AI is accelerating how business leaders process information to make informed decisions and add top- and bottom-line value to their organizations.
In the 2025 McKinsey Global Survey on AI, 78% of respondents reported that their organizations now use AI for at least one business function, a 55% jump from the previous year’s survey. As AI’s influence expands, so does the need for leaders to implement robust AI governance and ethics guidelines. CEO oversight of AI governance is the element of generative AI use most associated with increased self-reported bottom-line impact. Yet the public remains skeptical about AI. Only 23% of U.S. adults said they trust businesses to handle AI responsibly in a 2024 Gallup/Bentley University survey.
Deepti Kunupudi, former chief data and analytics officer and current advisory board member for multiple companies—who has led AI teams in the tourism, fintech, and pharmaceutical industries—emphasizes that good AI models don’t exist without good data.
“Every industry is different, but the underlying issues around AI ethics considerations are the same,” says Kunupudi. “AI governance is so important because the models are being trained on the data that we provide. If the data is confined to one set of variables but ignores another, then we see problems with the model not behaving correctly. We first need to understand the risks associated with AI and categorize the data and its impact. There’s always a risk associated with the data we use, whether it impacts revenue generation or cost reduction.”
BEST PRACTICES FOR AI GOVERNANCE
AI governance shouldn’t be an afterthought. Kunupudi encourages business leaders to build AI governance into their systems and processes by prioritizing four key components: fairness, transparency, accountability, and privacy.
“Organizations typically want to experiment with AI, but they don’t want to fully invest in it,” says Kunupudi. “Leaders need to know that when they build AI, even if it’s a prototype, they have to think through what data they’re using and how they’re using it. Can the data be trusted? Is it transparent enough that they can share the factors that are shaping their decisions?”
1. Fairness
AI is simply a tool that helps us generate output or insights—and it’s only as good as the data we feed it. If training data is inconsistent or biased, a model may hallucinate or produce unreliable results.
Businesses use customer data to make decisions. To mitigate outputs that can perpetuate discrimination, leaders need to identify bias in training data and verify that no sensitive attributes are shared in any step of the training’s feature selection. If their customer data contains more information about one race or gender group over another, for example, that can interfere with the AI model’s outcomes. Leaders might see the outputs and assume that their female customers don’t buy a certain product, but flawed foundational data has influenced the entire decision-making process.
“Whenever we train a model on data, we need to make sure we are using diverse, inclusive data sets,” says Kunupudi. “We can also make sure that we implement algorithmic fairness techniques. That means that we are looking closely at how and when we are pre-processing, in-processing, and post-processing the data in model training. Having a clear path is crucial. Otherwise, you’re already introducing bias into the data. And once biased data is in place, it’s very hard to identify and undo.”
2. Transparency
It’s not enough for leaders to show stakeholders the outputs of an AI model; they also need to be able to explain how the model generated these outputs. They must understand the technology and the data behind every AI model, building well-defined documentation processes and monitoring systems that team members regularly evaluate and improve.
“For each use case, you need clear documentation around what data is being used and what variables exist within that particular data set,” says Kunupudi. “You cannot completely rely on the AI model to eliminate bias and provide reliable outputs. You need to have humans in the loop who are reviewing the data before it’s shared with senior management. How frequently are you going to use the model output? How do you enable feedback and reporting so that everything is aligned? It’s a continuous, iterative process.”
3. Accountability
If everyone is in charge of AI governance, then no one is in charge. Leaders should be able to answer the following questions: Who is managing each specific data set, customer transaction, or AI model? Who is accountable for making decisions with AI output?
Kunupudi recommends making one person ultimately responsible for decision-making and
establishing a dedicated AI governance committee to oversee the implementation of ethical standards and guidelines. In her company, an AI steering committee meets bi-weekly to evaluate all use cases, monitor and improve processes to meet guidelines, and align resources across different teams.
“Multiple cross-functional teams are involved in governance,” she explains. “There’s an engineering team working on data pipelines, an AI team building models, and a business team leveraging outputs. When you’re working with cross-functional teams, you need to understand who is accountable for what in multiple data sets.”
4. Privacy
Privacy regulations are constantly evolving across geographic regions. To compete in a global marketplace, a business must create an AI governance framework that meets data privacy laws for each of the places it operates.
“If you’re getting data from Europe, Asia, and North America, it’s all integrated, but the rules are different,” says Kunupudi. “If you’re working in just one model, you’re setting yourself up for failure from an audit standpoint. It’s critical that you understand each country’s limitations and work with them so you protect individual and organizational data and avoid paying fines.”
Kunupudi stresses that AI governance should enable, not block, business growth. By incorporating fairness, transparency, accountability, and privacy into their AI frameworks from the beginning, organizations can save significant time, money, and hassle.
“Let’s reframe governance as a way to reduce future work and improve the quality and impact of AI solutions,” says Kunupudi. “You can start with a certain framework that aligns with regulations and builds best practices around your organization’s goals. It doesn’t have to be perfect; get something in place, and refine it over time.”