The Critical Link Between Data Quality and AI Systems

For Generative and Agentic AI, High-Quality Data Is Essential

As the use of artificial intelligence (AI) tools continues to expand, and observers debate its practical and ethical nuances, an important point is frequently overlooked: The quality and value of the outputs of AI applications hinge greatly on the caliber of data that is "educating" and informing those applications.

Businesses increasingly rely on AI systems to drive decision-making and innovation. In this environment, it's not an option to brush off or gloss over the relationship between data quality and AI performance. Data quality, perhaps regarded by many as a technical concern for IT and data management professionals, has been promoted to the level of a strategic business priority — something that directly impacts the value organisations will realise from AI investments.

A recent survey conducted by Dun & Bradstreet reflects this growing awareness; the survey showed that a vast majority of organisations (88%) are now implementing AI, but more than half (54%) have concerns over the trustworthiness and quality of the data they are leveraging for their AI systems.

Advances in generative and agentic AI further amplify the need for accurate, complete, and unbiased data sets, as these technologies learn and generate outputs from the information they ingest. As data volumes and complexity continue to grow, maintaining data quality becomes central to managing operational risks, meeting regulatory requirements, and building stakeholder confidence in AI-driven initiatives.

Consequently, organisations that embrace best practices in data quality management are positioned to achieve more sustainable AI adoption. These organisations will also lead their businesses in driving innovation and establishing a competitive edge in a data-centric economy.

This article explores the importance of providing high-quality data within AI projects, the challenges posed by poor data quality, and the future implications for data in AI technologies.

Understanding Data Quality in AI

Data quality refers to the condition of a data set and its ability to serve its intended purpose effectively. In the context of AI, high-quality data is crucial for training models that produce accurate and reliable outputs. Poor data quality can lead to flawed insights, biased results, and ultimately, a lack of trust in AI systems.

The Importance of Data Quality

High-quality data is the foundation of any successful AI initiative. It ensures that AI models are trained on accurate, relevant, and comprehensive data sets, leading to more reliable predictions and insights. Businesses that prioritise data quality can harness AI to outpace competitors, streamline their processes, and deliver superior experiences to clients.

Challenges of Ensuring Data Quality

The sheer volume of data generated in the world today makes it difficult to maintain data quality across all sources. Data can often be incomplete, inconsistent, or outdated, leading to inaccurate AI model outputs. Organisations must implement robust data quality management practices to address these challenges and ensure the integrity of their AI systems.

A Short Definition of Quality Data

Identifying high-quality data isn’t always straightforward. Its usefulness often depends on who’s using it and for what purpose. To help guide businesses in shaping a strong data strategy, here are four essential characteristics that generally define quality data:

Accuracy

At its core, accuracy means the data is correct. For instance, if you're sending a package to a client, the address must be precise enough to ensure successful delivery.

However, accuracy can be relative. Consider a business with multiple phone numbers — if you're trying to reach the main office, the investor relations line won’t help. Accuracy must be evaluated in context. It’s not always about aligning with a universal truth, but rather about how well the data fits the specific need or scenario.

Completeness

This refers to whether a data set captures all relevant information for a given case. If a customer’s record only includes a third of their purchases, their value to the business will be underestimated. Incomplete data can lead to missed opportunities, especially when trying to identify top-tier customers.

Standardisation

Standardised data makes it easier to compare and analyse across different sources. For example, having consistent formatting for addresses allows companies to match and evaluate customer information globally — even when dealing with vastly different address structures, like those in Tokyo versus New York. Standardisation is key not only for data entry but also for spotting duplicates and ensuring consistency.

Authority

Reliable data comes from trusted sources. Your data should originate from credible, well-regarded providers that are appropriate for your business needs. When your data sources are authoritative, your decisions are more likely to be sound. Without trustworthy inputs, even the most sophisticated analysis can lead to flawed conclusions.

The Impact of Poor Data Quality on AI Systems

Poor data quality can significantly and negatively impact the performance and trustworthiness of AI systems. Understanding these impacts is crucial for businesses looking to harness the full potential of AI technologies.

Consequences of Poor Data Quality

1. Inaccurate Predictions: AI models trained on poor-quality data are likely to produce inaccurate predictions, leading to misguided business decisions.
2. Bias and Discrimination: Incomplete or biased data can result in AI systems that perpetuate existing biases, leading to unfair outcomes and missed opportunities.
3. Loss of Trust: When AI systems produce unreliable results, stakeholders may lose trust in the technology, hindering its adoption and effectiveness.

Examples of AI Failures Due to Bad Data

Several high-profile AI failures have highlighted the importance of data quality. For instance, AI systems used in hiring processes have been criticised for perpetuating gender and racial biases due to biased training data. In another example, healthcare AI algorithms designed to predict patient outcomes have delivered less effective recommendations for underrepresented populations when trained on data that lacks diversity or completeness.

Furthermore, chatbots and language models have produced offensive or misleading content when exposed to large volumes of uncurated and unchecked text. For instance, some early language models inadvertently repeated biased stereotypes or used discriminatory language in their responses. In other cases, chatbots have perpetuated conspiracy theories because inaccurate data was included in their training sources. There have also been incidents where chatbots adopted toxic language or made inappropriate comments when prompted by manipulative users.

These scenarios demonstrate that AI is only as trustworthy as the data it learns from, making rigorous data quality management not only a technical requirement but a fundamental business responsibility. These cases illustrate what can go wrong when organisations neglect the integrity of their training data — flawed outcomes are not just possible, but probable.

Generative AI and Data Quality

Generative AI, a subset of AI that focuses on creating new content, relies heavily on high-quality data to function effectively. This section explores the relationship between generative AI and data quality, highlighting the challenges and opportunities it presents.

The Role of Data in Generative AI

Generative AI models, such as large language models (LLMs), require vast amounts of data to generate coherent and contextually relevant content. The quality of this data directly impacts the model's ability to produce accurate and meaningful outputs. High-quality data ensures that generative AI models can create content that meets user expectations and aligns with business objectives.

Challenges of Generative AI With Respect to Data

• Data Availability: Generative AI models require diverse and comprehensive data sets to function effectively. Ensuring the availability of such data can be challenging, particularly in niche domains.
• Data Bias: Generative AI models are susceptible to biases present in their training data, which can lead to biased outputs. Addressing data bias is crucial to ensure fair and equitable AI systems.
• Privacy and Data Protection: Businesses must now walk a fine line between innovation and responsibility, ensuring that their use of AI respects individual privacy and complies with evolving legal standards. While this data may be publicly accessible, it doesn’t always mean it’s ethically or legally appropriate to use. Even where it may be ethical or legal, such use may be conditioned on a number or requirements that need to be met to ensure that the rights of individuals are transparent and honored.

Agentic AI and Data Quality

Agentic AI, which refers to AI systems capable of autonomous decision-making, also relies on high-quality data to function effectively. This section explores the concept of agentic AI and its relationship with data quality.

What Is Agentic AI?

Agentic AI systems are designed to mimic human decision-making processes, allowing them to operate with limited supervision. These systems rely on accurate data to make informed decisions and adapt to changing environments. Unlike traditional AI models, such as expert systems. that often require explicit programming or rule-based logic, agentic AI leverages advanced probabilistic algorithms and real-time data streams to assess situations, learn from outcomes, and adjust strategies accordingly. As these AI agents are tasked with supporting or automating complex business functions — such as supply chain logistics, risk assessment, and customer engagement — the quality of the information they consume becomes directly linked to performance outcomes and stakeholder trust. 

The Importance of Data Quality in Agentic AI

For agentic AI systems to function effectively, they require access to data that is trusted, complete, and relevant. High-quality data ensures that these systems can make informed decisions, respond to real-time changes, and achieve their intended goals. Poor data quality can hinder the performance of agentic AI systems, leading to flawed reasoning, suboptimal outcomes, and user distrust.

Best Practices for Ensuring Data Quality in AI

To harness the full potential of AI technologies, businesses must implement best practices for ensuring data quality. This section outlines key strategies for maintaining high-quality data in AI projects.

Data Quality Management

1. Data Governance: Establishing a robust data governance framework is essential for maintaining data quality. This includes defining data standards, roles, and responsibilities to ensure data integrity and compliance.
2. Continuous Monitoring: Implementing continuous monitoring practices allows businesses to identify and address data quality issues in real-time, ensuring that AI models are always trained on accurate and up-to-date data.
3. Data Cleaning and Preprocessing: Regularly cleaning and preprocessing data helps eliminate errors, inconsistencies, and biases, improving the overall quality of the dataset.

Assessing Data Quality

1. Data Quality Frameworks: Utilising data quality frameworks can help businesses assess the quality of their data and identify areas for improvement. A robust data quality framework typically includes setting clear definitions of data quality requirements, establishing metrics for each dimension, ongoing monitoring and measuring, and implementing improvement initiatives based on regular assessments.
       a. Examples of Data Quality Frameworks: Some widely recognised frameworks include the Total Data Quality Management (TDQM) framework, the Data Management Association’s DAMA-DMBOK (Data Management Body of Knowledge) data quality framework, and ISO 8000. For example, the TDQM framework emphasises the continuous assessment and improvement of data across four phases: define, measure, analyse, and improve. The DAMA framework breaks data quality down into key dimensions such as accuracy, completeness, consistency, currency, and relevance, offering metrics and processes for each. ISO 8000 provides standardised guidelines for ensuring accurate and portable data quality across diverse systems.
2. Human-in-the-Loop: Incorporating human oversight in AI processes can help identify and address data quality issues, helping to ensure that AI models are trained on high-quality data.

The Role of Data Quality Analysts in Enhancing AI Performance

Data quality analysts play a pivotal role in optimising the performance of artificial intelligence systems. By carefully monitoring, assessing, and refining data sets, these professionals ensure that the data feeding AI algorithms is accurate, consistent, and reliable. Their expertise makes it possible to identify discrepancies, missing values, and errors that could negatively affect model outcomes.

Data quality analysts use advanced tools and methodologies to cleanse and standardise information, collaborate with data scientists and engineers, and implement protocols for ongoing data validation. Their rigorous approach not only boosts AI accuracy but also helps prevent model bias, supports compliance efforts, and helps enterprise leaders make better decisions.

Improve Data Quality With Master Data Management

For AI to operate effectively, it relies on data that is trustworthy, thorough, and easy to access. That’s why many organisations turn to master data management (MDM) to maintain consistency and structure across their data assets.

Master data management encompasses a set of strategies and tools designed to unify enterprise data into a single, authoritative source — often referred to as a “golden record.” Its goal is to break down data silos and reconcile duplicate or conflicting entries, ensuring that everyone in the organisation is working from the same current and reliable information. This alignment is especially valuable for large enterprises that need to stay coordinated.

When a company begins using generative or agentic AI, the importance of master data management becomes even greater. AI systems aren’t yet capable of critically evaluating questionable data the way humans can. To avoid training models on flawed or biased information, it’s essential to keep data well-maintained, accurate, and systematically organised.

The Future of Data Quality in AI

As AI technologies continue to evolve, the importance of data quality will only increase. Businesses that capitalise on these emerging trends and innovation opportunities will be better positioned to leverage AI for innovation and growth:

• AI-Driven Data Quality Solutions: The use of AI to enhance data quality management is an emerging trend. AI-driven solutions can automate key tasks such as:
  • Integrating, cleansing, and verifying data to minimise inconsistencies and enhance overall reliability
  • Enriching data by adding context or supplementary information to boost its usefulness
  • Generating meaningful insights that support smarter, data-driven decisions
  • Spotting nuanced patterns and emerging trends that may be overlooked by human analysts
  • Detecting anomalies and potential fraud through advanced pattern recognition
• Data Observability: Data observability, which involves monitoring data pipelines and systems for anomalies, is becoming increasingly important for maintaining data quality in AI projects.
• Synthetic Data: The use of synthetic data, which is artificially generated data that mimics real-world data, presents opportunities for improving data quality in AI projects. Synthetic data can help address data availability and privacy challenges, enabling businesses to train AI models on high-quality data sets.
• Collaboration and Partnerships: Collaborating with industry partners and stakeholders can help businesses access diverse and high-quality data sources, enhancing the performance of their AI systems.

Conclusion

The relationship between data quality and AI is critical to the success of AI initiatives. High-quality data helps ensure that AI models produce valid and reliable outputs; the phrase "garbage in, garbage out" is perhaps overused but not inaccurate in this context. As AI continues to evolve, the importance of data quality will only increase as AI outputs train other AI models in a continuing cycle. By prioritising data quality and implementing data management best practices, businesses will be able to harness the full potential of AI technologies and stake leading-edge positions in their industry sectors.