Karen Ostdiek, Ph.D.
About
I work at the intersection of marketing analytics, attribution, and data modeling—helping teams invest smarter with rigorous analysis and trustworthy pipelines.
Hello and welcome. I’m a Principal Business Intelligence Analyst based in Boston, Massachusetts, focused on marketing and product analytics—and I’ve been taking on more machine learning work as well. I build attribution models, ROI reporting, and the Snowflake/dbt pipelines that make those answers reproducible.
One recent ML project: a social listening relevance model. Our feed had a lot of noise (for example, competitors whose names are common words). I used an LLM to generate truth labels on a subset of posts, then trained a TF–IDF + logistic regression classifier to label the rest of the dataset and score new data as it arrives—so the team gets cleaner signal without hand-labeling everything.
Before my current role, I worked in health-plan risk adjustment—reconciliation, regulated submissions, and analytics in Snowflake, dbt, Airflow, and Python. My foundation is a Ph.D. in nuclear physics from Notre Dame: experimental design, careful measurement, and communicating results to both experts and general audiences.
Most of my recent work lives in private repositories; below you’ll find anonymized project summaries. My public code is on GitHub. I’m always happy to connect on LinkedIn or via the contact form.
Thank you for visiting.
Capabilities
A deliberately unscientific chart—because humility and humor belong on a résumé too.
Selected work
Representative themes from my work. Details are anonymized where code and data are proprietary.
Social listening data was cluttered with irrelevant mentions—e.g. competitors whose names are common words.
Created gold-standard labels on a subset using an LLM, then trained a TF–IDF + logistic regression model to classify the full historical set and score incoming posts automatically.
Outcome: Higher-precision relevance signal for reporting and follow-up without labeling every row by hand.
Stakeholders needed a consistent view of how paid and lifecycle campaigns translated into revenue.
Built and iterated ROI reporting and a rebuilt attribution data model in Snowflake and dbt (with modern git-based workflows), aligning analytics to a major CRM transition.
Outcome: More credible campaign decisions and a durable foundation for downstream reporting.
The business needed an agreed revenue forecast for the SMS line of business to align company-wide planning.
Partnered with stakeholders to develop and socialize forecasts grounded in usage, churn, and upgrade behavior—translating complex time series into numbers leadership could adopt as targets.
Outcome: Forecast adopted as a primary planning benchmark.
Submissions for government programs required tight reconciliation between operational systems and reporting.
Maintained analytical layers in Snowflake; collaborated across engineering and data science on pipelines (GitHub, CircleCI, Airflow, dbt) and surfaced issues through dashboards and deep dives.
Outcome: More reliable submissions and faster insight into risk-adjustment performance.
Background
Klaviyo · Boston, MA
April 2023 – present
Klaviyo · Boston, MA
October 2021 – April 2023
Devoted Health Plans · Remote / Watertown, MA
August 2019 – September 2021
PacificSource Health Plans · Eugene, OR
April 2017 – August 2019
April 2017 – August 2019
United Way Day of Caring and company garden work—produce donated to a local food bank.
2016 – 2017
Edited a county Emergency Operations Plan for grammar, spelling, and style.
University of Notre Dame · Notre Dame, IN
August 2016
2010 – 2016
Nuclear Science Lab tour guide for students, scouts, and officials. Coordinator for Physics Olympics across universities.
University of Wisconsin–Stevens Point · Stevens Point, WI
May 2009
2007 – 2009 · University of Wisconsin–Stevens Point
Contact
Send a message below: