Lead Principal Technical Program Manager

Opening Statement (~60 sec)

I owned the cross-functional program to fix a broken ad-deduplication system inside Amazon's Prime Video and Freevee AdTech stack — a system I'd inherited as 'already solved' but that was actually catching only ~2% of true ad repetition in production, with CSAT down 8% over six months and the issue escalated to board level%%. Rather than keep reporting the gaps as tuning edge cases, I became the escalation point myself: I named the root cause publicly, before anyone directed me to, and ran the three-team fix — ML Science, AdTech Infrastructure, and Ad Ops — through to a validated global launch. Detection went from 2% to 98.5% accuracy across 17 markets, recovering $87M+ in annualized revenue, and the review process I introduced to capture the original team's trade-offs was adopted by two other programs as standard practice.

Situation

1. I inherited a Creative ID-based ad-deduplication system the org considered closed, but it was matching ads by exact file ID and couldn't recognize a re-uploaded creative or a 15s/30s cut of the same commercial as duplicates. 2. Viewers were seeing the same ad up to five times an hour, CSAT had dropped 8% over six months, and the issue had already escalated to board level based on customer-facing signals alone — before I had named the internal root cause.

Task

1. My task was to run the technical diagnosis to confirm whether the system itself — not advertiser behavior — was the root cause, then decide, and own, what to do about it. 2. A VP gave me direct feedback that I'd been framing a structural failure as a tuning gap to protect stakeholder comfort — my task shifted to calling the diagnosis clearly and leading a cross-team rebuild, without an existing mandate to redesign a system the org had already accepted as solved.

Action

1. 1.Ran an independent root-cause analysis, outside my assigned mandate, and documented the structural failure modes rather than continuing to call it a tuning issue.
2. 2.Reframed the problem publicly at the next steering committee, walking the engineering leads who had built the original system through the failure data so the conversation stayed evidence-driven rather than defensive.
3. 3.Structured the fix as a cross-team co-design — two weeks of joint working sessions with ML Science, AdTech Infrastructure, and Ad Ops — instead of handing down a top-down spec.
4. 4.Ran a dedicated Architecture Decision Record (ADR) session with the infrastructure team specifically to capture the trade-offs they hadn't been consulted on before the pivot, converting a team with reason to be defensive into a co-owner of the replacement.
5. 5.Sequenced delivery to rebuild trust before asking for it: shipped a measurement-only phase first — profiling every ad and publishing similarity scores with no enforcement — so skeptical stakeholders could validate the new approach against the old system's blind spots before anything went live.

Result

1. True ad-repetition detection rose from ~2% to ~40% in the measurement-only phase alone, making the business case undeniable before any filtering shipped. 2. The rebuilt system reached 98.5% global detection accuracy across 17 markets, recovering $87M+ in annualized revenue. 3. The infrastructure team that built the original system became active co-owners, and the ADR process I introduced was adopted by two other programs as standard practice for mid-program architecture pivots.

Closing Statement (~60 sec)

What made this an escalation-ownership problem, not just a technical one, was that I had the data before I had the resolve to call it — I'd been softening a structural finding into a comfortable narrative. The change I made afterward wasn't just the fix; it was the ADR practice itself, so the next team facing a similar pivot had a formal way to register trade-offs instead of going quiet and defensive. That's the model I'd bring to being OCI's escalation point: report the root cause, not the most comfortable interim explanation, and build the mechanism that keeps the same failure from recurring — not just patch the one instance.