Accelerating Responsible AI components for a large telecom company
Business problem and objectives
Our client, a large telecom company, needed to implement a Responsible AI framework for two specific use cases: Authorization to Operate (ATO) and Fraud Detection.
The primary objective was to develop a comprehensive framework that outlined Responsible AI best practices and standards from ideation to decommissioning. Additionally, the client wanted to implement a Governance Framework that assessed scalability and integrated platform and custom changes for both legacy and new solutions.
Key challenges
The client faced several challenges in implementing Responsible AI, including ensuring adherence to correct processes, aligning diverse teams and stakeholders, and establishing governance-compliant procedures. These tasks required attention to detail, thorough documentation, and leveraging platform capabilities for ethical deployment and management in production environments.
The complex nature of these challenges emphasized the need for a strategic approach to address the inherent complexities of Responsible AI implementation.
Solution
To tackle the challenges, we conducted a thorough analysis of the existing platform and identified areas related to governance, metadata management, model registry, and logging. Based on this assessment, we designed a recommended architecture that was not tied to any specific platform, thus reducing the risk of future challenges associated with platform changes or upgrades.
We developed Standard Operating Procedures (SOPs) to cover various aspects, including monitoring, scalability, ML development, and reusability. Additionally, we operationalized and automated the governance framework in collaboration with our client’s operations team, ensuring seamless integration with existing processes and workflows.
Impact created
Introducing a Responsible AI framework, checklists/scorecards, and SOPs allowed the client delivery team to swiftly enhance their Responsible AI practices.
Streamlining the onboarding process for new projects resulted in a 45% reduction in assessment effort by establishing entry and exit criteria.
Additionally, setting up a Responsible AI Center of Excellence (CoE) led to a 40% boost in use case efficiency by adopting industrialization techniques at scale.