Redesigning all-mile planning at Amazon Japan

Context

A rapidly scaling logistics program required a stronger planning operating model across all logistics miles and weekly, campaign, and long-term horizons. Data collection, analysis, forecasting, planning transformations, routing inputs, stakeholder decisions, and execution controls needed to work as one connected system.

What I did

As the single-threaded owner of planning across all miles and horizons, I redesigned the workflow connecting data collection, analysis, forecasting, planning transformations, and integration with an existing routing tool.

I translated operating requirements into executable business rules and a connected suite of deterministic Python tools that I use regularly. Weekly planning includes rolling 13-week seller forecasts and planning outputs for every mile. Campaign planning combines historical baselines, reusable peak factors, and commercial uplifts, with performance tracking that informs controlled improvements for subsequent campaigns. Long-term tools connect topology, volume scenarios, order–destination mix, and sort-center capacity to support faster iteration.

The workflows automate validation and corrections for known issues while stopping unresolved exceptions for human review. Operational findings are translated into reviewed changes to rules and tools rather than being applied automatically.

Kiro supports the development process through requirements, specifications, documentation, and implementation. AI assists how the tools are designed and changed; the production calculations themselves remain deterministic and do not depend on runtime AI. For controlled changes, an agent validates the reported issue, reads the existing code, and produces a specification and implementation plan before implementation proceeds.

The connected tools are in regular operational use by me. Broader team distribution is still in progress, so this work does not yet represent independent team-wide adoption.

Impact

Across the wider planning program, the business outcomes during this period included:

• Supported significant program growth to multi-million-package scale.
• Substantially reduced First Mile cost per package as the program scaled.
• Improved First Mile forecast accuracy by more than 20 percentage points.
• Maintained downstream plan-versus-actual performance within its target range.

What this demonstrates

The ability to own a complex planning operating model, align stakeholders around shared mechanisms, and extend product and operations leadership into hands-on implementation through AI-assisted development—without attributing the wider business outcomes to the tools alone.