into interactive 3D and augmented reality experiences that embedded directly on retailer product pages.
The enterprise system I redesigned powers how these models are created, priced, assigned, and delivered at scale.
Mobile & AR Experience
The Challenge
While the shopper experience felt seamless, the internal enterprise workflow powering it was fragmented, manual, and difficult to scale. Although ARitize3D was already supporting enterprise clients at scale, the Amazon intake and production setup workflow had grown fragmented and manual over time.
What began as a design led self-serve platform was later adapted for enterprise use without dedicated design involvement. As Amazon became the company’s highest-volume client, the existing process struggled to keep up.
.png)
The Problem: A Fragmented, Manual Workflow
-
Orders were spread across multiple tools
-
Frequent back-and-forth between spreadsheets and the ARitize3D platform
-
Duplicate work wasn’t caught early
-
Model complexity and pricing decided manually for high-volume enterprise clients
-
Progress depended heavily on individual PM judgment
.png)
The Business Impact: Slower Output and Limited Growth
-
Slow setup before production could begin
-
Duplicate and unnecessary work
-
Higher operational costs
-
High risk of errors and rework
-
Limited ability to scale efficiently
From a business perspective, every manual step reduced throughput and constrained revenue. From a user perspective, the workflow relied heavily on memory, judgment, and workarounds rather than system support.
The challenge was to simplify and streamline this process without disrupting live production, while thoughtfully introducing AI in a way that improved speed and consistency without removing human oversight where it mattered.
Research & Discovery
Understanding the Real Amazon Enterprise Workflow
This project moved quickly, so discovery focused on learning how work actually happened, not producing heavy artifacts.
The goal was to understand where setup slowed down, where decisions broke down, and what prevented the team from scaling Amazon orders efficiently.
How I learned the workflow
Fast, targeted discovery with the people doing the work
I worked closely with the Project Manager responsible for Amazon intake and production setup, observing how large batches were prepared and validated day to day.
-
Walkthroughs of real Amazon orders and CSV uploads
-
Step-by-step review of how orders were created, validated, and exported
-
Identifying manual decision points and handoffs
-
Review of where duplicate model work and rework occurred
-
Ongoing validation with Operations and Engineering
Rather than redesigning in isolation, I designed directly against the constraints of live production.
What made Amazon different
Not all enterprise workflows were the same
Amazon orders had unique characteristics that amplified the problem:
-
Very large batch sizes
-
Strict requirements around pricing and complexity
-
High risk if setup errors weren’t caught early
-
Heavy reliance on speed to meet production demand
These constraints made the existing manual process especially fragile.
Key Insights
Where the system was breaking down
-
Setup work took hours or days before production could begin
-
Searching the database for duplicate models was tedious and slow taking days to complete
-
Complexity (pricing) assignment was a major bottleneck
-
PMs had to leave the platform often to complete critical steps
The biggest issue wasn’t effort, it was that the workflow was never designed as a system.
Strategy: Designing for Speed, Scale, and Confidence
With a clear understanding of how Amazon enterprise orders actually moved through production, the strategy focused on reducing setup time, eliminating manual decision points, and creating a workflow that could scale without relying on individual judgment.
The goal wasn’t just to make the flow faster, it was to make it predictable, repeatable, and system-driven.
1. Centralize Job Preparation
Key setup decisions; model matching, pricing inputs, and job readiness were spread across spreadsheets and tools.
The strategy was to:
-
Bring all setup decisions into one flow
-
Reduce context switching and manual coordination
-
Give PMs a clear, reliable path from intake to production
2. Reduce Manual Judgment Without Removing Human Oversight
The existing process relied heavily on individual PM and QA judgment, which didn’t scale well for large Amazon batches.
The system needed to:
-
Handle repetitive and error-prone work automatically
-
Surface clear recommendations
-
Keep humans in control of final decisions
Auto-complexity supported this by providing a fast starting point for pricing decisions, not replacing review, but removing guesswork.
3. Design for Volume, Not Edge Cases
Enterprise orders often included hundreds of SKUs, making per-item manual review slow and costly.
The strategy prioritized:
-
Batch-level workflows
-
Clear completion states
-
Confidence that an order was truly ready, without extra checks or follow-ups
.png){kind=small}
What Success Looked Like
-
Faster job preparation and handoff to production
-
Fewer duplicate or mis-scoped models
-
More consistent pricing and output
-
Increased throughput without additional headcount
Project Goal: Create a system-driven workflow that improved speed, accuracy, and scalability across enterprise intake and production.
Constraints
Designing this workflow required balancing speed, accuracy, and feasibility within a live production system under tight time constraints.
Extremely Limited Timeline
This project was designed and shipped in two weeks, with no opportunity for extended discovery or iteration cycles.
-
Decisions needed to be made quickly
-
Designs had to be production-ready, not exploratory
-
There was no margin for rework after launch
High-Volume Enterprise
Use Case
The Amazon intake flow handled large batches with hundreds of SKUs, amplifying the impact of any inefficiency.
-
Manual steps became exponentially more expensive at scale
-
Small errors led to significant downstream rework
-
The system needed to optimize for volume, not edge cases
Live Production System
This was not a new product or isolated experiment.
-
The workflow was actively used by PMs every day
-
Any change risked slowing down or blocking revenue
-
Backwards compatibility with existing orders was critical
Partial AI Readiness
AI capabilities existed, but they were not turnkey solutions.
-
Model matching had an early technical prototype, not a product
-
Auto-complexity would require human review and validation
-
Design needed to account for uncertainty and trust calibration
Trade-offs
Speed vs. Perfection
The goal was to deliver meaningful operational improvement quickly, rather than pursuing a fully polished or future-complete solution.
-
Prioritized core workflow improvements over secondary enhancements
-
Accepted UI roughness in favor of operational clarity and adoption
-
Focused on decisions that reduced friction immediately
Scope vs. Impact
Instead of redesigning both enterprise workflows, I focused on the Amazon pipeline to deliver meaningful impact within the two-week timeline.
-
Prioritized the workflow with the greatest business impact
-
Delivered measurable improvements where operational pressure was highest
-
Established a scalable foundation that could later extend to additional enterprise workflows
Flexibility vs. Consistency
Too much flexibility created inconsistent decisions and operational risk across teams.
-
Standardized decision paths supported consistent outcomes
-
Clear completion states reduced subjective interpretation
-
Structure improved scalability across large enterprise batches
Automation vs. Human Control
Automation needed to reduce manual effort without removing accountability from critical decisions.
-
AI provided recommendations, not final outcomes
-
Humans remained responsible for validation and approval
-
Clear system states replaced silent or invisible automation
Rebuilding the Enterprise Workflow
To understand the impact of this redesign, it’s important to first see how the system worked before.
The following diagrams show the intake and production setup process for Amazon enterprise orders first as it existed, and then after the redesign.
Before: A Fragmented Intake & Setup Process
_edited.png){kind=small}
Outcome: Progress depended on individual effort, spreadsheets, and manual coordination, limiting speed and reliability.
After: A Unified, System-Driven Workflow
_edited.png){kind=small}
Outcome: Faster job preparation, fewer errors, and scalable production planning to support higher volume.
The Enterprise Redesign
To solve the "speed vs. scale" dilemma, I redesigned the production pipeline to move away from individual PM judgment and toward a centralized, system-driven workflow. By bringing fragmented tools into a single interface, we eliminated the reliance on spreadsheets and replaced manual guesswork with AI-assisted decision-making.
The following modules represent the core interventions that transformed our operations, allowing the team to handle Amazon’s high-volume requirements without increasing headcount.
1. Order Intake & Creation: Structuring the Production Entry Point
Order intake was the foundation of the production system. Every downstream step, including pricing, job generation, and workload distribution, depended on the clarity of this first action.
As volume increased, intake lacked structure and guidance. Orders could be created, but they were not validated or progressed intentionally. PMs were left to manually interpret what needed to happen next and coordinate across teams to keep the work moving.
I redesigned intake as a guided, state-aware workflow that transformed it from a manual handoff into a reliable, system-driven entry point built for scale.
Before: Manual, Error-Prone Setup
The intake experience relied heavily on manual coordination and implicit knowledge.
Orders were created, but not guided, leaving PMs to interpret what to do next and slowing down every downstream step.
What happened
-
PM submitted a CSV to create an order
-
System returned them to a dense list view
-
No visible confirmation, progress, or next step
-
Job creation required a separate manual action
Why it was a problem
-
Primary actions were buried in operational data
-
PMs had to rescan and re-locate newly created orders
-
No sense of system state or completion
-
Every batch started with avoidable coordination overhead
Outcome: Order intake lacked structure and guidance, slowing everything that followed.
After: Guided Order Creation for Amazon Batches
Order creation was redesigned as a guided, state-aware workflow.
Instead of submitting a form and returning to a list, PMs now move through a structured flow, from validated intake to automated job generation, with clear system feedback at every step.
1.Orders Page, 2. Auto-filled Order Form, 3. Order Creation & Jobs Generation 4.Order Details Page & Success State
What changed
-
Clear visual hierarchy and simplified layout
-
Amazon-specific fields auto-filled and validated in real time
-
Explicit system states (loading → success confirmation)
-
Order creation automatically generates associated jobs
-
PM is redirected directly to the new order detail page
Why this mattered
-
Removed scanning and manual follow-up steps
-
Reduced setup time for every batch
-
Increased confidence through visible system feedback
-
Prevented downstream errors before production began
-
Turned intake into a repeatable, scalable entry point
Outcome: Order intake became structured, guided, and scalable, removing friction from the very first step.
2. AI Model Matching: Preventing Unnecessary Production Cost
As ARitize3D scaled to over 200,000 existing 3D models, identifying model duplicates became increasingly manual and time-consuming. QA teams searched the database by hand to determine whether a model already existed or if something similar could be adapted.
While this occasionally prevented redundant work, the process could take days for large enterprise batches. This manual process delayed job creation and tied up production capacity.
I designed and introduced a net-new AI-powered model matching workflow that scans incoming SKUs against the full model database before production work is assigned, shifting match detection from manual effort to system-level automation.
1.Match Results, 2. Enlarged Comparison, 3. Confirmation
What this introduced
-
AI scans new product requests against 200K+ existing 3D models
-
Flags potential duplicates before jobs are created
-
Introduces structured job types (new model, retexture, duplicate)
-
Integrates reuse decisions directly into the production workflow
-
Keeps humans in the loop through side-by-side validation
Why it mattered
-
Reduced unnecessary 3D modeling hours and AI processing
-
Lowered cost per asset as order volume scaled
-
Protected modeling capacity for net-new production
-
Enabled revenue from previously created assets
-
Prevented duplicate production before labor was allocated
Outcome: Duplicate production was systematically reduced, embedding cost-aware reuse directly into the enterprise pipeline.
Extending AI Model Matching Across the Production System
Model matching wasn’t limited to order intake.
I also designed a dedicated model search tool for 3D modelers, enabling them to scan the existing asset library directly from within their workflow.
When a job was assigned, artists could search against the 200,000+ model database to identify reusable models before beginning production.
Rather than rebuilding from scratch, they could:
-
Search multiple product images at once
-
Reuse an existing model
-
Modify a near match
This ensured cost savings weren’t dependent on one checkpoint.
Impact: Model matching evolved from a feature into production infrastructure, scaling cost savings across the entire pipeline.
3. AI Complexity Assignment: Automating Pricing & Workload Distribution
As production scaled, assigning model complexity became one of the most operationally heavy and strategically sensitive steps in the workflow.
Complexity directly determined pricing, production effort, and how work was allocated across modeling capacity. Yet the process was manual, fragmented, and dependent on coordination between QA, PMs, and production teams.
I redesigned this step as an in-flow, AI-powered pricing and effort assignment system, embedding it directly into order setup rather than treating it as a downstream task.
Before: Manual, Distributed Complexity Assignment
Despite being pricing-critical, complexity assignment lived outside the core order flow. It required CSV exports, manual SKU filtering, and coordination across QA and PMs before any production work could begin.
There was no system feedback that pricing was complete, only manual confirmation and follow-up.
What happened
-
PM downloaded a batch CSV from the portal
-
CSV was split and distributed manually to QA via Teams
-
QA copied SKUs into filters to review and assign complexity one by one
-
Completion status relied on manual confirmation from QA
-
PM then filtered by complexity to assign work
-
Lists were exported again and sent to the client externally via CSV
-
Price discrepancies were returned and manually updated SKU by SKU
Why it was a problem
-
Pricing assignment blocked job creation and revenue readiness
-
No system visibility into completion state
-
Manual CSV distribution created version risk
-
Pricing discrepancies with the client triggered repetitive rework loops
-
PM workload increased with every large batch
-
The process didn’t scale with Amazon-level volume
Outcome: Pricing decisions were slow, manual, and dependent on back-and-forth coordination delaying production before it even began.
After: In-Flow AI Complexity Assignment
I transformed complexity assignment from a manual, disconnected review task into an in-flow, AI-powered pricing system embedded directly within order setup.
Instead of downloading CSVs, splitting lists across QA, coordinating through Teams, and waiting for confirmation, the system now evaluates every SKU in the batch automatically. Within minutes, AI assigns a recommended complexity level, which directly determines pricing and production effort and surfaces it for immediate review.
PMs can validate, adjust if needed, and confirm in one continuous flow. Once approved, the order is clearly marked ready for export and job distribution.
This shifted complexity from a coordination-heavy bottleneck into a fast, system-driven pricing checkpoint.
1.Complexitiy Results, 2. Confirmation
What this introduced
-
AI-driven complexity recommendations across entire batches
-
Pricing decisions embedded directly into order setup
-
Human review layered into the same interface, not a separate step
-
Clear system states (assigning → reviewing → confirmed)
-
Automatic confirmation that the order is ready to move forward
Why this mattered
-
Removed a major operational bottleneck in a pricing-critical step
-
Reduced complexity assignment from days of coordination to minutes
-
Improved consistency in how pricing and effort were determined
-
Eliminated uncertainty about whether a batch was ready
-
Enabled faster, more predictable workload distribution
Outcome: Complexity assignment evolved from a manual, coordination-heavy task into a structured, system-driven pricing checkpoint, giving the team speed, clarity, and confidence at one of the most financially sensitive steps in the workflow.
Impact: Enabling Faster, More Reliable Production
By restructuring intake, embedding model matching earlier, and bringing complexity (pricing) into the flow, the production system became significantly more reliable.
Manual coordination across PMs, QA, and modeling teams was reduced. Duplicate production was caught before work began. Pricing decisions became faster and more consistent.
Most importantly, the team was able to increase production capacity without increasing headcount.
This wasn’t just feature-level improvement, it was a shift from a people-dependent workflow to a system-driven one.
How We Validated It
I worked closely with the PM responsible for Amazon intake to validate the changes in real workflows.
The impact showed up quickly:
-
Batches were prepared faster
-
Fewer manual follow-ups were needed
-
Complexity and pricing decisions were clearer and more consistent
-
The workflow was adopted across PM, QA, and modeling teams
The system reduced friction in daily operations, which is where real product impact lives.
Company-Reported Metrics
After rollout, the company reported the following improvements tied to the capabilities enabled by this redesign:
While I did not directly own measurement, this system redesign enabled the operational adoption of these capabilities at scale.
Reflection: Designing for Scale in Complex Systems
I was asked to redesign a fragmented, developer-led workflow with minimal documentation, no dedicated product support, and a two-week deadline. The requirements were ambiguous, but the impact was not. This system directly affected pricing, workload distribution, and production capacity.
To move quickly and responsibly, I stayed closely connected to the people using it every day. Conversations with PMs surfaced the real friction points. Ongoing collaboration with engineering and data ensured what I designed was technically feasible and scalable. Clear communication was critical. There was no room for misalignment.
With limited time, I focused on what would deliver the most operational clarity. I narrowed scope to the highest-impact enterprise workflow, structured the system around decision states, and embedded AI directly into the flow instead of layering it on top. It required creative thinking, practical tradeoffs, and confidence in shipping a meaningful improvement rather than waiting for perfection.
I’m proud of what we achieved. The metrics reflect real operational gains, but just as importantly, the team felt the difference. Work moved faster. Decisions were clearer. Coordination became simpler. The business gained capacity without adding headcount.
This project strengthened how I approach complex systems: stay close to users, align tightly with engineering, make ambiguity tangible, and design for operational reality. That mindset continues to shape how I lead and build today.
If Time and Resources Allowed
-
Map and instrument the full lifecycle from client intake through model publication and analytics
-
Bring remaining offline coordination steps (price discrepancy handling, job distribution refinements) fully into the platform
-
Introduce deeper capacity visibility and forecasting tools
-
Expand feature-level measurement to quantify cycle time and pricing accuracy improvements
-
Polish and refine the UI to further reduce cognitive load at scale
This project demonstrates how structured thinking and domain fluency can transform a fragmented workflow into a scalable production system, quickly and responsibly.
Thanks for Reading
Check out more of my projects, or reach out if you want to learn more about this one!
