Kelun Wang
  • Home
  • Resume
  • Work

Work

Selected work · 2026

Projects built around decisions, not dashboards.

Marketing analytics, experimentation, causal inference, and statistical modeling. Every case includes the question, method, result, and what I would recommend next.

07 projects
Python · R · SQL

01 / Causal targeting

Who should see the ad?

Uplift modeling separates persuadable players from customers who would convert anyway, then optimizes audience depth for incremental profit.

$59,161maximum projected
incremental profit
Uplift modeling · XGBoost · profit optimization
View case study ↗ Books, code, and customer paths illustrating uplift targeting
02 / Lifecycle marketing

QuickBooks upgrade targeting

Model selection based on campaign economics, not AUC alone.

$377Kprojected population profit
Logistic regression · MLP · feature engineering
Open project ↗ QuickBooks customer messages and upgrade targeting scene
03 / Mobile targeting

Send fewer messages. Earn more.

Turning the hidden cost of customer messaging into a profitable rollout policy.

7.9×realized profit lift
Response modeling · campaign economics · ROME
Open project ↗ Tuango restaurant, entertainment, and mobile marketing scene
04 / Experimentation

A/B testing, explained from first principles

Simulation-based intuition for estimation, uncertainty, the bootstrap, and peeking risk.

Frequentist statistics · simulation · communication
Read explainer ↗ Code window and balanced A/B experiment groups
05 / Econometrics

Minimum wage and employment

A replication of Card and Krueger's natural experiment in fast-food restaurants.

Difference-in-differences · regression · replication
Read replication ↗ New Jersey and Pennsylvania restaurant comparison
06 / Count modeling

Software and patent awards

Estimating Blueprinty's association with patent outcomes using Poisson regression.

Maximum likelihood · Poisson GLM · counterfactuals
Open analysis ↗ Blueprint, patent document, and software count modeling scene
07 / Choice modeling

How students rank the MSBA core

Three approaches to turning Best-Worst choices into relative preference estimates.

MaxDiff · MNL · Bayesian estimation
Open analysis ↗ Preference survey cards and ranked course choices

Interested in how I think?

The resume gives the short version. These projects are the evidence behind it.

View resume ↗ Start a conversation ↗

Kelun Wang

Technical portfolio for analytics, experimentation, and product-minded data work.

LinkedIn · GitHub