The Intersection of AI and IG: Getting the (Data) House in Order
Information Governance Pros Can Shape How AI Is Optimized and Governed Across Organizations
-
noviembre 04, 2025
-
As enterprises increasingly adopt artificial intelligence, information governance programs will need to adapt and information governance professionals will have a unique opportunity to influence how AI is deployed, optimized and governed across their organization. Proactive attention to the cross-section between AI projects and information governance fundamentals helps position organizations to transform data from a potential liability into a strategic asset for innovation and growth.
Across many enterprise functions, AI is equal parts potential solution and risk: the technology can be powerful when pointed at sound, quality data, but it will also generate new data that itself requires management and governance. Solving that paradox requires organizations to first determine the purpose for which they want to use AI and what data will be used within an AI system. This is deeper than selecting a use case. It’s about understanding the strategic imperative for the use of AI and its business case, including whether and how a new AI implementation will accelerate activity and amplify the ability to do more with less. Or, conversely, diminish enterprise capacity and capabilities. Defining the data elements needed to achieve the objectives is also critical.
Once the business case is established and confirmed, teams must then define what data will be needed to accomplish the goal and whether that data is available and ready to meet the need. Selecting the right data requires clear definition of numerous factors, including the business need, the model that is being used and how that model has been trained. For example, the team must consider whether the data that’s used in the model is appropriate and consistent with the effectiveness of the model based on its training set.
All AI projects are data projects, because even after an AI model is trained on data, the AI model will consume (different) data to generate an output. Many AI projects break down at the point when it’s time to deploy a new tool with a specific set of data. Organizations may not know what data they need or if that data is appropriate for consumption within the AI system, or if they do know and appropriate data is identified, that data is often not ready, organized or accessible.
So, how to avoid this? By leaning into information governance basics so that the data estate can be prepared to effectively support advanced technologies and so that processes can evolve to meet the demands of an environment that includes AI. While information and data governance practices will span numerous aspects of AI use across an enterprise, there are five critical factors that must be addressed to ensure clean data before a project begins. These include:
- Lineage. Knowing where the data came from and whether the organization has rights (e.g., privacy, intellectual property and contractual rights) to use it will impact whether it can be included in an AI project and what level of risk may coincide with that use.
- Alignment. A lot of attention is given to model testing and training, however, that isn’t always aligned to the type of data that the AI will be pointed to. There needs to be alignment between the data the organization has, the kind of data the model has been trained on and the needs of the business case.
- Access. New AI tools may create new access control risks when certain data is loaded into the model. Typically, organizations manage permissions for which users can see which types of data. However, some enterprise-wide generative AI tools will work across the entire corpus of data and may inadvertently give improper access to an employee using the tool. This is particularly true about generative AI systems, which might include restricted data as part of an output. Information governance, legal and compliance teams must fully understand how an AI application will impact and interact with existing user permissions, to avoid inadvertent disclosure of sensitive or protected information.
- Preparation. To the extent that an organization is preparing its own models, data preparation is critical, including for predictive AI. Models will improve relative to the type of munging of data that takes place before data is loaded into the model and how the model is trained. Even when using a pre-trained model, there are cases when data needs to be prepared prior to its input.
- Stability. Models will require consistency and monitoring, so that the data being fed into the model continues to be aligned, or that the model can be adapted if the input changes. This requires a mechanism to monitor for data or model degradation over the course of time.
An organization’s data can be among its most valuable assets and existing data can have new value as an input into an AI system. When an organization’s data assets are combined with AI to derive insights and shed light on patterns that were otherwise unknown, the results can be powerful. Adapting existing information governance programs for AI by revisiting retention and disposal, data classification, policies, procedures and more will reduce the inherent risks within an organization’s data. At the same time, organizations can maximize the impact of their technology investments when each AI project is approached as the data-driven initiative that it is, with care and priority given to information governance before new tools are deployed.
Related Insights
Related Information
Published
noviembre 04, 2025
Key Contacts
Managing Director