Establishing an AI Center of Excellence (CoE): Strategy, Roadmap & KPIs

Sharing some observations from a recent engagement establishing an AI Center-of-Excellence (CoE) for a large Fortune 500 Enterprise.

The goal was to setup an AI CoE that would drive AI thought leadership and an AI first mindset across the enterprise.

Building Blocks

Given this, we started with an AI CoE mandate to set the AI strategy and roadmap for the enterprise, maturing the CoE services and offerings across its core building blocks:

1. Innovation: Establish strong AI foundation with a culture of AI innovation and thought leadership.
2. Talent: Acquire and retain top AI talent. Foster collaboration with key business stakeholders and global teams on AI use-cases and applications.
3. Governance: Develop AI strategy aligned with organizational goals, integrating Responsible AI principles into the AI lifecycle to manage risk.
4. Data as an Asset: Improve enterprise data quality and focus on unstructured data to maximize the potential of AI.
5. Enablement: Drive AI-first mindset by AI literacy, trainings, knowledge-sharing and change management.

AI CoE?Maturity

The journey starts with an assessment / understanding of the organization’s AI maturity in terms of:

• Extent to which AI is currently embedded in products, services, and operations
• Integration of Responsible AI principles in the AI lifecycle
• Data readiness and availability for developing robust AI models
• Knowledge of AI among employees & leadership
• Existing skillsets and headcount in AI/ML

Given this understanding, the AI CoE works towards improving the AI maturity of the organisation along all axes corresponding to the building blocks outlined earlier, e.g.,

Ad-hoc usage of AI in Products & Services →? AI is deployed at scale, but in selected Business Units / Products → AI is fully integrated in Org. DNA - driving strategic differentiation and disruption.

Data as an?Asset

Most enterprises today recognise Structured data (available in Relational?—?SQL Databases and Datawarehouses) as a strategic asset, having made significant investments in Data Governance & Data Quality tooling over the last few years.

With Generative AI (Gen AI), the main differentiator is that unstructured data has become equally important, with most use-cases focused on document (text data) processing.

(Pre-trained) Foundational LLMs are generic in nature. To realize the full potential of LLMs for Enterprises, they need to be contextualized with enterprise knowledge captured in terms of documents, wikis, etc.

Given this, the focus is on unstructured data management for Gen AI, and it is important to implement the same Data Governance (Data Quality, Lineage, Observability, Privacy, etc.) frameworks for unstructured data (documents, logs, invoices, service tickets, etc.) now, e.g., establishing metadata tagging standards for documents.

AI CoE Roadmap &?KPIs

Overall, the AI CoE serves as a strategic enabler for the organization, driving value creation, innovation, and sustainable growth through responsible adoption and effective utilisation of AI technologies.

A sample roadmap to achieve the same is illustrated in the figure below, focusing on 3 three phases:

Understanding (Assessment) & Design → Pilot & Deploy → Scale & Industrialise

KPIs are the key here, and need to be achieved by continuous monitoring and adaptation based on quantifiable KPIs, such as,

• number of AI PoCs, models in Production
• time to market from identification of ideas to AI products
• improvement in data quality measures
• % of employees who have received training on Responsible AI
• % of AI projects compliant with Responsible AI principles and guidelines
• improvement in the average explainability scores of deployed AI/ML models and algorithms
• number of partnerships established with external organizations, academia, or civil communities to promote Responsible AI practices and address societal and employee concerns
• increase in revenue attributed to AI-driven innovations
• cost reduction achieved through automation and optimization enabled by AI technologies
• assess (customer) satisfaction levels of business sponsors based on surveys, ratings, and qualitative feedback