Every piece starts with a question
Samples of real work across data engineering, analytics, and marketing intelligence: warehouse architecture, identity resolution, attribution, BQML value modeling, market analysis, and a crowdfunding launch I ran myself.
Raw events to revenue
The whole architecture on one page: raw sources through staging, Kimball marts, identity, BQML, and attribution, ending in ad platforms bidding on modeled value. One pattern, run at 20M and 200M rows.
Identity resolution
48,117 identities unified for a service contractor, and a probabilistic layer that reconnected 89% of the pipeline value a global enterprise's CRM never saw. Same discipline, two scales.
Attribution recovery
Warehouse attribution took a contractor's paid-revenue visibility from 8.8% to 26.5% of won revenue, and stage-level data-driven models re-priced every channel's credit against the linear baseline for an enterprise funnel.
Value-based bidding
A BQML model prices every lead before sales touches it. Top-decile leads convert at 33x the bottom, so Google bids on modeled value instead of raw lead counts.
The Amazon diagnosis
A brand that owned its Amazon niche kept failing off-platform. 28M rows across 16 sources showed why: the market had moved, and competition was the story.
Five forces, measured
Porter's five forces scored from 28M rows of consolidated market data. The structural read explained why a healthy Amazon brand kept losing ground everywhere else.
Decisions from data
45K orders re-sequenced into buyer journeys surfaced numbers a genealogy education company had never seen: premium-entry buyers carry roughly 7x the lifetime value, and 16% of buyers drive 48% of revenue.
The crowdfunding pre-launch
A $13.4K pre-launch engine, 4,277 leads at $3.06 each distilled into a VIP core, that funded a Kickstarter 2,070% over goal.