$172M+ of receipts.

Case 1

Saved the platform. Then it became the core.

T-Mobile's Next Best Action system was at risk of decommission. I rebuilt the seam between AI outputs and agent incentives, not the model.

→ $26M CLV uplift · +37% NPS · 40M decisions/month at 250ms

Case 2

AI caught what audits missed.

Built an LLM+ML risk detection system connecting contracts and invoices to financial databases at Comcast.

→ $140M exposure surfaced · <5% false positives

Case 3

18 million decisions a month, one operating model.

Led the global data science team overhauling SiriusXM CRM decisioning.

→ +5pt save rate · +3pt CLV · forecasting at ~10% MAPE

Case 4

Killing the tests that can't produce decisions.

Built an MDE-based experimentation gate: tests that couldn't reach a decision never launched. Prioritization became objective, not political.

→ 35% faster time-to-decision · enterprise-wide adoption. Live now: the Experimentation Toolkit on this site — same logic, free to use.

Case 5

Dashboards show data. This made decisions.

Built an insight engine that decomposes variance and recommends the action.

→ 4x faster SLT decisions · 20% cancellation improvement

The pattern: enterprise AI fails at the seams — between model and workflow, data and decision, incentive and output. I build the systems that close those seams.

MIT-educated · Fortune 20 background · now leading AI & analytics strategy at NiCE.