AI in Action Beyond Experimentation to Transform Industry 2025

Key components of a digital core   FIGURE 4 Top/bottom margin area 8mmAI applications and digital platforms 1 Data and AI backbone 2 Physical and digital infrastructure 3 Source: Adapted from Accenture and World Economic Forum stakeholder community input. To unlock value, the digital core should lay the groundwork for sustained growth and transformation. GenAI requires enhanced capabilities from the digital core, making a modern, adaptable infrastructure non-negotiable. Key insights from community discussions emphasize the need for centralized, classified-for-purpose and secure systems. Seamless integration across platforms and automated data management is critical to enhancing efficiency and agility. An open architecture is vital, offering the flexibility and interoperability needed to address diverse use cases. However, different architectures are evolving at varying speeds, requiring a thoughtful approach to integration and scaling. 1. AI applications and digital platforms This layer enables seamless, personalized and intelligent user experiences through applications and interfaces like virtual assistants and personalized recommendations. These systems use real-time data from diverse sources, such as user interactions, social media and internet of things (IoT) devices, to understand individual preferences and deliver context-aware interactions. Complementing this, analytics and adaptive learning systems continuously refine these interactions, becoming smarter over time by integrating feedback through real-time analytics.49,50,512. Data and AI backbone The data backbone layer ensures the flow and usability of high-quality structured and unstructured data, which is crucial for training and deploying effective AI models. Data cleaning enhances model accuracy by reducing noise and biases, using methods like handling missing values, removing duplicates and standardizing formats. When data is not available, synthetic data complements real- world data by addressing scarcity and sensitivity, using techniques like generative adversarial networks (GANs) to improve model diversity and robustness while maintaining privacy. High-quality generation is essential, as poor synthetic data can harm model performance.52 This layer also includes tools for data management, such as synthetic data generation and advanced databases (like vector databases) that can store unstructured data like images, voice or natural language via mathematical representations of data (vectors) and knowledge graphs to capture context.53 Organizations should employ a robust strategy for their data and AI backbone, including the deployment of modern data stacks, cloud migration and effective staging systems. Companies should commit to ensuring metadata accuracy, consistent labelling and continuous monitoring. These actions will ensure the trustworthiness of data, which is core to trusted front-end systems.54 When data is not available, synthetic data complements real-world data by addressing scarcity and sensitivity. AI Governance Alliance: Transformation of Industries in the Age of AI 19