Cases
Work in high-consequence decision environments
Selected engagements where model behaviour, governance, and evidence affected real decisions — with real consequences for customers, money, operations, and reputation.
Banking
Interest Calculation and Regulatory Remediation
A credit card interest calculation model was producing errors across customer balances, creating financial and regulatory risk at scale. The issue drew regulatory attention and required changes to the underlying approach. Assurance work was carried out around calculation accuracy, control design, and output validation, including the development of assurance tooling and close work with compliance and legal teams. This helped strengthen confidence in the calculations and support remediation under scrutiny.
Financial Crime
Fraud Model Review in a Banking Environment
Fraud detection models were being used to identify suspicious behaviour across large volumes of transactions. The challenge was not only model performance, but whether decisions could be understood, reviewed, and defended when false positives or missed risks occurred. Assurance and review work strengthened traceability around model outputs and highlighted where decision control needed to be clearer. This reduced uncertainty in a high-risk operational environment.
Financial Crime
Fraud Detection Across Banking and Payments
Fraud detection operated across both banking and payment environments, where decisions had to be made quickly and at scale. Small weaknesses in logic, data quality, or review processes could lead to customer impact, financial loss, or challenge from stakeholders. Work focused on improving confidence in decision outcomes and making supporting controls easier to evidence. This helped create a more reliable and defensible decision process.
Government
COVID-19 Data and Decision Support
During the COVID-19 period, data was used to support urgent public-sector decisions under intense pressure. The environment changed rapidly, while decisions still needed to be based on information that could be trusted and explained. Work focused on improving the reliability, structure, and usability of data feeding decision processes. This helped reduce uncertainty in a time-critical setting.
Public Health
Tooling and Data Control in a High-Scrutiny Environment
In a public health setting, tools were established to support operational and analytical decision-making where poor data handling could have immediate consequences. The challenge was to make systems usable at pace while preserving control, consistency, and accountability. Work focused on helping set up tooling and improving confidence in how outputs were generated and used. This supported better visibility and stronger decision discipline under scrutiny.
Banking
Driver-Based Financial Modelling and Management Decisions
Driver-based financial models were used to support management understanding of performance, planning, and decision-making. The risk was not only model error, but misplaced confidence in assumptions that could materially affect business choices. Work in this area supported clearer validation, stronger challenge of outputs, and better alignment between modelling logic and decision use. This improved confidence in how model-driven insights were interpreted.
Transactions
Data, Assurance and Investor Scrutiny
In a transaction environment, the business was being prepared for sale to overseas investors, which increased the importance of data quality, consistency, and explainability. Information used to support value and performance needed to withstand close challenge from external parties. Work focused on strengthening confidence in the underlying data and in the way outputs were presented and understood. This reduced risk in a commercially sensitive process.
Education
AI in Teaching, Feedback and Marking
AI tools were used to help prepare lessons, generate student feedback, and support marking activity. While this improved speed, outputs were not always consistent and could not simply be accepted without review. Structured controls were introduced around how AI-generated material was used before reaching students. This improved consistency and reduced the risk of poor or unfair educational outcomes.
Accounting
AI in Bookkeeping and Submission Preparation
AI was used to automate bookkeeping tasks and support the preparation of financial and tax submissions. Small classification errors or unsupported assumptions could flow through into final outputs and create compliance problems. Review and validation controls were added before submission stages. This improved reliability and reduced the risk of avoidable reporting errors.
Education Technology
AI-Driven Learning Recommendations
AI was used to personalise learning pathways and generate recommendations for students. The main challenge was ensuring that outputs were suitable, explainable, and not misleading when used in real educational settings. Control was introduced around how recommendations were reviewed and applied. This helped ensure learning decisions remained appropriate, consistent, and defensible.
Retail Finance
Buy Now Pay Later Calculation Risk
Interest on Buy Now Pay Later balances had been calculated incorrectly over time, creating widespread customer impact and regulatory risk. The lack of clear validation and oversight allowed errors to persist unnoticed. A full review of calculation logic was undertaken, and assurance controls were introduced to test and monitor outputs. This enabled remediation and ensured calculations were accurate, controlled, and suitable for scrutiny.
Discuss a related situation
If your business uses AI in similar ways, the risk is real
A short discussion is usually enough to determine whether the issue is one of risk, governance, evidence, or some combination of the three.
Not sure of your exposure? Try the AI Risk Check