Artificial Intelligence for Banking Risk and Fraud Management Training Course

Course overview

Bank risk and fraud teams face a difficult operational balance: detect suspicious activity early, reduce credit and financial-crime losses, and meet regulatory expectations without overwhelming investigators or declining legitimate customers. Rules engines alone struggle with evolving fraud typologies, mule-account networks, synthetic identities, and changing transaction behaviour. AI can improve prioritisation and detection, but poorly designed models can introduce bias, weak audit trails, model-risk findings, and costly false positives. This course equips banking professionals to make sound decisions about where machine learning adds value and how it must be governed.

Participants examine AI use cases across transaction monitoring, card and payment fraud, anti-money laundering alert triage, credit risk, collections, customer due diligence, and operational-risk surveillance. They learn to frame risk problems, prepare transaction and customer data, engineer behavioural and network features, select supervised and unsupervised models, evaluate precision, recall, ROC-AUC, PR-AUC, and alert-rate trade-offs, and explain model decisions using SHAP. The programme also addresses data leakage, concept drift, fairness testing, human-in-the-loop review, model validation, and documentation aligned to banking model-risk management practices.

Instruction combines bank-specific case studies, guided Python and SQL analysis, model-review workshops, and structured discussion of control design. Participants build a fraud-risk modelling pack using a realistic transaction dataset: a feature specification, model evaluation scorecard, threshold recommendation, explainability output, monitoring plan, and model-governance evidence checklist. This tangible deliverable gives attendees a defensible starting point for a pilot, validation review, or enhancement to an existing fraud and risk analytics programme.

Course objectives

By the end of this course, participants will be able to:

• Frame banking fraud and risk use cases as measurable classification, anomaly-detection, or prioritisation problems
• Prepare customer, account, transaction, device, and alert data using SQL-based quality and leakage checks
• Engineer behavioural, velocity, peer-group, and network-derived features for fraud detection models
• Build and compare supervised and unsupervised fraud models using Python and scikit-learn
• Evaluate models with precision, recall, PR-AUC, ROC-AUC, calibration, and investigator-capacity metrics
• Set risk-score thresholds using expected-loss, false-positive, and operational-review trade-off analysis
• Explain individual model decisions with SHAP and document reason codes for investigator review
• Produce a model monitoring and governance pack covering drift, bias, validation, approval, and escalation controls

Benefits of attending

Target competencies

Who should attend

• Fraud Risk Managers — who need to improve detection performance while controlling investigator workload and customer friction
• Financial Crime and AML Managers — who oversee transaction-monitoring and alert-triage processes
• Credit Risk Managers — who assess AI applications for underwriting, early-warning, collections, and portfolio surveillance
• Banking Data Scientists — who need to develop models that meet fraud, explainability, and governance requirements
• Model Risk and Validation Specialists — who must challenge AI model design, performance, and monitoring evidence
• Risk Analytics and Digital Banking Leaders — who sponsor fraud and risk AI initiatives and need defensible implementation decisions

Requirements and prerequisites

Participants should understand core banking risk or fraud processes and be comfortable interpreting basic descriptive statistics, such as distributions, rates, averages, and correlations. Familiarity with transaction data, customer records, fraud alerts, credit portfolios, AML monitoring, or risk reporting is expected. Participants should also have basic SQL literacy and be able to read simple Python notebooks; prior coding experience does not need to be extensive. No prior machine-learning model-building experience is required, and no advanced mathematics, cloud engineering, or production-system deployment knowledge is assumed. Pre-course materials provide a refresher on Python, classification terminology, and evaluation metrics.

Training methodology

The course uses instructor-led briefings to establish banking risk concepts, followed by guided analysis in Python notebooks and SQL queries. Participants work with a realistic transaction dataset to profile fraud patterns, create features, train and compare models, and test alert thresholds against finite investigation capacity. Case discussions examine failed controls, explainability challenges, and model-risk review findings. Small groups conduct a model-approval workshop, then complete an individual application plan identifying a priority use case, required data, control owners, success measures, and first 90-day actions.

Course outline

Day 1: Banking AI risk foundations and use-case design

• Fraud, AML, credit, and operational-risk AI use-case taxonomy
• Classification, anomaly detection, ranking, and network-analysis problem types
• Bank fraud loss economics and customer-friction trade-offs
• Risk appetite, investigator capacity, and alert-volume constraints
• Data lineage across customer, account, transaction, device, and case-management systems
• Target definitions, label quality, and delayed fraud-confirmation outcomes
• AI governance roles across first line, compliance, model risk, and internal audit

Workshop: Participants map a selected banking fraud or risk use case into a problem statement, target variable, decision owner, loss measure, and control requirements.

Day 2: Data preparation and fraud feature engineering

• SQL profiling of transaction volumes, nulls, duplicates, and outlier values
• Temporal train-validation-test splits for transaction and customer data
• Data leakage detection in post-event fields and investigation outcomes
• Velocity features for payments, logins, beneficiary changes, and cash activity
• Behavioural baselines and peer-group deviation features
• Customer-account-device relationship features and entity resolution concepts
• Feature documentation, provenance, and reproducibility controls

Workshop: Using SQL and Python, participants profile a transaction dataset and produce a documented feature specification for a fraud-detection model.

Day 3: Fraud modelling and performance evaluation

• Logistic regression as an interpretable banking risk baseline
• Random forest and gradient-boosting model comparison in scikit-learn
• Isolation Forest for novel-pattern and anomaly detection
• Class imbalance handling through weighting, sampling, and stratified evaluation
• Precision, recall, F1, ROC-AUC, PR-AUC, and confusion-matrix interpretation
• Probability calibration and risk-score reliability assessment
• Champion-challenger testing and model-selection rationale

Workshop: Participants train baseline and challenger models, then produce a scorecard comparing detection performance, calibration, and operational implications.

Day 4: Explainability, fairness, and model-risk controls

• SHAP global feature importance and local transaction-level explanations
• Reason-code design for investigators and customer-facing adverse-action contexts
• Fairness testing across protected and vulnerable customer segments
• Threshold bias, proxy variables, and disparate-impact investigation
• Concept drift, data drift, and fraud-typology change detection
• Independent validation testing and model limitations documentation
• Human-in-the-loop review, override controls, and escalation governance

Workshop: Participants conduct a model-review workshop, creating SHAP-based reason codes, a fairness test summary, and a list of validation challenges.

Day 5: Deployment planning, monitoring, and banking implementation

• Risk-score threshold optimisation using expected loss and review capacity
• Alert queues, case-management integration, and investigator feedback loops
• Production monitoring for performance, drift, stability, and data quality
• Model inventory, approval evidence, change control, and audit trails
• Incident response for model degradation and emerging fraud typologies
• Pilot design, success metrics, and phased rollout governance
• Executive reporting for fraud reduction, false positives, and control effectiveness

Workshop: Participants assemble and present a fraud-risk modelling pack containing a threshold recommendation, monitoring dashboard specification, governance checklist, and 90-day implementation plan.

Tools & standards covered

Python, SQL, scikit-learn, SHAP

A typical training day

Live online deliveries follow the same structure in the East Africa Time zone, with shorter screen blocks and longer breaks.

What the fee includes

How you can take this course

Certification

Participants who complete the full five days receive the Skillset Development Certificate of Completion, stating the course title, course code, dates and delivery format — suitable for professional-development records and employer reimbursement.

Frequently asked questions