Insurance broker
Evidence retrieval for an insurance broker
- Headcount
- Approx. 200
- Operations
- UK
- Engagement
- Prototype & evaluation
- The brief
- Help a broking team use evidence from client interviews, with each answer traceable to its source.
- Our work
- Built and evaluated a prototype that extracts evidence from interview transcripts and makes it searchable through an AI agent. Answers link back to source passages, while missing, qualified and contradictory evidence remains visible to the analyst.
Engineering consultancy
Independent review of an AI application platform
- Headcount
- Approx. 20,000
- Operations
- Global
- Engagement
- Architecture & security review
- The brief
- Assess a reusable platform for deploying internal AI-enabled applications before those applications go live.
- Our work
- Reviewing the architecture with the engineering and security teams, covering identity, data access, deployment boundaries and agent connectivity. The review examines which controls the platform enforces and what individual application teams need to address.
International Banking Organisation
AI project triage and agent controls
- Headcount
- Approx. 3,500
- Operations
- Europe, Asia & Africa
- Engagement
- Governance design
- The brief
- Define how an AI centre of excellence assesses new projects and AI features introduced by existing software vendors.
- Our work
- Developed a tiering framework, a vendor-feature triage process and draft security requirements for agent connectivity. The design links the level of review to a system’s data access and autonomy, with a route to reuse approved architecture patterns.
Investment management firm
Selecting controls for AI deployments
- Headcount
- Approx. 1,000
- Operations
- Global
- Engagement
- Governance review & control framework
- The brief
- Connect an organisation’s AI policies and control catalogue to the decisions teams need to make for individual projects.
- Our work
- Reviewed the governance documents and developed a project triage process and control-selection workbook. The workbook maps data sensitivity, autonomy, connected systems and architecture to an applicable control set, with the reason for each selection.