Senior Data Scientist

Modelling Build, validate and maintain predictive models on customer level data: churn and attrition, propensity and next best offer, response and redemption likelihood, credit and collections risk, lifetime value, segment migration, and time to event. Own validation. Out of time testing, calibration, class imbalance handling, leakage detection, stability monitoring. Set the internal standard for what counts as a model fit to ship. Build uplift and holdout designs so campaign impact can be measured rather than claimed. Design and analyse A/B tests for new features and treatments. Randomisation and assignment logic, sample size and power, primary and guardrail metrics, test duration, and a defensible ship or stop call at the end. Produce portfolio level forecasts and anomaly detection over transaction and engagement data. Data Do the work that comes before the model. Profiling, cleaning, entity resolution across tables, feature construction, and clear documentation of what a dataset can and cannot support. Assemble customer level datasets yourself from normalised transactional schemas. You will not be handed a feature store. Engineering Write production Python that runs unattended. Automated cleaning, model selection, scoring, and the evaluation layer that decides whether a model output is good enough to surface. Build pipelines that work on datasets you have not seen in advance, where column names, data quality and structure vary from one dataset to the next. Package models so they can be called by an application rather than run by hand. Scoring interfaces, scheduled retraining, versioning, and monitoring for drift. Integrate LLM APIs into automated workflows, for schema interpretation, plain language explanation of model output, and generated rationale. Communication Turn model output into a recommended action. A ranked list with a probability attached is the input, not the deliverable. The deliverable is a decision with an expected outcome and a stated assumption set. Size opportunities before work starts. Say what is worth modelling and what is not. Present to non-technical stakeholders, including marketing, risk and C-level audiences who will not read a confusion matrix. Assess feasibility on a sample extract and give a realistic estimate of what the data can deliver. Stack Python is the primary language for everything we productionise. Expect to work daily in pandas, scikit-learn, and gradient boosting libraries such as XGBoost or LightGBM, plus whatever the problem needs. SQL for data assembly. R for part of the existing analytics work, including Shiny. LLM APIs as a component inside automated pipelines. Some of what we look for High ownership. We do not have people managers. You have core responsibilities that take about half your day. The rest is on you to find the hardest problems, chase them down, and solve them independently. Curiosity. We are solving unsolved problems and the rules change daily. You need to find joy in learning new and difficult things. Horsepower. Raw capability and throughput. You get a lot done, and the quality holds. Strive to challenge and be challenged. Push on decisions, including ours. Expect the same in return.

Desired Candidate Profile

Requirements Required 5 or more years building predictive models on real, messy commercial data at customer level. Strong production Python. You write code another engineer will run and maintain, not only notebook code. Deep grounding in classification and propensity modelling. Gradient boosting and regularised regression as a baseline, and the judgment to know when something more or less complex is the right call. You can explain why AUC alone is not enough and what you would look at instead. Experimental design. You can size a test, pick the metric before it runs, spot a broken , and call a result honestly when it is flat. SQL at the level of building an analytical dataset from a transactional schema unaided. The judgment to know when a dataset cannot answer the question being asked, and the willingness to say so early. Clear written and spoken English. Strongly preferred R, including RStudio and Shiny. Experience in banking, fintech, lending, telecom or retail loyalty. Familiarity with terms like roll rates, collections treatment, RFM, breakage or redemption liability means less ramp-up time. Experience building automated or configurable pipelines that run on unfamiliar data. Experience integrating LLM APIs into a production workflow.