Why AI is only as good as your data

Dubai: Data is the foundation of modern business, but its value depends on one critical factor: trust. As AI becomes more capable and autonomous, organisations face new questions around accuracy, governance, accountability and risk. Reece Clifford, Public Sector Pre-Sales Manager for the Middle East, Turkey and Africa at SAS, shares his views on the evolving relationship between data, AI and business decision-making, and why strong governance is becoming a competitive advantage rather than a compliance exercise.

Data accuracy, trust and reliability

How accurate is the data organisations rely on for critical business decisions today? What are the biggest threats to data quality and reliability? How do you define data accuracy in a modern enterprise?

Reece Clifford: Before answering how accurate data is, I want to emphasise how important and fragile trust in data and AI is within an organisation. Trust is the reason we worry about data accuracy and is what we try to preserve through our Data Quality and Data Governance programmes. With the introduction of Agentic AI, trust becomes even more critical because it lowers the barrier to entry for many specialist tasks or areas of knowledge that users did not previously possess. When using different providers, a single incorrect answer can quickly erode trust.

To answer your question, data accuracy depends on the specific use case. Does a credit decisioning model have accurate, complete, timely, relevant and traceable data? Most probably, yes. However, with the introduction of Agentic AI and its ability to contextualise vast amounts of data, overall data accuracy across the enterprise is not as high.

Many organisations have improved their ability to collect vast amounts of data, but the core challenge is ensuring it remains reliable. The major threats to data quality are data silos, inconsistent definitions, incomplete records and, increasingly, poor-quality AI-generated content entering business workflows.

What misconceptions do business leaders have about data quality and reliability?

Reece Clifford: Analytics and AI only amplify the quality of the data on which they are built. It is a principle known since the inception of computer science: garbage in, garbage out.

The primary misconception is that data quality is purely an IT issue. It is not. Data quality is a business-led issue because it directly affects customer experience, risk management, forecasting and strategic planning. It requires collaboration between business teams responsible for definitions and usage, and IT teams responsible for data values and storage.

Leaders also frequently overestimate the maturity of their data foundations. A SAS report developed with research insights from IDC found that many organisations are still developing the infrastructure needed to support explainability, validation and governance at scale, particularly as AI becomes more autonomous.

How easy is it to manipulate data, intentionally or unintentionally? To what extent does human bias influence data?

Reece Clifford: Humans influence every stage of the data lifecycle: what data to collect, which variables matter, how results are interpreted and what actions are taken. Most data issues are not malicious; they are usually the result of inconsistent processes, sampling errors or unconscious bias, compounded by the increasing volume and complexity of data.

Because decision-making requires explainability and transparency to maintain trust by reducing and ultimately removing manipulation or bias, investing in specialists remains a priority for organisations. The concept of complete automation is closer through the introduction of Agentic AI, but we are not there yet.

Have you seen organisations make major decisions based on flawed data? What lessons were learned?

Reece Clifford: Every organisation has experienced some form of a data quality issue, and numerous examples are publicly documented.

Overall, given the rapid pace of innovation in data insights, the most important lesson is not to skip guardrails and security measures. This is especially vital when a system or output is externally facing. Data quality checks and governance policies are essential components of these protective measures.

Data verification and governance

What frameworks and controls should organisations implement to verify data authenticity and maintain trust?

Reece Clifford: The best results are achieved when people, processes and technology are combined into a single framework.

At a practical level, that requires strong data lineage, quality monitoring, audit trails, access controls, clear business definitions and defined ownership. Governance works best when embedded into day-to-day decision-making rather than managed as a standalone compliance exercise. The most advanced approaches allow in-production issues and errors to be identified and rectified before becoming problems because the underlying processes are well defined.

What warning signs suggest a dataset, report or dashboard cannot be fully trusted?

Reece Clifford: The simplest warning signs are often the most useful:

How can organisations balance governance requirements with the demand for real-time insights?

Reece Clifford: Governance and speed are often presented as conflicting priorities, but they are not. Quality checks, lineage tracking, model monitoring and policy enforcement can be built directly into workflows. Modern technological integration has made this significantly easier.

In fact, the organisations moving fastest today are often those with the strongest governance foundations. Because they do not build "quick and dirty" systems initially, they avoid paying the price later and can scale subsequent innovation seamlessly. Governance is increasingly an accelerator rather than a barrier.

Artificial intelligence and data

How is AI changing the way organisations collect, process and analyse data?

Reece Clifford: The change can be summed up in one word: automation.

AI capabilities and their supporting workflows allow tasks to be completed faster or executed when they were previously impossible. Collection, processing and analysis now happen in subseconds, seconds or minutes rather than hours, days or months. In many cases, human intervention is no longer required. Automating tasks such as pulling data sources, processing documents and triggering analytical predictions in near real time is now standard practice.

Tasks including contextualising vast data sources to create expert assistants or generating plain-language and voice-driven insights from enterprise data are now achievable through Large Language Models (LLMs), Retrieval-Augmented Generation (RAG) applications and Model Context Protocols (MCPs).

The caveat remains: not all processes should be fully automated.

Has AI improved decision-making overall, or introduced new risks?

Reece Clifford: The answer is both.

AI has undoubtedly improved the accuracy and insight available for decision-making across operational efficiency, fraud detection, customer intelligence and risk assessment. Many organisations have realised measurable value from these applications for years.

However, as innovation in the field of AI continues, new complexities introduce new risks. With generative AI, risks such as hallucinations or model bias can lead to incorrect or unbalanced outputs. To combat these risks, new frameworks are emerging, including the progression towards Graph Engineering specifically for Agentic AI.

According to research undertaken by SAS, the organisations seeing the greatest return from AI are not necessarily the fastest adopters; they are the organisations that invest in trustworthy AI practices and human oversight.

How serious are threats such as data poisoning, misinformation and low-quality AI-generated content?

Reece Clifford: They are becoming increasingly significant. As AI systems consume larger volumes of external content, the quality of those inputs becomes critically important. A simple example is the rise of AI-generated images on visual search engines or in property advertisements, which create misleading expectations and erode consumer trust. The same principle applies to organisational decision-making at every level: poor inputs pose a serious threat to decision integrity and diminish trust in the technology.

The solution is not to avoid AI. Strategically, organisations must strengthen governance, validation and monitoring. On an individual level, users must consistently ask: "Would I trust the output if this specific input was used?"

Can AI be trusted to verify data independently, or is human oversight still essential?

Reece Clifford: AI is exceptionally effective at identifying anomalies, detecting patterns and automating quality assessments. However, human oversight will always remain essential because AI is inherently limited by the data on which it is trained. For instance, ongoing discussions about whether AI-generated images should be explicitly labelled demonstrate where human oversight directly impacts policy and trust.

At SAS, we view AI as a decision-support technology rather than a replacement for human judgement. The optimal outcomes occur when AI and humans collaborate, with human oversight providing necessary context, ethical judgement and supervision.

Future risks and challenges

What challenges should organisations prepare for as AI evolves towards real-time and autonomous decision-making?

Reece Clifford: The primary area of evolution, and confusion, is automation. Agentic AI presents opportunities for automation that were previously impossible. However, Agentic AI does not equal total automation; just because a process can be completely automated does not mean it should be. For example, few people would want their entire healthcare experience to be automated without human intervention. The human element cannot be ignored in such contexts.

More broadly, fundamental questions such as "Who is responsible when an AI system makes a mistake?" and "How do we audit autonomous decisions?" will become increasingly critical. AI governance is a boardroom-level priority rather than a purely technical concern.

What risks do organisations face if AI adoption outpaces governance?

Reece Clifford: Risks range from operational errors and biased outcomes to regulatory penalties and, most critically, reputational damage.

Reputational damage resulting from a loss of trust can define the future of a business. Once customers, employees or regulators lose confidence, the business suffers regardless of whether a human or an AI made the decision. Rebuilding that lost confidence is far more difficult than establishing proper guardrails in the first place.

Ethics, privacy and regulation

How can organisations balance innovation with privacy, transparency and responsible data use?

Reece Clifford: Responsible innovation starts with a fundamental question: not merely "Can we do this?" but "Should we do this?" The most successful organisations embed ethical thinking into every stage of development as a core component of the process rather than as an afterthought.

Utilising technology that provides these capabilities natively simplifies the task. For example, maintaining a central repository for all models, regardless of programming language or origin, that automatically conducts performance testing, communicates outcomes in plain language and performs health checks to monitor data drift ensures ongoing compliance and control.

Do current regulations adequately address AI risks?

Reece Clifford: Regulations are advancing, but technology continues to evolve rapidly.

Important progress is occurring globally, with regulatory frameworks in the European Union, the United States, South Korea and elsewhere focusing on transparency, accountability and risk management. These frameworks provide a baseline that all organisations can adopt, even if they are not strictly mandatory in their jurisdiction. Adherence to these standards offers a competitive advantage and lays the foundation for scalable, sustainable innovation using cutting-edge technology.

Looking ahead

How do you expect the relationship between data, AI and business decision-making to evolve over the next decade?

Reece Clifford: The pace of innovation has accelerated so rapidly that predicting where the market will be in two years, let alone a decade, is challenging. However, organisations will continue to push the boundaries of what AI systems are trusted to execute. My team already uses multiple agents to supplement their work, and we continually evaluate whether and how we can trust those outputs and what that means for our roles.

AI will not replace humans over the next ten years, but it will fundamentally alter how we work. The faster we can trust AI to perform specific tasks, the faster that transformation will occur: decisions will be made more quickly, outputs will become more accurate and automation will become more widespread.

If you could give business leaders one piece of advice about using data and AI responsibly, what would it be?

Reece Clifford: Try the technology yourself. The barrier to entry is lowering rapidly; you no longer need a computer science degree to create a model or user interface development skills to build a dashboard. Hands-on experience helps leaders build trust in the technology while understanding its practical limitations.

Direct experience ensures technology is deployed responsibly, data is managed correctly and trust is maintained among internal users and external customers. Trust is the foundation that turns data into decisions and AI investment into tangible business impact.