Why Data Quality, Security And Governance Will Always Drive AI Success

Forbes

20-05-2025

Dr. TJ Jiang, Chief Executive Officer and Co-Founder of AvePoint.
getty
The arrival of DeepSeek's R1 large language model (LLM) shocked the global AI ecosystem, causing many in the U.S. and Europe to reevaluate how we approach AI development.
While LLMs from large organizations like Open AI, Anthropic, Meta and others were trained at tremendous expense, DeepSeek was trained for a fraction of the cost, though the financial numbers from DeepSeek have been called into question.
Still, now that we know lean teams can develop AI models for significantly less money, a democratic revolution in AI development is already taking place, with businesses now empowered to develop their own models—a radical shift from the pre-DeepSeek world.
Let me be clear: This is a good thing. More competition inevitably leads to greater innovation. We've seen this happen with cloud computing, open-source software, mobile operating systems and now AI. But this development doesn't solve the challenges that some organizations were facing before the arrival of DeepSeek.
To do that, we must rethink the way we approach data security, data quality and data governance.
A low-cost, open-source AI model means that high-quality LLMs are no longer the primary differentiator for businesses. Now, instead of focusing on developing or adopting the most complex model, it's about ensuring that models are efficiently and securely applied, refined and integrated within specific industries and business processes.
This is why enterprise data is an increasingly key focal point. Companies like OpenAI, Anthropic, DeepSeek and others can train their models on vast public datasets, but they do not have access to enterprises' specific information. That proprietary data—such as customer interactions, operational insights and historical records—is what can make AI a true competitive advantage for businesses.
In a recent survey, IBM found that 15% of forward-thinking AI businesses (whom IBM calls 'AI leaders') are already realizing measurable benefits from their AI initiatives. One key differentiator for these AI leaders is their ability to adapt AI models using their proprietary data. Rather than developing models from the ground up, these organizations refine and enhance existing AI platforms with unique datasets that competitors lack.
To do this effectively, businesses will need to strengthen data quality, data governance and data security.
AI models trained on unreliable, inconsistent, outdated or biased data produce poor results that can erode trust and hinder adoption. My company's "AI and Information Management Report" found 95% of organizations are experiencing challenges during AI implementation, with more than half (52%) specifically challenged with data quality and categorization during implementation. This highlights the critical importance of:
• Data cleaning and augmentation to ensure high-quality training datasets.
• Metadata tagging and structuring to enable seamless AI integration.
• Bias detection and mitigation to improve model fairness and accuracy.
Data governance is no longer just a compliance requirement; it is a fundamental enabler of AI success. Businesses need robust governance frameworks that:
• Maintain data security and privacy, especially for AI training data.
• Ensure compliance with evolving regulations like GDPR and CCPA, as well as new regulations such as the EU AI Act.
• Standardize data access policies to prevent information silos.
Transparent data governance isn't just regulatory. It's foundational to AI user adoption.
The proliferation of open-source models driven by DeepSeek has highlighted the significant (but manageable) risks of open-source models. However, this is just one of the potential threats associated with AI adoption and use today.
AI models, particularly LLMs, are highly vulnerable, among many other risks, to attacks like training data poisoning that can compromise AI integrity, prompt injection attacks where malicious inputs can manipulate AI responses and accidental exposure of confidential data.
As AI moves from being a hypothetical prospect to a concrete business reality—something that I've spoken about recently—data security must always be top of mind.
There are several steps that organizations can pursue to help mitigate vulnerabilities today:
1. Implement strong data governance. Weak data governance policies can lead to accidental disclosures from poorly organized and secured data. By classifying or tagging data according to confidentiality and risk, organizations can limit their vulnerability. Strong data governance also helps organizations improve the output of their AI tools.
2. Be diligent about security and compliance. While the U.S. federal government is taking an increasingly hands-off approach to AI regulation, state governments are still drafting their own legislation and other regions like the EU are taking aggressive steps to legislate compliance. The EU AI Act, for example, is sweeping and powerful and affects organizations and individuals well beyond the EU's borders. As these regulations continue to proliferate, organizations need effective, model-agnostic solutions to manage compliance to avoid consequences from regulators.
3. Proactively measure and enhance AI-readiness. Organizations can measure and enhance AI readiness both by using technology to strengthen data security and compliance and by implementing governance strategies to regularly evaluate their AI models for vulnerabilities. Doing this proactively helps mitigate risks such as training data poisoning, prompt injection attacks and accidental data exposure, driving robust and secure AI adoption.
AI's value creation has shifted. No longer is the most advanced model the key differentiator. Instead, it is the proprietary data that fuels these models. With foundational AI models becoming commoditized, businesses must now focus on structuring, governing and using their own datasets to drive innovation and competitive advantage.
With enterprises prioritizing secure AI deployments, it's essential to ensure proprietary data is truly an asset, not a liability. Companies that invest in robust data management strategies will be the ones that thrive in the AI-driven future.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?