Portfolio · Amazon · Enterprise UX

Designing for systems that cannot fail.

Four case studies from Amazon – each addressing high-stakes, large-scale operational complexity for a global workforce of 1.2 million associates.

Case Study 01

WAGE — Wage Analytics & Guidance Engine

Transforming Amazon’s compensation decision-making from a 2–3 month manual cycle into a real-time, AI-guided platform serving 1.2M hourly associates.

The Problem

Amazon manages compensation for over 1 million L1–L3 hourly employees. A single $0.50 wage adjustment translates to an $800M annualized investment — yet the network wage review decision process was broken. Each team used different aggregations, different interpretations, and different tools.

“What used to require extensive meetings and manual analysis can now be accomplished in minutes.” — John Tagawa, VP of North America Operations

Research

• The bottleneck was not data availability but synthesis — nothing connected data into a single decision-ready view
• Leaders needed directional confidence — AI-guided recommendations they could challenge, not raw data
• Approval anxiety was high: the fear of making a wrong call at scale was paralyzing
• A single off-cycle wage review averaged 14 discrete handoffs and takes about 3 weeks

Design

I structured the design around three core tenets: Data Visibility, Assisted Decision, and Frictionless Approval. A consolidated interface surfaced base wages, shift differentials, premiums, incentives, KPIs, and market benchmarks together in real time.

• Node-level and site-level views with clear drill-down paths
• AI summaries lead with the “so what” — connecting metrics to operational impact first
• Scenario modeling with side-by-side comparison and automated impact analysis
• Compression and ripple effects surfaced upfront, not discovered after decisions

Outcomes

Case Study 02

VCM — Variable Compensation Management

Replacing a 2009 PeopleSoft legacy system with a purpose-built platform for managing variable compensation across 43 plans in 26 countries — eliminating 1,500 annual defects.

The Problem

Amazon’s variable compensation was managed through PeopleSoft’s VCE, built in 2009 and never designed for Amazon’s scale. PeopleSoft could only assign either a sales incentive plan or a local allowance to an employee — but not both. In countries like Mexico, Argentina, and Uruguay where employees are legally entitled to multiple allowances, teams manually hardcoded workarounds requiring monthly audits.

In 2023, a single misconfigured rule incorrectly assigned variable compensation plans to 1.4 million employees over two days. Correcting it required legal involvement and taking other systems offline.

Research & Design

The fundamental insight: the problem was not UI polish but data modeling. Comp admins needed confidence before committing changes — safety features were not optional, they were the core UX requirement.

• Simulation as mandatory gate — no plan submittable without viewing projected employee impact first
• 2-step approval with unique approvers — no self-approval allowed
• Rollback treated as an edit — same approval workflow, full auditability preserved
• Tenure auto-updates: Morocco’s graduated allowances configured once, applied automatically

Outcomes

Case Study 03

ACE Control Center

A single administrative platform replacing PeopleSoft, Excel calculators, and engineering-dependent workflows — enabling non-technical comp admins to manage policy changes globally.

The Problem

Every Amazon compensation event — annual review, promotion, transfer — depends on configurable inputs: stock prices, exchange rates, haircut policies, formulas, and eligibility rules. In 2024, none of these had a single administrative home. Formula changes in Excel had no version control. Eligibility changes required engineering to update backend code — average turnaround was weeks.

An inaccurate entry in PeopleSoft would automatically propagate to compensation journey products, potentially affecting millions of employees before anyone noticed.

Design

Control Center was designed as a multi-tenant platform — not a monolithic application. VCM and future GTMC teams could build their own administrative modules independently. The design centrepiece: simulation as a mandatory gate, rollback as a first-class operation, and rule transparency that translates backend compensation code into plain language any manager can understand.

• Simulation mandatory — no change submittable without reviewing projected impact
• Rollback activated with the same approval workflow — normal operation, not emergency
• Tiered permissions layered by blast radius — not binary admin/read-only
• Zero engineering dependency for standard control data updates

Outcomes

Case Study 04

Shift Differential Analysis

AI-assisted self-service analytics for shift differential compensation decisions — replacing 700+ hours of annual manual analysis across 72 recurring requests.

The Problem

Amazon’s shift differential system compensates for shift burden across all US Operations sites using a 14-category matrix. Each of the ~72 annual SD review requests involved 3–5 people and several hours of manual analysis — pulling data from HAC, G&C, and OCEANS — with no standard methodology and no self-service capability for L8+ leaders.

Design

I designed an AI summary system that weaves shift differential status directly into the compensation health narrative — integrated with fill/attrition analysis, not siloed. Site-level heat maps show SD status relative to node standard. A decision tree guides users from matrix category classification through headcount thresholds and health checks to APPROVE/DENY/MANUAL REVIEW outcomes.

• Health status always includes SD status (Below/At/Above Node Standard)
• Fill rate and attrition thresholds user-adjustable — different nodes have different operational contexts
• Rule (R) / Exception (E) indicators with color-coded risk: Red (<90% fill AND >2% attrition), Yellow, Green
• Filter selections and view state persist during session navigation

Outcomes