Data Stewards determine AI Success

Artificial Intelligence rests on three pillars: software, hardware and data.

These first two receive a lot of hype.  OpenAI, maker of ChatGPT is the world’s most valuable private company. Nvidia, maker of AI chips is the world’s most valuable public company. While data is coined the ‘lifeblood’ of AI, as I have argued in previous posts, it is not managed with the same discipline of these more traditional assets.

While it is relatively quick and easy to acquire the necessary hardware and software, AI projects inevitably slowdown once executives start asking questions like: 

• How was this output produced?
• Is the same decision made in the same way across the organisation?
• Is this compliant with privacy, security and regulatory obligations?

At their core, these are questions of trust, consistency and compliance. They are not unique to AI, and they are not necessarily tied to the hardware or software.

These are data management questions that require data management solutions.

This is where data stewardship comes in. Data stewards sit at the intersection between business and technology and are responsible for the data and processes that ensure the effective control and use of data assets. As a Certified Data Management Professional (CDMP) it is my view that data stewards have the tools to enable the effective deployment of AI solutions.

Four Activities That Enable AI

Creating and managing metadata

Metadata is the context that makes data comprehensible. Business glossaries, definitions, classifications, data models, lineage, and ownership allow both humans and AI systems to understand what data represents. Metadata is a pre-requisite. Without it, data is representation without meaning.

Documenting rules and standards

Data stewards document and define business rules, data standards and data quality rules. This is critical in ensuring consistency in the processes across the enterprise that collect and consume data. Without stewardship across the organisation, you are condemned to inconsistency.

Managing data quality

Poor data quality eliminates trust in a data asset. AI amplifies these errors with confidence. Stewards must benchmark the quality of their assets across the nine dimensions of data quality and address issues at the source.  Once trust is lost it will not be regained.

Operationalising Data Governance

Stewards must define and enforce data policies (the what) and standards, procedures (the how). By demonstrating privacy and security compliance, stewards are able to ensure that AI can be trusted internally, defensible externally and scalable across the organisation.

Data Management is the enabler of Value

In the AI age, trustworthy data assets are a competitive advantage. Data centric organisations align their business and data strategies, ensuring that data stewardship is not limited to a job title but is an enterprise-wide practice. For most organisations, successful AI initiatives will not be defined by their choice of hardware or software, they will be defined by the maturity of their data management practices.

These challenges are not unique to AI. They are problems stewards of data already know how to solve.