Data Scientist applying machine learning, causal inference, and LLMs to understand human behavior and build decision-support tools. Former Bloomberg Philanthropies Data Scientist with experience analyzing large-scale behavioral data at major consumer platforms, including cloud systems with tens of terabytes of streaming data.
I developed an ML-powered application that compares a user’s responses to global value profiles from the World Values Survey to identify the closest country matches. The system combines statistical modeling, feature selection, and an LLM explanation layer to translate complex survey data into interpretable user insights.
Portfolio of Work
Causal Analysis of Conversion Funnel Behavior
Determined whether a key action in an e-commerce funnel (clicking a link to view additional insurance options) actually increased conversion or simply reflected already high-intent users. Using SQL and R, I applied k-nearest-neighbor statistical matching to construct a near-identical comparison group of users who did not take the action, enabling an apples-to-apples estimate of the action’s true causal impact on revenue and conversion.
This approach isolates behavioral effects that simple correlation or funnel metrics cannot capture, helping product teams understand which user actions genuinely drive revenue versus those that merely signal high-intent customers.
Equity Analysis of Climate Policy Using Scottish Deprivation Data
Data science project analyzing how the benefits of climate action policies are distributed across Scottish communities. Combines UK climate co-benefits modeling with the Scottish Index of Multiple Deprivation to examine equity implications of climate policy. Shortlisted as a top-10 entry in a national climate data challenge.
Feature ROI Simulator
Built an interactive ROI model linking user behavior changes to revenue outcomes. Decision trees and random forests identified key funnel drivers, and propensity-score causal inference estimated the true impact of product actions. The tool allows teams to simulate KPI improvements and prioritize features based on expected revenue impact.
Predict the number of users and revenue based on marketing spend
It's vital to know the expected impact of a marketing campaign to evaluate its success before, during and after. I created an interactive tool using Tableau and R allowing marketing managers to enter their marketing budget on a daily, weekly or monthly level and receive back a predicted number of users and revenue by channel type using polynomial regression, ARIMA modeling, and autocorrelation tests.
Natural Language Processing to predict Net Promoter scores
What a person says about your product can be just as important as how they score it on a net promoter score (NPS). Using natural language processing, I found which words are associated with high and low NPS scores using a bag of words model and random forest algorithm, achieving 97% accuracy in its predictive power. R code here.
Co-founder and Co-lead for Canada's longest running civic technology meetup
My work crafting the portfolio of Beta City led to creation of more than 20 projects that brought together private sector tech volunteers with area politicians and nonprofits to create lasting positive public impact. The work led to me being chosen as a 2018 Top 3 Finalist for Public Sector Digest's Open City Champion for Canada and one of the top ten municipal innovation from Federation for Canadian Municipalities, the largest public sector professional development organization in Canada.
Report writer for the City of Austin's Racial Profiling Report 2018-2019
Lead writer and data analyst for the first Racial Profiling Report for the City of Austin in 3 years for the Office of Police Oversight while working as a data scientist in a Bloomberg Philanthropies-funded iTeam embedded in the City of Austin. Combined Austin Police Department motor vehicle stop data with Census Bureau data to understand racial disparities in motor vehicle stops, citations, arrests, searches, and police discretion. Findings are now used as an annual performance metric for City Council's Equity focus, and report is released annually based on the template. Click read more to see report.
Marketing to those in need with data science
The City of Austin, Texas has a large federally-funded grant program to help low-income people stay in their homes through forgivable home repair loans. One of the challenges is making sure low-income homeowners know about the program and apply. I combined data from the City, Travis County, and the Census using ArcGIS and R to create a needs-based likelihood score for every home in Austin, to be used to market the services to those most in need. The goal was to provide high-touch marketing to those most in need in the form of hand-written letters or door-knocking campaigns.
Product Manage creation of a social benefit screening tool 'You Can Benefit'
When the mayor of the fifth largest city in Canada asked what the community could do to end poverty in that city in a generation, I decided to create and lead the product management of an open-source tool so people could determine their eligibility for 39 social benefit programs across federal, provincial, local, and nonprofit programs. The tool has been used more than 60,000 times since 2017 and was debuted by the City of Edmonton mayor in a major press conference for the End Poverty Edmonton initiative.
Product initiator and manager for open-source pedestrian counter
Urban planning is all about vibrancy, but we all know that vibrancy is very subjective. Urban planning guru Jan Gehl undertook rigorous research to find that the most quantifiable factor related to vibrancy in cities is pedestrian traffic. Instead of buying industry-standard $3K pedestrian counters, I led a team to create low-cost and non-privacy invasive pedestrian counters using infrared imaging and solar power. The project is open source (code here). See article about creation of the project.
Product manager for AR and VR apps to envision a future city
Worked with local augmented and virtual reality software developers to create two apps to help people in Canada's fifth largest city see the future of their city and provide their feedback. The VR app used Google Cardboard to create an immersive experience of a downtown block outfitted with smart city technologies for comment. The AR app allowed residents to see new public buildings through their phone before they were built.
Product Manager for the Pi-Ano, making a public piano a digital sensation
I created and led a project called OpenPianoYEG in Canada's fifth largest city to place 7 pianos in outdoor public spaces. One piano was placed in a downtown subway stop and was outfitted inside with a Raspberry Pi connected to a microphone and wifi. When people played the piano, it recorded the performance and automatically uploaded the recording to Soundcloud. Thousands of performance were recorded, creating the world's first digital public piano, using open-source technology. Details and code can be found here.
Professional Photographer and Musician
When not working, I take photos for theater and dance groups. As a former photojournalist (sample photos here), I appreciate being able to stay engaged in my craft. Sample of dance photos.
I am also a multi-instrumentalist, recording artist and songwriter represented on Capital City Records (link to album here).
Contact
Please get in touch to find out more about me and my work. You can contact me at davidwrauch[at]gmail[dot]com
New York City
