I build end-to-end data systems that turn messy real-world data into decisions about where to focus, who to prioritize, and how to allocate limited resources.
Previously built systems used across:
→ $16B+ in modeled fundraising portfolios
→ 500K+ leads scored in production CRM pipelines
→ 150K+ voter outreach interactions across 13 U.S. states and 5 Canadian provinces, spanning 1200+ communities
Most of my work lives at the intersection of:
→ analytics engineering
→ applied modeling
→ product thinking
Often in environments where the data is incomplete, biased, or operationally messy (fundraising, civic tech, campaigns, early-stage startups).
In practice, much of my work involves building data systems that help organizations decide who to ask, how much to ask, and when to ask — whether the audience is donors, voters, or customers.
End-to-end data pipeline and targeting system for a national voter outreach campaign across 13 U.S. swing states, uncovering previously unknown outreach locations and shaped targeting and deployment strategy across 15,000+ engagement sites.
→ designed and operationalized the geospatial targeting pipeline (Google Places → Census enrichment → BigQuery)
→ partnered with messaging analytics pipeline (ThruText → S3 → GCS → BigQuery) to enable downstream reporting
→ developed a service density KPI to quantify outreach coverage vs real-world infrastructure
→ identified 400K+ locations and prioritized high-impact communities for outreach
🔗 https://github.com/k10sj02/barbershop-voter-engagement-analytics
Donor propensity scoring system for nonprofits modeled on real fundraising workflows, built to prioritize outreach toward highest value donors most likely to give again.
→ engineered 15+ RFM and donor profile features from transactional data to capture giving lifecycle behavior
→ trained Random Forest classifier (ROC-AUC 0.87, 2.5x lift at top 33%) to predict likelihood of repeat giving
→ designed four-tier segmentation framework (High / Medium / Low / Very Low) with actionable outreach guidance per tier
→ deployed as an interactive Streamlit app with CSV ingestion, column mapping, live filtering, and export for fundraising teams
🔗 https://github.com/k10sj02/nonprofit-donor-scoring
Production-style partner data ingestion pipeline for a voter registration reporting model, handling the full transformation lifecycle from raw extract to unified reporting table.
→ designed a three-layer architecture (raw → staging → mart) with an auditable, non-destructive working copy at each stage
→ built deterministic validation rules for contact data (email, ZIP, NANP phone) and demographic bounds (age 18–105, recency constraints)
→ engineered county enrichment via ZIP lookup using LEFT JOIN to preserve record fidelity over silent data loss
→ implemented window-function deduplication with explicit tie-breaking logic (Complete status → recency → email)
→ produced schema-aligned UNION ALL integration with semantic field mapping and UUID surrogate keys for lineage tracking
🔗 https://github.com/k10sj02/voter-reg-pipeline
Predictive modeling and validation study on how social norms shape behavior, highlighting the limits of prediction in complex, real-world data.
→ trained classification models (Random Forest, Logistic Regression) on attitudinal survey data
→ identified a hard performance ceiling (~0.65 AUC) and investigated underlying causes of limited predictability
→ traced constraints to survey design, measurement limitations, and noise in self-reported attitudes
→ paired modeling with a literature review to contextualize results within social science research
→ deployed an interactive app to surface predictions and feature-level drivers
🔗 https://github.com/k10sj02/gender-norms-predictor
- building systems, not one-off analysis
- turning data into decisions, not dashboards
- designing metrics that actually reflect real-world behavior
- working in imperfect, real-world data environments
Core: SQL, Python
Warehousing: BigQuery, Snowflake, PostgreSQL, dbt
Apps & Viz: Streamlit, Tableau, Power BI, Looker
Infra: GCP, AWS, Docker, Git
Portfolio: https://stannomarjones.com
LinkedIn: https://linkedin.com/in/stannomarjones