Unlocking a strategic approach to data and AI

Artificial intelligence (AI) is the promise of our age. It will change industries, enhance decision-making and unlock unprecedented efficiency and productivity.

But beneath the dazzling headlines and futuristic visions lies a fundamental truth: AI is only as good as the data that powers it. It is time for the spotlight to shift to this groundbreaking technology’s often-overlooked foundation – data.

And the truth is, you’re sitting on a goldmine of untapped potential. Every day, your business generates vast amounts of data (i.e., IoT sensor data, sales receipts, emails, patient health records). Some of those data sets have billions of columns and rows; some are spreadsheets with fewer than 1,000 columns and rows.

If you're not using the power of your data with AI, you're leaving money on the table. According to a study by McKinsey, companies that embrace data-driven decision making see up to a 25 percent increase in EBITDA. In addition, 86 percent of CEOs believe AI will help their companies maintain or grow revenue in 2024 and 2025.

The message is clear. Organizations must embrace data and AI to be business-forward. Data and AI aren't just for data engineers and data scientists but all roles within the data and AI life cycle.

In this article, we’ll explore an AI-powered approach to data management in the new era – one that covers practical considerations to ensure data accuracy, trustworthiness and responsible AI.

It's bigger than big data: The complicated nature of data and AI

Too often, when we discuss data and AI, we get stuck on the "big data analytics" narrative. Having a massive data set can be important, but it's just one piece of this convoluted puzzle. The true power of AI lies in the diversity, quality, timeliness and contextual relevance of the underlying data.

Consider the role of data in each of these emerging AI scenarios:

• Self-driving cars. Autonomous vehicles rely on a vast array of data types – from real-time sensor readings to historical traffic patterns to detailed map info. Deep learning models are trained on this data. The quality and timeliness of that data can be the difference between safely reaching your destination and ending up as part of a multi-car pileup.
• AI-assisted medical diagnoses. In health care, AI systems are trained on diverse data sets, including medical imaging, patient histories and genetic information. Natural language processing is used to analyze a mix of structured and unstructured data. The accuracy and representativeness of this data directly affect the reliability of AI-generated diagnoses.
• AI copilots. From coding assistants to writing aids, AI copilots are becoming ubiquitous. Their effectiveness hinges on training data that spans multiple domains, languages and stylistic variations.
• AI-driven marketing. Customer experiences driven by AI and machine learning will be the future. A marketing campaign that relies on large amounts of data allows a more personalized approach to understanding behavior and preferences with precision.

In each of these cases – and in every successful AI initiative – the challenge goes beyond the volume of data. It's about ensuring data accuracy, eliminating bias and providing the right context for AI to make informed decisions. And yet, when AI stumbles, produces biased results, or generates nonsensical outputs, our reaction is usually to blame the algorithms.

What's behind AI shortcomings?

According to Dan Soceanu, Senior Manager, Technology Product Marketing at SAS, this surface-level analysis misses the mark. AI’s shortcomings can almost always be traced back to the data or the question.

“The crazy thing about AI is that it's rarely a bad algorithm or a bad learning model that causes AI failures,” explains Soceanu. “It's not the math or the science; more often it's the quality of the data being used to answer the question.”

High-quality data is accurate, complete, consistent and timely. It's data you can trust to make critical business decisions. Poor quality data without proper data processing can lead to flawed insights, misguided strategies and costly mistakes.

When you prioritize data management, you're not just improving your data. You're laying the groundwork for trustworthy data models and successful AI implementation.

Data management: The foundation of data and AI success

Data management includes robust processes for handling data throughout its life cycle – from acquisition and integration to cleansing, governance, storage and preparation for analysis. It transforms raw data into fuel for your AI engine. Without effective data management, your organization will be data-rich in a data lake, but insight-poor.

Core considerations for effective data management

Here are some key data management-related techniques and concepts:

• Data quality assessment. Regularly conduct data analysis to evaluate data accuracy, completeness and consistency in analytics and AI.
• Data cleansing. Systematically identify and correct errors, inconsistencies and inaccuracies in data sets to improve AI model performance. (See this fun data cleansing challenge in SAS Community.)
• Metadata management. Maintain comprehensive metadata to provide context and facilitate proper data interpretation by AI systems.
• Data lineage. Track data's origin, transformations and movements to ensure transparency and aid in AI troubleshooting processes. “Just as it’s hard to have a good sense of direction without GPS, it’s hard to have a good sense of the data feeding AI without data lineage,” writes Jim Harris, Blogger-in-Chief at Obsessive-Compulsive Data Quality.
• Access control and security. Implement robust security measures and access protocols to protect sensitive data used in AI applications.
• Data integration. Effectively combine data from various sources to create a unified view for more comprehensive AI analysis.
• DataOps and AIOps. Align IT and business to ensure accurate results, reduce costs and increase productivity.
• Ethical considerations. Establish guidelines for responsible AI use and data flow, including detection and mitigation of bias in training data.

Responsible AI starts with data management

SAS established formal responsible innovation standards in response to the fast-paced AI era. When asked about SAS’ blueprint for responsible AI, Reggie Townsend, vice president of the SAS’ Data Ethics Practice, said: “Before you start writing a line of code, [you have] to activate a trustworthy AI environment … you must work on building a culture that is ethical by design.”

AI and generative AI (GenAI) will continue to push data management boundaries – not just now, but in the future. Considering the data source, organizations will need a strategy for modern approaches like synthetic data, which promises to improve the robustness of data sets or substitute for sensitive real-world data. Data management will continue to be the foundation of AI to ensure responsible use and outputs.

Faster, more productive AI

Innovation from data science teams will require flexibility in tools and choice across the data and AI life cycle. Whether open-sourced methods are used or not, outputs will need to be explainable, and data management will be key to improving decision making and business processes.

The road ahead: A modern data management approach to AI

Algorithmic innovations will continue to advance what’s possible, but the true key to unlocking AI's potential includes data management. Organizations that prioritize data quality, implement robust governance practices and ensure accountability and transparency in their use of AI will be in the best position to benefit from AI while mitigating its risks.