Data Benefits AI; AI Benefits Data

Without data, there is no AI. Without AI, data is… well, it’s complicated.

Getting one’s data house in order tops the list in AWS’s recently published Blueprint Executive Blueprint for Enterprise AI Transformation, highlighting the urgency of rethinking data strategy for the times ahead. “As large language models and foundation models become widely accessible via out-of-the-box apps, differentiation lies not in the model itself but in the quality, structure, and accessibility of the data powering it,” the blueprint’s authors state.

Absolutely.

This means opening up information from a variety of new or previously underused resources. “Unlike conventional systems that rely primarily on structured data, genAI demands comprehensive access to all data types – including unstructured and multimodal formats such as video, audio, text, and code – with real-time accessibility.”

Along with the need to manage data comes the need to transform architecture and put strong governance in place. This is the challenge, however. Only 26% of chief data officers are confident their data capabilities can support new AI-enabled revenue streams, a survey out of the IBM Institute for Business Value finds. They struggle to use their data to power AI. The top data barriers they face on this front are accessibility, completeness, integrity, accuracy, and consistency.

It’s notable that the relationship between AI and data flows both ways. Not only is having the right data at the right time key to AI, but AI can also assure that the data is being put to good use.

At this time, “agentic AI is the technology most impacting data management today,” said Ashwin Patil, principal and data engineering and analytics practice leader at Deloitte Consulting. “Most organizations handle large volumes of structured and unstructured data in the context of genAI. Agentic AI significantly augments or automates a once very manual process of profiling data, performing quality checks, building business rules and integrating data across applications.”

LLMs can be deployed directly into a data infrastructure, said Jim Liddle, AI entrepreneur and former chief innovation officer of data intelligence and AI at Nasuni. “This goes beyond traditional AI assistants layered on top of legacy file systems. Instead, it introduces a semantic understanding layer that fundamentally changes how unstructured data is stored, discovered, classified, and acted upon.”

In the near future, “platforms won’t just sync files; they’ll interpret content, context, and usage patterns,” Liddle predicted. “With access to all data through a unified namespace, organizations can unlock dormant data that has historically been trapped in archives or file shares. This transition marks a shift from passive file storage to intelligent, business-aware data ecosystems.”

There is also a notable benefit to the people handling data. “The real impact is on the people responsible for putting that data together, determining what that data is and orchestrating its movements,” said Aron Semle, chief technology officer at HighByte.
For example, LLM capabilities are being built into data management projects to help users answer questions like ‘what’s the relationship between these two data sets?’ and helping to write an SQL query.”

In the process, “LLMs are helping experts speed up the time to discover, deliver, move and troubleshoot data across data management platforms while lowering the bar of entry for less experienced users,” Semle added. “One benefit we’re already seeing from LLMs is that they are making data more accessible to a broader audience, which could grant decision-makers more control over their data and help them use it more effectively.”

Thus, an AI-first data strategy pays off in many ways. The AWS authors recommend five steps to properly integrate data into an AI-focused system:

Conduct a data audit: Establish use cases that take advantage of low-hanging fruit, to quickly show success: “For example, reducing service-call handling time by 30%,” the authors suggest. “Unify the relevant data in a secure, scalable storage
solution and implement appropriate guardrails immediately.”

Modernize your data architecture: “This consists of breaking down silos and
defining data product owners for key business units. Establish common
governance structures.”

Build internal capabilities: Upskill data teams “in prompt engineering,
vector databases, and responsible AI, while training teams across the
organization on AI fundamentals and responsible use to maximize adoption.”

Make sure there’s a human in the loop: A human in the loop, along with the use of LLMs to provide feedback, can be employed to “continuously monitor and improve data quality and model performance.”

Measure progress: “Track business outcomes, operational performance metrics, and data and trust metrics such as retrieval precision rate, factual consistency score, and daily active users.”

Forbes