It's Time to Start Governing Data and AI Together

Home / All Blogs

It's Time to Start Governing Data and AI Together

View Other Blogs by Category

Everyone is talking about data governance for AI. That’s a good thing. Over the last year awareness has grown that data governance is necessary for AI. But that isn’t enough. And it might even be dangerous. Data Governance FOR AI is very different from governing data AND AI together. Very different. And those differences matter.

Here’s a reason to prioritize governing data AND AI – regulations. Laws are here and more are coming. Take the EU Act – a pioneering regulation designed to govern the ethical use of AI. The EU has always been at the forefront of regulations for technology – compliance, data usage, data privacy, and now AI. Smart organizations take heed of these laws early, because inevitably other countries follow suit with similar regulations. Oh, and by the way, there are large fines up to 30 million Euros for non-compliance. In case you needed an additional incentive for your governance program, there it is.

The guidelines are just the beginning for governing data and AI. They start by classifying the risk of the AI use case based on outcomes that could violate fundamental rights, security, privacy and ethics. The guidelines also require data governance “to ensure quality and representativeness to minimize biases.” In addition, transparency is required on decisions and how AI works.

Upcoming Webinar - Prevent AI from Creating Customer Mishaps
Why Data Governance is the Key

October 29 - 2 PM ET / 1 PM CT / 11 AM PT

Without a filter, customer-facing AI could make mistakes. Embarrassing mistakes.

Learn how to govern data & AI together to prevent it.

In my opinion, this is a good start, but a lot more is needed. Here are a few areas where I think organizations should go above and beyond the current regulations, in order to get a strategic advantage in their markets.

Create a single governance and catalog technology for all data and all AI models and apps. Right now, data and AI governance is often tracked in different systems. If we are being honest, a lot of organizations have different technologies for data governance on its own (governance of operational data vs. analytical data in a lake or warehouse, governance of structured vs. unstructured data).
Think hard about data representativeness. This isn’t usually tracked or managed in data catalogs or governance tools. But it easily could be. Profile your data sets for key attributes. Incorporate third party data and statistics to determine what ideal representation means. Then define acceptable levels of deviation from that ideal. Now you can measure the representativeness of your data for training and operating AI.
Get serious about bias. AI models are tested and scored for bias today. Are you running detailed simulations of bias based on different data sets at different levels of representation? You should be. Using the new metadata from step 2, you can take AI bias analysis and decisioning to the next level.
Make governance decisions on data and AI in the same process. Building from step 1, you can now have governance approval processes for data (training and operations) and AI.
It’s time to write down your ethics. Ethics has been a vague concept of ‘what’s right’ for too long. Write it down. What sort of organization do you want to be? How will you treat external stakeholders, like your customers? Specifically how will you market, sell and service them? And what data is ok for you to use, what insights are ok to generate and action on, and how can AI be used? Ethics are policies. They sit one level above data governance and AI policies and can be mapped to them. Your data governance tools can be configured to do this. Without setting those broad set of rules, how will you be able to set governance policies that reflect who you want to be?

This may seem daunting, but it doesn’t have to be. Aside from regulations from governments, there is an even more important party who will hold you in compliance. Your customers. The stakes for AI are much higher than past programs with analytics. As you develop AI models, co-pilots and applications, these tools will interact directly with your customers. Unfiltered. The consequences of an AI mishap will cost you much more than 30 million Euros in terms of brand value and customer attrition. In fact, ethical AI and data management can be a brand differentiator and actually grow your revenues.

The next step is a maturity assessment and audit of your current capabilities for governing data AND AI. Now is the time to do that, as 2025 planning is upon us. I’d like to help. Q Spark is offering a free AI and data governance mini-audit until November 30. Send me an email at dcorrigan@qsparkgroup.com if you’d like more details. Until then, remember this – governing data and AI isn’t just a moral or legal issue, it is a customer issue. Act now to preserve your valuable customer relationships.