A structured, practice-oriented program exploring how ML models process individual financial behaviour and generate contextually relevant guidance - built for analysts, developers, and product teams.
Personalised financial recommendations sit at the intersection of behavioural data, model design, and product constraints. This program treats each of those as a distinct skill area. Participants work through real datasets drawn from lending, savings, and investment contexts - not synthetic examples. Each module builds directly on the previous one, so gaps in understanding surface early rather than at the end.
Participants examine how collaborative filtering and content-based methods behave differently when applied to financial products versus general consumer goods. The module covers cold-start problems, sparse interaction matrices, and the regulatory constraints that shape what a model can and cannot recommend.
Raw transaction logs rarely arrive in a model-ready format. This module covers segmentation by spending category, recency-frequency-monetary aggregation, and the construction of user embeddings from sequential behaviour. Participants build feature pipelines using Python and evaluate which signals carry genuine predictive weight.
Choosing between gradient boosting, neural collaborative filtering, and hybrid architectures depends on data volume, latency requirements, and explainability needs. Participants run controlled comparisons, interpret precision-recall trade-offs in a financial recommendation context, and calibrate probability outputs to match real-world acceptance rates.
A recommendation system that works in a notebook does not automatically work in production. This module addresses serving infrastructure, A/B testing design for financial products, and the feedback loop problem - where model outputs influence future training data in ways that can silently degrade performance over time.
The program is structured around deliverables, not just concepts. Each participant leaves with working artefacts they can reference, extend, or present in a professional setting.
Participants assemble and test a complete ML pipeline - from raw data ingestion through feature engineering to ranked output - using a curated financial dataset provided during the program.
Choosing the right metric for a financial recommendation task is not straightforward. Participants practice interpreting offline evaluation results and connecting them to business-relevant outcomes like product uptake and user retention.
Understanding how a recommendation model behaves after release matters as much as building it. Participants design monitoring dashboards and work through scenarios where model drift would affect recommendation quality in a live financial product environment.
"The feedback loop module changed how I think about model maintenance. I had not considered how recommendations themselves become training signal - and how quietly that can go wrong."
Daria Okonkwo - Data Analyst, Financial Services