0 of 34 items completed 0%
Business Case Validation Business problem is clearly defined — not just 'we want to use AI' Quantified value case prepared — revenue uplift, cost reduction, or risk reduction Business sponsor identified and committed Success metrics defined — measurable KPIs with baseline and target values Time to value estimated — when will the first measurable benefit be realised? Build vs. buy vs. partner decision made and rationale documented Opportunity cost assessed — what else could this investment fund?
Data Readiness Required data sources identified and their owners confirmed Data availability confirmed — data exists and can be accessed Data quality assessed — sufficient volume, accuracy, and completeness Historical data available for model training — minimum 12 months recommended Data labelling requirements assessed — is labelled data available or must it be created? Data privacy and consent issues assessed — is the data permissioned for AI use? Data lineage documented — can the model's inputs be traced and audited?
Technical Feasibility Technical approach validated by data science or AI team Proof of concept or prototype completed — model performance assessed Model accuracy meets the minimum threshold required for the business case Inference latency meets the operational requirements (real-time vs. batch) Infrastructure requirements assessed — compute, storage, and integration Integration with existing systems is technically feasible Monitoring and retraining approach defined
Risk & Ethics AI risk assessment completed — bias, explainability, and drift risks documented Protected characteristics assessed — model does not discriminate unlawfully Explainability requirements assessed — can decisions be explained to customers and regulators? Human oversight mechanism defined — who reviews and overrides model decisions? Regulatory requirements for AI in this use case assessed Ethical review completed — use case is consistent with the institution's values Reputational risk assessed — how would this use case be perceived publicly?
Governance & Approval Use case registered in the AI use case inventory Risk rating assigned — low, medium, or high based on impact and complexity Approval obtained from the appropriate governance body (AI committee, Risk committee, etc.) Model risk management requirements confirmed Regulatory notification or approval obtained if required Exit criteria defined — under what conditions would the model be switched off?