Senior Data Scientist (Contract): Churn, CLV, Propensity & Demand Models, EDA to Production

WHO WE ARE devx labs is an AI-native consulting company that fundamentally reimagines business operations around AI capabilities. We don't just implement AI tools, we transform how enterprises operate through first-principles thinking and outcome-driven execution. ABOUT THE ROLE You own the modelling layer of client engagements: turning customer, transaction and operational data into propensity, value, demand and pricing models that drive decisions across Customer Interactions and AI-Led Business Operations. Clients span retail, D2C, telecom, financial services and manufacturing. The problem changes every few months; the discipline doesn't: frame the question, prove the signal exists, ship a model that holds in production, show the business the number moved. Not a notebook-only role. You are accountable for the model from EDA through deployment, monitoring and the A/B read that proves it worked. PROBLEMS YOU'LL WORK ON CLV and customer health scoring; churn, retention and win-back propensity; upsell, cross-sell and Next Best Action; segmentation and behavioural clustering; SKU, store and channel demand forecasting; price elasticity and markdown optimisation; uplift modelling; anomaly and fraud scoring; attribution; recommendation and personalisation. RESPONSIBILITIES Build and Ship Models (60%) - Frame the problem: target, label definition, unit of analysis and success metric before writing code - Own the data work: EDA, data quality, feature engineering and leakage checks in SQL and Python - Build the model: GBMs, GLMs, survival models, clustering, time-series, and deep learning where it earns its place - Evaluate honestly: out-of-time validation, calibration, lift/gain curves, business backtests. A good AUC on a leaky split is a failure. - Deploy and operate: batch or API serving on Vertex AI or SageMaker, MLflow tracking, drift monitoring - Prove impact: design and read A/B tests, uplift tests and holdouts - Explain the model: SHAP, partial dependence and segment diagnostics Drive Client Success (30%) - Join discovery workshops, assess data readiness and shape the analytics approach - Present results clearly: what the model does, where it's uncertain, how to act on it - Work with client teams to land models in CRM, marketing, planning or pricing workflows - Track adoption post-launch and recalibrate as the business changes Multiply the Practice (10%) - Build accelerators: feature libraries, evaluation templates, model cards, playbooks - Mentor junior team members (DS 2 expectation) - Use Claude Code, Cursor or similar to move faster while owning correctness REQUIREMENTS - Rigour first: you distrust a result until you've checked the split, the leakage and the baseline - Stats and ML fundamentals, including causal basics (uplift, DiD, propensity matching) - Production experience with at least two of: CLV, churn, propensity, segmentation, demand forecasting, pricing or uplift. Kaggle-only or coursework-only doesn't clear the bar. - Python (pandas, scikit-learn, XGBoost/LightGBM/CatBoost, statsmodels) and strong SQL on BigQuery, Redshift, Snowflake or Databricks - Working knowledge of PyTorch or TensorFlow, and judgement on when DL is warranted - Has taken a model to production with experiment tracking, versioning and monitoring - PySpark or Dask; comfort with messy, multi-source enterprise data - SHAP/LIME, fairness checks and clear model documentation - Data storytelling: turn a confusion matrix into a decision a CMO can act on GOOD TO HAVE LLM APIs or embeddings, optimisation (LP/IP, bandits), marketing mix modelling or Bayesian methods, retail/D2C/telecom/BFSI/manufacturing exposure, GCP ML Engineer or AWS ML Specialty certification. EXPERIENCE - DS 1: 2-3 years hands-on, at least one model shipped to production - DS 2: 3-5 years, multiple models shipped and owned end to end, plus some mentoring or project ownership - Degree in Statistics, Maths, CS, Economics, Engineering or related field - Experience working directly with business stakeholders HOW TO APPLY Start your proposal with "HOLDOUT". Include: (1) one production model you built, how you validated it and its business impact, (2) a GitHub repo, case study or write-up, (3) whether you're applying at DS 1 or DS 2 level, your hourly rate and weekly availability.