AI Governance Process for Research and Insight Teams
This project focused on designing a governance framework for AI usage within research workflows. The challenge was to enable experimentation and innovation while ensuring compliance, data security, and output reliability.
Context and brief
AI tools were being adopted inconsistently across teams, creating risks around data privacy, quality of outputs, and duplication of effort. There was no standardised approach to evaluating or using these tools.
Objective
Create a structured framework that allowed teams to experiment with AI safely while maintaining quality, compliance, and transparency, as well as adhering to company AI policy guidelines.
Role and responsibilities
I designed the governance framework and defined how AI should be evaluated, tested, and used across research workflows.
This included creating a structured tool evaluation scorecard, defining different usage tiers based on risk, and establishing clear rules for experimentation, validation, and documentation.
Approach
The framework distinguished between low-risk and high-stakes use cases.
For low-risk scenarios, AI could be used for speed and efficiency with minimal oversight.
For higher-stakes work, I defined requirements for parallel human benchmarking, statistical validation, and peer review.
I also introduced a structured evaluation process for new tools, covering accuracy, reliability, compliance, and integration, and created a central registry to document all experiments and outcomes.
Supplier AI capabilities were also assessed to ensure they met data handling and compliance requirements before use.
Output
A comprehensive AI governance framework, tool evaluation scorecards, experimentation guidelines, and centralized documentation processes.
Impact
This standardized AI usage across the team, reduced risk, and accelerated adoption of tools that met quality thresholds, while maintaining alignment with broader organizational governance.
It also reduced duplication of work related to testing new AI tools, accounting for approximately 5 man days per month.