Overview

This role is about using behavioral data to predict what premium consumers want next — and doing it with enough precision that both our B2C concierge product and our B2B enterprise customers can act on the output. You'll own the predictive modeling stack end-to-end: feature engineering, model development, experimentation, deployment, and measurement. The work spans ranking, propensity, lifetime value, churn, next-best-action, and recommendation systems across a consumer base that skews high-intent and high-value.

Responsibilities

Build, train, and deploy predictive models across the consumer lifecycle — acquisition, activation, engagement, conversion, retention, and expansion.
Develop recommendation and ranking systems that power the concierge experience: what to suggest, when to suggest it, and how to order the options.
Own feature engineering on behavioral, transactional, geographic, and temporal signals — and build the feature store that makes those features reusable.
Design and run experiments (A/B, multi-armed bandits, uplift modeling) to measure model impact on the metrics that matter: conversion, AOV, repeat rate, and LTV.
Productionize models with proper monitoring — drift detection, performance alerts, and retraining pipelines — so nothing silently decays.
Partner with the platform engineer to ensure ingestion outputs match modeling needs, and with product to translate business questions into modeling problems.
Contribute to the B2B data product: build the propensity and segmentation outputs that enterprise customers will consume through our APIs.

Qualifications

4+ years building and shipping predictive models in production — consumer, marketplace, or commerce contexts preferred.
Deep command of classical ML (gradient boosting, regression, classification) plus hands-on experience with deep learning for tabular and sequential data.
Experience with recommender systems: collaborative filtering, matrix factorization, two-tower models, or LLM-assisted retrieval.
Strong experimentation instincts — you know when a lift is real, when it's noise, and how to design the test that tells you the difference.
Fluency in Python and its ML stack (PyTorch, scikit-learn, XGBoost/LightGBM, pandas, numpy); comfortable with SQL at depth.
Production MLOps experience: feature stores (Feast, Tecton, or homegrown), model registries, CI/CD for models, and observability (Evidently, WhyLabs, or equivalent).
Clear thinking about causal inference and the difference between predicting behavior and changing it.

Nice to have

Experience with LTV modeling, uplift modeling, or causal forests.
Background in marketplaces, travel, hospitality, luxury retail, or subscription consumer products.
Work with geospatial data or time-series forecasting.
Publications, blog posts, or open-source contributions that show how you think.