Four detailed accounts of complex product and programme challenges — the problem defined, the approach taken, the solution built, and the impact delivered.
Building a fair, ML-powered enforcement system that protects Amazon from bad-debt while delivering the right experience to millions of legitimate customers.
Amazon's existing enforcement approach treated all flagged customers the same — leading to false-positive enforcement on legitimate buyers, poor customer experience, and missed bad-debt recovery. A more nuanced, ML-driven system was needed.
Authored a comprehensive 3-year product vision document, designed the ML customer behaviour model architecture, secured executive funding for 6 SDEs + 2 ML Scientists, and built a 25+ team governance model across engineering, legal, data science, and operations.
Delivered $26M in DSI savings in 2023-24, reduced false-positive enforcement for 2.27M customers, achieved 4800 bps YoY coverage improvement, and built a roadmap to $110M bad-debt reduction by 2027.
Converting a static, manual knowledge base into a dynamic, automated self-service platform — driving 145% usage growth in under a month.
Amazon's internal Knowledge Centre was difficult to navigate, relied heavily on manual email handling, and had poor response quality metrics. Agents couldn't find information efficiently, leading to incorrect responses and repeat customer contacts.
Designed and launched KC Navigator — a structured navigation system for 119 policies and 200+ blurbs. Implemented a systematic blurb optimisation programme driven by PRR data, and built an automation layer to convert email handling into self-service workflows.
KC usage increased 145% within 4 weeks. PRR improved 600 bps, repeat PRR +248 bps, understandability +124 bps. Automation coverage increased 2030 bps, dramatically reducing manual email handling cost.
Expanding Amazon's customer service infrastructure to serve millions of users in their native languages — at enterprise scale and speed.
As Amazon India expanded into Tier 2 and Tier 3 cities, a growing portion of customers preferred to interact in vernacular languages (Telugu, Tamil, Kannada, Malayalam). Existing customer service infrastructure was English/Hindi-centric, creating a significant experience gap.
Designed a comprehensive readiness programme covering contact routing, agent training, knowledge content creation, quality frameworks, and technology integration for four simultaneous vernacular language streams — each serving 300K+ daily users.
Successfully launched customer service readiness across Telugu, Tamil, Kannada, and Malayalam — each with 300K+ daily users — establishing a repeatable playbook for future language expansion across Amazon India.
Influencing executive decision-making with a rigorous business case to fund a transformative AI/ML-powered enforcement platform — without direct budget authority.
Scaling Moderated Enforcement to its full potential required significant engineering and ML investment — 6 SDEs and 2 ML Scientists — that needed to be justified through a compelling business case to Amazon's senior leadership in a highly competitive resource environment.
Built a comprehensive BRD and 3-year product vision, anchored in hypothesis study results, customer data, and modelled financial projections. Presented a credible path to $220M bad-debt reduction with clearly defined milestones, risk mitigation strategies, and ROI analysis.
Secured full executive approval for 6 SDEs and 2 ML Scientists. Programme on track to deliver $220M in bad-debt reduction with the ML customer behaviour model now under active development — alongside $110M annual bad-debt target by 2027.
Each case study represents months of complex work across multiple teams. I'm happy to walk through the details, decision points, and learnings in a conversation.
Schedule a Conversation