How data quality makes the difference in successful AI applications

Few organizations truly have their enterprise-level data in order. Our recent research shows that 36% of CIOs report their data platform infrastructure is inadequately prepared. According to Databricks, only 22% of organizations have sufficiently built their data infrastructure to implement AI effectively.We live in an era where data is generated in unprecedented volumes—billions of bytes every second across platforms and devices. This explosion of information presents enormous opportunities, but also significant challenges for both businesses and individuals.

AI promises improved automation, higher productivity, and lower costs. But the increasing volume, velocity, and variety of data introduces new obstacles. Regulations like the EU AI Act add pressure on organizations to manage their data properly. That’s why a solid data platform is essential. But how do you combine internal data with relevant external information from the internet? Where does an AI agent source its data, and what criteria guide its decisions? The output of one agent influences the actions of the next—a chain of decisions that must be carefully managed to ensure reliable results.

Without strong data quality and infrastructure, deploying AI effectively—especially autonomous agents—is a challenge. The key lies in building data products and agents that perform well, comply with regulations, and minimize risks such as errors, hallucinations, and bias.

Data: The fuel for AI success

In today’s digital world, AI is a hot topic attracting companies across sectors. Its promise to automate processes, reduce costs, and generate insights drives growing interest. But the key to successful AI implementation isn’t just the technology—it’s the quality of the data that feeds these systems. Simply put: garbage in is garbage out.

Our research explores the critical role of data in the AI revolution and offers insights for organizations looking to benefit from this technology. EY asked 500 CIOs about their biggest concerns when adopting AI, and 36% cited insufficient data and infrastructure platforms.

A fundamental challenge

These insights highlight a critical business challenge: how do you use data effectively in practice? As Gartner notes, the real differentiator today isn’t the large language models (LLMs) or AI models you build, but the quality and completeness of your data.

What CIOs are saying

• 21% struggle to meet regulations or face increased ethical and legal risks.
• 36% lack the right data and infrastructure platforms to implement generative AI.
• 31% are concerned about rising cyber and data privacy risks (e.g., data leaks, data poisoning).
• 5% risk reputational damage due to inaccurate data.
• 4% lack the right skills and capabilities; employees worry about job security.
• 2% don’t know, as ROI remains unclear.
  

What can go wrong?

Real-world examples show how things can go seriously wrong. Headlines speak for themselves:

• An airline was held liable after a chatbot gave passengers incorrect advice.
• Zillow’s failed AI project revealed the difficulty of valuing real estate with AI.
• New York City defended an AI chatbot that advised entrepreneurs to break laws.
• A major healthcare algorithm showed racial bias.
• Amazon scrapped a secret AI recruitment tool that discriminated against women.

It’s about trust, reliability, and security

Data—and how we differentiate ourselves with it—is the core. But as AI matures and scales, it brings challenges in data management, security, infrastructure, and governance/MLOps. To trust AI in decision-making, it must be fed accurate and reliable information.

Quantity vs. quality: The data dilemma

Organizations must ask themselves a crucial question: do we have access to all the data needed to achieve optimal results with AI? Many models are trained on publicly available data, but the real opportunity lies in effectively combining external data with internal business data. That integration is what sets organizations apart.

To trust AI in decision-making, it must be fed accurate and reliable information. Data quality is essential: complete, accurate, and error-free. But security is also increasingly important. As AI becomes more widespread, the risk of malicious actors exploiting faulty or manipulated data grows. Organizations must invest not only in combining internal and external data sources, but also in securing them. Only then can AI be deployed in a way that is trustworthy, responsible, and distinctive.