The opportunity
The Statsig team within OpenAI builds the experimentation, feature rollout, dynamic configuration, and analytics systems that help OpenAI ship products with speed, safety, and evidence. Our work sits on the critical path for how product, engineering, research, and go-to-market…
What you'll do
Set and execute the technical roadmap for Generative Insights and Predictive: Experimentation, from early prototypes through production adoption.
Build cross-experiment learning systems that retrieve and synthesize: historical experiments, detect recurring effects and segment behavior, reanalyze prior results when data or methods improve, and generate hypotheses with clear evidence and provenance.
Develop predictive models and simulation workflows, including: simulation-based evaluation approaches, to estimate likely impact, affected segments, regression risk, and uncertainty before a full live experiment.
Create high-quality datasets and feature or retrieval pipelines from: exposures, events, metrics, experiment metadata, and replay data, with strong lineage, freshness, privacy, and data-quality controls.
Establish rigorous evaluation through offline benchmarks, backtests,: calibration, drift monitoring, prediction-to-outcome comparisons, and explicit failure or abstention behavior.
Turn models into durable product, API, and agent workflows that move from an: insight to experiment design, approval-gated action, and measured learning.
What they're looking for
- Partner deeply with data science and product teams on experiment design,: causal inference, sequential decision-making, variance reduction, and the boundary between prediction and causal evidence.
- Build reliable services and intuitive workflows so sophisticated ML: capabilities are understandable and useful to teams making high-stakes product decisions.
- Provide technical leadership across engineering, product, data science, and: research partners, and raise the bar for production ML quality across the platform.
- Have led ambiguous 0-to-1 production ML products where success was measured: by better real-world decisions, not only offline model metrics.
