Effective AI Requires Effective Data Governance

How to Make Data AI-Ready

AI applications require good data to be effective. Consequently, universities and colleges are fine-tuning their data governance strategies and technologies to support their AI initiatives.

Data governance starts with developing policies driven by four principles: data quality, security and privacy, regulatory responsibilities and ethics, says Forrester Principal Analyst Michele Goetz.

Universities must then use processes and technology to implement those policies and create the appropriate AI guardrails to manage risk and produce good, accurate and relevant results, she says.

“It’s creating a consistent data governance plan and stewardship model and investing in technology to help you do this,” Goetz says.

Higher education can deploy cloud-based data governance tools from Amazon Web Services, Google Cloud or Microsoft Azure. They can also implement more comprehensive governance software from vendors that are increasingly adding features for AI use cases, she says.

These tools enable institutions to curate data and create data catalogs so users can search for available information. They also allow universities to document and enforce policies, manage user access and track the lineage or flow of data to ensure security, privacy and compliance.

Today, many universities have cobbled together homegrown solutions, but some campuses have begun using generative AI to govern their data, Goetz says.

SURVEY SAYS: The pace of AI evolution demands a sense of urgency.

UC San Diego Adapts Data Governance for AI    

UC San Diego developed a data governance framework about eight years ago when IT leaders built a data analytics platform using Tableau and IBM Cognos reporting tools and an SAP HANA data warehouse on AWS, which allows staff to run reports and get insights on institutional data.

A data analytics governance committee (including an interdisciplinary team of data stewards who curate the data) collaborates with the campus privacy office and IT security team to create and fine-tune data governance policies, says Daniel Suchy, senior director of Academic Technology Services at UC San Diego.

TritonGPT is hosted securely on-premises at the university’s San Diego Supercomputer Center, giving IT department complete control over the data used. When the university rolled out the AI assistants in 2024, the IT team leveraged a mix of existing and new technology to govern data.

For example, the AI assistant that will allow employees to use natural language to query the analytics platform is in the proof-of-concept stage. It uses the same security and data governance workflow that currently gives employees access to the data warehouse, Pollak says.

Through ServiceNow software, employees can request data access, and through an electronic workflow, their requests are approved by managers and data stewards. Once requests are approved, the IT staff uses Active Directory to provide access, he says.

UC San Diego is migrating away from a legacy tool to catalog data. It’s a metadata repository that allows employees to find the data they need to build reports. The IT department has placed the metadata in an in-house AI assistant. So, instead of manually searching for data, users can simply ask the AI assistant for the data they need, Pollak says.

A Fund Manager Coach AI assistant, in its first iteration, is trained on knowledge base articles and provides fund managers advice on managing grants. To increase the tool’s functionality, the IT staff has begun incorporating financial data from the data warehouse.

As part of data governance, Pollak’s team is collaborating with the finance data steward to identify questions that fund managers might ask. Based on the data steward’s feedback, developers will beef up the metadata or make other tweaks to ensure questions are mapped to the correct data so they are answered correctly.

IT leaders will test the AI assistant this year and plan to broaden it to all the data in the data warehouse so they can create AI assistants that can query student, employee, research and facility data.

“We feel we can replace a good portion of the reporting done on Cognos and Tableau,” Pollak says. “People can ask the AI assistant a question instead of creating a report.”

University Readies New Cloud-Based Data Governance Software 

Within the next year, the University of Tennessee Health Science Center plans to deploy Microsoft Purview, a data governance management tool in Azure. But it currently uses a mix of technologies as a stopgap solution, says Todd Barber, UTHSC’s director of enterprise applications and data services.

Memphis-based UTHSC comprises six colleges, including dentistry, medicine and nursing. It uses homegrown software to catalog data, SharePoint to share governance documents and TeamDynamix ticketing software to handle data access requests.

The campus’s data analytics platform — which includes Power BI for report building and Microsoft Fabric, a cloud-based data lake for centralized data sharing — allows IT administrators to manage aspects of users’ data access, he says.

Purview will provide UTHSC a more robust tool for managing all aspects of data governance, including data cataloging, access control and data loss prevention, Barber says.

“Purview will allow people to find the data available to them, while the governance process works behind the scenes to make sure that the data we allow them to see has been vetted and is good, quality data,” he says.

The IT department provides staff and faculty access to Microsoft Copilot, which they can use within Power BI and Microsoft 365 tools. If employees get wrong answers on Copilot, Barber and his team investigate. Sometimes, it happens because the AI is hallucinating, but they make sure it’s not a data quality issue.

“We look back through data quality and integrity to make sure we’re providing the right data, so the AI model is making the right decisions,” he says.

Notre Dame Reviews Data Governance for AI

In Indiana, the University of Notre Dame is re-evaluating its data classification structure in the context of its AI initiatives to ensure the proper handling and sharing of data, says Jane Livingston, Notre Dame vice President for IT and CIO.

Notre Dame, which has a mature data governance program going back to 2013, currently classifies data as either public, internal, sensitive or highly sensitive.

“To provide guidance to our faculty, staff and students when they are using AI, we wanted to dig a little bit deeper and provide more clarity and examples of what kinds of data are in which category, what kinds of data are safe to use with generative AI chatbots, and which data you should be careful using or not use at all,” Livingston says.

Notre Dame has begun implementing generative AI tools and has created data usage guidelines for each. Through its software license, Google has extended data privacy protections to its education users. As a result, the university allows campus users to utilize Google Gemini and NotebookLM.

“We’re talking to our community and doing a lot of listening about what gaps there are, so that we can come up with helpful guidance,” she says.