Artificial Intelligence in Banking Training Course

Participants can frame a banking AI problem with measurable objectives, boundaries and ownership.

• Problem statements linked to customer, risk or operational decision points
• SMART success measures for accuracy, savings, losses and service levels
• Decision boundaries separating recommendations, approvals and customer communications
• Human-in-the-loop design for material decisions and exception handling
• Prioritisation matrix using value, feasibility, risk and data readiness
• Use-case owners assigned across business, technology, risk and compliance

Participants can specify data requirements, quality controls and lineage for an AI use case.

• Source systems mapped across core banking, CRM, payments and case management
• Data lineage traced from transaction capture to model feature
• Completeness, accuracy, timeliness and consistency checks defined for datasets
• Missing values, duplicates and outliers treated before model development
• Feature definitions documented for balances, behaviours and customer interactions
• Personally identifiable information classified under data protection requirements

Participants can explain common machine-learning methods and select an appropriate approach for a banking decision.

• Supervised learning applied to default, fraud and attrition prediction
• Unsupervised learning used for clustering and unusual transaction detection
• Classification, regression and ranking selected by target decision type
• Training, validation and test sets separated to prevent leakage
• Feature engineering converts banking events into measurable predictive signals
• Baseline rules compared against machine-learning model performance

Participants can interpret model performance and set thresholds aligned to banking risk appetite.

• Confusion matrices quantify true positives, false positives and false negatives
• Precision and recall balanced for fraud alerts and investigation capacity
• ROC curves and AUC compared across candidate classification models
• Calibration checks align predicted probabilities with observed customer outcomes
• Decision thresholds set against losses, workload and customer friction
• Champion-challenger testing supports controlled model selection and replacement

Participants can assess AI-supported credit decisioning for performance, fairness and governance requirements.

• Credit lifecycle mapped from application intake through collections treatment
• Alternative data assessed for predictive value, consent and conduct risk
• Scorecard variables reviewed for stability, explainability and prohibited proxies
• Probability of default models linked to underwriting decision policies
• Adverse action reasons translated into understandable customer communications
• Override governance records human decisions, rationale and outcome patterns

Participants can design controlled AI approaches for fraud detection and financial crime monitoring.

• Fraud typologies translated into behavioural and transactional detection features
• Graph analytics identifies connected accounts, devices and payment beneficiaries
• Alert triage models rank cases by risk and investigation priority
• False-positive reduction measured without weakening suspicious activity coverage
• Scenario tuning compares rules-based monitoring with machine-learning detection
• Investigator feedback loops improve labels, thresholds and model monitoring

Participants can define safe AI applications for customer interaction, servicing and collections processes.

• Intent classification routes customer requests to appropriate service journeys
• Chatbot escalation rules protect customers during vulnerable or disputed cases
• Speech analytics detects complaints, sentiment and quality assurance indicators
• Next-best-action models prioritise relevant servicing and retention interventions
• Collections treatments segmented by affordability, engagement and vulnerability indicators
• Customer outcome testing checks service equity and conduct impacts

Participants can evaluate generative AI uses and define safeguards for banking content and workflows.

• Large language model capabilities compared with retrieval-augmented generation architectures
• Prompt design specifies role, context, source constraints and output format
• Hallucination controls require grounded sources, citations and human review
• Customer-facing content reviewed for accuracy, fairness and regulatory disclosures
• Knowledge bases curated from approved policies, procedures and product documents
• Sensitive-data controls prevent confidential information entering external model services

Participants can assess AI decisions for explainability, disparate impact and customer treatment risks.

• Explainability requirements differ for underwriting, fraud blocking and customer assistance
• SHAP values identify feature contributions to individual model predictions
• Proxy discrimination tested through correlated variables and protected characteristics
• Fairness metrics compare selection rates, errors and outcomes between groups
• Customer-impact assessment considers denial, delay, friction and remediation paths
• Model explanations documented for users, customers, auditors and regulators

Participants can apply model-risk management controls throughout an AI model lifecycle.

• Model inventory records purpose, owner, materiality and deployment status
• SR 11-7 principles applied to development, validation and governance
• Conceptual soundness review tests assumptions, methodology and intended use
• Independent validation challenges data, performance, outcomes and limitations
• Model limitations recorded with compensating controls and user restrictions
• Periodic review schedules reflect materiality, drift and regulatory change

Participants can establish accountable governance for banking AI initiatives and deployed systems.

• Three-lines model assigns business, risk, compliance and audit responsibilities
• AI policy defines permitted uses, prohibited uses and approval gates
• NIST AI RMF functions mapped to banking governance activities
• ISO/IEC 42001 management-system controls aligned with internal policies
• Risk appetite statements define tolerances for customer and model harm
• Governance committees review material models, incidents and remediation progress

Participants can assess privacy, security and vendor risks associated with banking AI solutions.

• Data minimisation limits personal information collected for stated purposes
• Consent, lawful basis and retention requirements documented for model data
• Threat modelling identifies prompt injection, data poisoning and model extraction
• Access controls separate development, testing, production and privileged administration
• Vendor due diligence reviews training data, hosting, security and audit rights
• Contract clauses address incident notification, data deletion and model changes

Participants can create monitoring and incident processes for deployed banking AI systems.

• Performance dashboards track accuracy, calibration, alerts and customer outcomes
• Data drift detected through distribution changes in production input features
• Concept drift identified when relationships between features and outcomes change
• Threshold breaches trigger investigation, rollback or enhanced human review
• Incident playbooks define containment, customer remediation and regulatory escalation
• Post-incident reviews document root causes, actions and control improvements

Participants can organise cross-functional delivery from AI concept through controlled production release.

• Product operating model aligns business sponsors, users and technical delivery
• Agile backlog converts use-case requirements into testable work items
• MLOps pipeline controls code, data, models and deployment approvals
• Acceptance criteria include performance, explainability, security and operational readiness
• User training addresses decision authority, exceptions and escalation routes
• Change management prepares frontline teams for revised processes and controls

Participants can prepare an investment case that quantifies benefits, costs, risks and control requirements.

• Benefits model estimates loss avoidance, productivity, conversion and service improvements
• Cost baseline includes data preparation, technology, validation and ongoing monitoring
• Financial assumptions tested through sensitivity analysis and downside scenarios
• Control costs incorporated for governance, assurance and human review
• Pilot design defines population, duration, comparison group and exit criteria
• Investment decisions documented with assumptions, dependencies and residual risks

Participants can prepare clear evidence of AI controls for internal audit and regulatory engagement.

• Regulatory horizon scanning identifies applicable AI, privacy and conduct obligations
• Model documentation links business purpose, data, methodology and limitations
• Audit trails retain approvals, versions, user actions and decision evidence
• Control testing samples model outputs, overrides and customer outcomes
• Management information reports risk indicators, incidents and remediation status
• Regulatory briefing packs present governance, validation and monitoring evidence

Participants can develop a detailed design for a selected banking AI use case.

• Selected use case refined through stakeholder needs and process mapping
• Target operating process distinguishes automated actions from human decisions
• Data requirements translated into source, quality and access specifications
• Risk assessment identifies conduct, model, privacy and security exposures
• Control design specifies approvals, monitoring, escalation and audit evidence
• Implementation roadmap sequences pilot, validation, release and review milestones

Participants can build, review and present a complete governed banking AI initiative pack for approval.

• Business problem, customer impact and value hypothesis consolidated into proposal
• Use-case scorecard assesses feasibility, materiality, risk and expected benefits
• Data lineage and quality controls documented for critical model inputs
• Model approach, thresholds and human oversight defined for decision workflow
• Governance, validation, monitoring and incident controls assembled into plan
• Stakeholder review challenges assumptions, evidence gaps and residual risks
• Executive presentation communicates investment case, controls and implementation milestones