Senior Data Scientist -- Clean Cooking PAYG LPG

What Success Looks like

• Your work will be measured against the commercial metrics of the PAYG LPG business, not model metrics alone. In your first 12 months you will be expected to:
• Ship 2–3 data products into production use by commercial or operations teams, each with a measured impact on at least one of: activation rate, refill frequency, dormancy/churn, ARPU or cost-to-serve.
• Establish the BU's approach to pricing and promotion analysis — elasticity estimates, uplift measurement and experiment design that commercial leaders actually use to make decisions.
• Build monitoring for the models you ship, with agreed retraining triggers and a clear owner for each.

What you will be expected to do:

Commercial

• Develop first-hand understanding of customer and commercial needs across our markets, including time in the field with sales agents and customers.
• Translate business problems into well-framed statistical questions, and present findings clearly to both technical and non-technical stakeholders — up to BU leadership.
• Prioritise ruthlessly: identify where a data product will move activation, refill frequency, retention or unit economics, and be willing to say where it won't.
• Manage stakeholder expectations, product scope and delivery timelines for your own workstreams.

Technical

• Design, build and evaluate machine learning models for business-critical use cases: churn/dormancy prediction, credit and payment-behaviour modelling, demand forecasting, anomaly detection and customer segmentation.
• Apply probabilistic and Bayesian methods to quantify uncertainty and support decisions under uncertainty — e.g. pricing elasticity, promotion uplift and media/marketing effectiveness.
• Design and analyse experiments (A/B tests, geo tests, quasi-experiments) in field conditions where clean randomisation is often impossible.
• Perform rigorous exploratory analysis, feature engineering and data wrangling on large structured and semi-structured datasets.
• Partner with data and analytics engineering, who own pipelines and production infrastructure: you own the model from problem framing through validated, deployment-ready handoff, and jointly own monitoring once live.
• Track and communicate model performance; identify degradation and recommend retraining or redesign.
• Maintain clean, reproducible, well-documented code following team engineering standards.

What this role is not about:

• Not a people-management role — this is a senior IC position (a path to leading a small team may open as the function grows).
• Not an MLOps/platform role — you'll work to production standards, but pipeline and deployment infrastructure is owned by MLOps Engineer.
• Not a reporting/BI role — dashboarding exists in the analytics team; this role builds models and data products.

You might be a strong candidate if you have:

• Degree in Computer Science, Statistics, Mathematics, Engineering, Economics or a closely related quantitative discipline. An advanced degree is a plus, not a requirement — evidence of shipped impact matters more.

Commercial

• A demonstrable track record of data products that measurably moved a business outcome — you can walk us through the problem, the model, the decision it changed and the number it moved.
• Ability to listen to and empathise with customers and colleagues, and to identify the P&L impact of a proposed data product before building it.
• Strong communication and storytelling: you can carry a room of non-technical commercial leaders.
• Experience managing upwards — product needs, trade-offs and timelines.

Technical

• 5–8 years of hands-on experience in data science or applied ML roles, with at least 2 years owning data products end to end.
• Strong command of classical ML (gradient boosting, regression, clustering, ranking, time-series forecasting) and the judgment to know when simple beats sophisticated.
• Solid grounding in probabilistic modelling, Bayesian inference and uncertainty quantification, with working experience in a PPL such as PyMC or Stan.
• High proficiency in Python (the standard scientific stack) and strong SQL, including complex multi-table queries and window functions.
• Deep familiarity with model evaluation: cross-validation, calibration, and choosing business-aligned metrics over convenient ones.
• Experience with experiment design and statistical hypothesis testing.
• Comfortable working with cloud data warehouses and experiment-tracking tooling (we use AWS and MLflow; equivalents are fine).
• Strongly preferred
• Experience in PAYG, fintech lending, telco or other emerging-market consumer businesses — you understand irregular incomes, mobile-money payment behaviour and thin, messy data.

Nice to have

• Survival modelling, causal inference or marketing mix modelling (MMM).
• Operations research / optimisation exposure (routing, scheduling) — relevant to our last-mile delivery problems.
• Familiarity with MLOps and model deployment on AWS (SageMaker, Lambda, ECS).