As machine learning moves into lending, claims handling, fraud screening, network planning and patient triage, model failure has become a business risk in its own right. Errors are often silent: a model keeps producing confident scores long after the population it was trained on has changed, or after an upstream system begins sending malformed values. New threats have appeared alongside the familiar ones, including data poisoning, prompt injection, model theft and privacy attacks on training data, and many organisations still have no register, appetite statement or named owner for any of them.

The Risk Management for Data Science & Machine Learning in Practice Training Course applies the established cycle of identification, assessment, treatment, monitoring and reporting to every stage of the model lifecycle. Days one to five build the foundations: a risk taxonomy, structured identification methods, data risk, model error and its business impact, and risk appetite. Days six to ten go further into quantification and control, with uncertainty estimation, stress testing, adversarial robustness, security threat modelling, third-party and generative AI risk, resilience planning, incident response and board-level reporting.

Each concept is tried on real models. Participants work in Python notebooks on credit, fraud and demand forecasting data, break the models deliberately and measure what happens, then design the controls that would have caught the failure. The course closes with an integrated risk assessment of a complete machine learning system, presented to a simulated risk committee.

Participants completing the course will be able to:

• Classify the risks that data science and machine learning introduce, from data quality and model error to security, privacy, conduct and third-party exposures.
• Apply FMEA, bowtie analysis and pre-mortem workshops to identify failure modes across a model pipeline.
• Quantify the business impact of model error using cost-weighted metrics, calibration checks and stress testing.
• Estimate prediction uncertainty with bootstrapping and conformal prediction, and set decision thresholds accordingly.
• Assess machine learning systems against adversarial, data-poisoning and prompt-injection threats.
• Define model risk appetite, tolerance limits and key risk indicators that trigger action.
• Design treatment plans that combine technical controls, human oversight and fallback procedures.
• Report model risk to executives, risk committees and auditors in a clear, evidence-based form.

The programme brings together the technical and risk specialists who share responsibility for models, such as:

• Model risk managers and model validation analysts
• Data scientists and machine learning engineers who own production models
• Enterprise, operational and credit risk officers
• Chief risk officers and heads of risk in banks, insurers and microfinance institutions
• Internal auditors and IT auditors reviewing analytics and automated decisions
• Information security and cyber risk specialists
• Actuaries and quantitative analysts who rely on predictive models
• Risk and compliance staff in telecoms, health, energy and public agencies

Participants return to work with:

• A machine learning risk taxonomy and register template populated for a real model.
• A completed FMEA worksheet and bowtie diagram for a model pipeline.
• A Python notebook for stress testing, calibration checks and conformal prediction intervals.
• A drift and stability monitoring script built on the population stability index.
• A threat model for one machine learning system, with mitigations mapped to MITRE ATLAS.
• A model risk appetite statement with key risk indicators and escalation thresholds.
• A fallback and human-override procedure for a high-impact model.
• A risk committee report summarising residual risk and recommended actions.

Learning is evidence-led: participants test risks on working models rather than discussing them in the abstract. Methods include:

• Python labs in which participants deliberately break models with bad data, shifted populations and adversarial inputs
• Structured risk workshops using FMEA templates, bowtie diagrams and pre-mortems
• Case reviews of documented model failures across finance, health, recruitment and public services
• Red-team exercises against a classifier and a chatbot
• Simulated risk committee sessions where participants defend their assessments
• An integrated risk assessment built up day by day and presented on the final day

Delegates who attend all ten days and complete the labs, risk workshops and final assessment receive a CPD-accredited Certificate of Completion issued by Vision Reach Global Consultancy.

Alongside the certificate, participants keep the risk register, threat model and committee report they produced, a practical record of model risk work that can support annual appraisals and continuing professional development files.