Wake Up People, Data Strategy is Core for AI Value!

You must be living under a rock if you haven’t noticed the excitement and buzz of AI all around us in business. The hype around AI has captured the thoughts and imagination of boardrooms and executives everywhere, where next-gen strategies are being created and solidified. The promise of generative AI’s transformative capabilities and the surprise of getting an automated response from ChatGPT have CxOs imagining a future where AI agents are seamlessly embedded into business processes. Voilà! With this magic touch, they envision attracting more customers, reducing operational costs, and becoming superstars in the business world. In a recent Gartner survey, most CEOs believe that AI will impact their business – positive or negative. And this is where the real challenge arises, I would argue most companies are not ready to get widespread benefits from this so-called AI revolution, happening in front of our eyes; and they will be on the “negative impact” side of AI-driven impact on businesses. And the main reason is that most companies still do not have a sound data strategy on which they can build a successful AI capability. To generate real value from AI, it is essential to have a sound and comprehensive data strategy. This strategy should include the ability to map business goals to data needs, establish a flexible data management architecture, provide practical data governance and accessibility, and ensure ongoing investment in the right level of data expertise.

Architecture to Democratize Data Access:

AI needs data—this is a fact. Yet, in most businesses, data is still trapped in silos. The era of data warehousing has come and gone, and many companies never fully embraced it to create a comprehensive repository of their business transactions. With approximately 80% of a business’s intellectual property and information locked within transactional datasets scattered across silos, a robust approach to data architecture is critical. Here, concepts like data warehouses, data lakes, and data lakehouses play pivotal roles. Each has value in a data strategy, though debates around the “best” data repository persist. What truly matters is a well-designed architecture that ensures data availability for AI. We don’t want data scientists endlessly hunting for data or multiple individuals duplicating data efforts. Instead, a centralized, accessible data repository should empower data scientists and democratize data access across the organization. Make data available for innovation without imposing excessive protectionism—trust your employees to use data responsibly. Democratizing data access not only accelerates AI initiatives but also fosters a culture of innovation and agility within the company.

Data Quality is a function of its Usage:

AI models require data for training, and the quality of data has a direct influence on the quality of AI outputs. This doesn’t mean we need a zero-quality-issue strategy; rather, data quality requirements should align with specific AI applications. Some insights can indeed be derived from lower-quality data, while certain AI applications—like autonomous driving or medical diagnostics—require exceptionally high data quality. A sound data strategy should therefore establish protocols, best practices, and frameworks to ensure data quality is appropriate for its intended use. A well-defined data strategy needs to have a tiered approach to data quality based on use cases. Let’sconsider the customer sentiment model of your product based on insights drawn from social media posts. Your model would be quite tolerant of inconsistency and data quality issues when measuring customer sentiment; contrasting it with the medical diagnostic use case identified above, the model would require very high-quality data to make goodrecommendations.

Data Governance expands to AI Governance:

With established practices in data governance that cover master data management, metadata, data audits, and data veracity, we need to broaden these practices to include AI governance. AI systems introduce new complexities and responsibilities that require oversight beyond the traditional data governance practices. AI governance addresses these challenges by focusing on the maintenance of AI artifacts, ethical considerations, transparency, and explainability of AI, as well as ensuring that model drift is identified and controlled.Moreover, AI governance provides the structural foundation needed to comply with regulatory standards and internal company policies. A comprehensive AI governance framework would integrate with existing data governance practices, covering AI-specific artifacts like model code, training datasets, and versioning. This approach also requires the incorporation of ethical guidelines and transparency protocols to clarify AI decision-making processes. For instance, by tracking model decisions and ensuring these are interpretable, financial institutions can justify a decision to approve or disapprove automated credit applications.

Organizational strategy to support data and AI:

As AI becoming pervasive and the center of attention for most companies, it is critical for companies to evaluate their organizational structure to support innovation and business outcomes driven by AI that is built on the shoulder of data and requires infrastructure and technology to run workloads. Numerous specialized teams, such as IT technologists, data engineers, data scientists, machine learning (ML) engineers, ML production engineers, and operational support teams, need to operate efficiently. To facilitate this, I recommend a"three-legged stool" structure, where the organization establishes three distinct pillars: a technology and infrastructure team, a data team, and an AI/ML team, each led by separate C-level leaders.

• Technology and Infrastructure Team – Led by a Chief Technology Officer (CTO) or CIO, this team would focus on building and maintaining the robust technology infrastructure needed to support scalable AI initiatives. This includes setting up cloud environments, dev-ops, and ml-ops capabilities to ensure seamless integration between data and AI/ML as well as smooth migration from model development to testing to production and support.
• Data Team – Headed by a Chief Data Officer (CDO), this team would be responsible for managing data governance, data quality, data strategy, and data engineering. Their goal would be to make sure that data is available and democratized for the organization with proper governance in place and quality that meets consumption needs.
• AI/ML Team – Reporting to a Chief AI Officer (CAIO) or Chief Analytics Officer (CAO), this team focuses on the development, deployment, and maintenance of AI models. This group includes data scientists, ML engineers, and ML production engineers who are tasked with building and optimizing AI models, monitoring model performance, and addressing model drift in production. Operational support teams within this function ensure that AI systems run smoothly, with protocols for updates and troubleshooting.

This three-pillared structure enables organizations to better allocate resources, streamline workflows, and assign accountability across the data and AI lifecycle. With each team led by a dedicated C-level executive, strategic decisions can be made with clarity and focus, fostering a culture where data and AI align with business objectives and operational support. This structure also allows for agility and scalability, ensuring that AI efforts are sustainable and that organizations are positioned to leverage data-driven insights effectively.

In the end, I would close my thoughts with a call for action. Anyone who is excited about the business opportunity of AI does not ignore the boring and not-so-sexy side represented by the data architecture and data strategy. A solid data strategy will position organizations to drive true AI-driven transformation. So, let’s all wake up and invest in the data strategy and watch our AI initiatives achieve results beyond expectations.