AI doesn't need an unstructured data dump -- move what matters
Bulk data migration pushes out AI timelines. Data leaders should invert the process: classify and catalog first, then move only relevant content to reduce delays and costs.
Enterprise AI adoption has exposed an uncomfortable truth: the data AI needs most is the hardest to deliver.
While organizations have spent years perfecting their structured data pipelines -- customer records, transactions, operational metrics and more -- AI models are increasingly hungry for something else entirely. They need documents, PDFs, images, videos, research archives, email threads and collaboration files. In short, they need unstructured data, and most enterprises have no efficient way to make it usable for AI.
This is the unstructured data bottleneck, and it's quietly killing AI projects across industries.
The scale problem no one talks about
The challenge isn't just volume, though that's significant. Enterprises routinely manage petabytes of unstructured content scattered across file shares, SharePoint sites, content management systems, cloud storage buckets and legacy archives.
The real problem is that current approaches to AI readiness require moving all this data first by copying it in bulk to centralized data lakehouse platforms or cloud AI services where models can access it.
At petabyte scale, this isn't a technical inconvenience. It's a months-long project with massive cost implications. Data transfer fees, storage duplication, network bandwidth constraints and the sheer time required to copy and validate data create delays that push AI initiatives from quarters into years. By the time the data arrives, business requirements have changed, models have evolved and the competitive window has closed.
The hidden costs of data centralization
Beyond time and money, the centralization approach introduces new risks. Copying sensitive documents, customer files or regulated content to new platforms expands the security perimeter and multiplies compliance obligations. Data sovereignty requirements -- keeping data within specific geographic boundaries or behind corporate firewalls -- make centralized cloud AI deployment impossible for many use cases.
Organizations are also discovering that not all unstructured data needs to move. AI projects often require access to specific document types, date ranges or content classifications, not entire repositories. But traditional ETL thinking treats data movement as all-or-nothing, leading to over-engineering and wasted resources.
Classification over migration emerges as an alternative
Forward-thinking organizations are taking a different approach. Instead of moving data first and asking questions later, they're investing in data classification, metadata enrichment and intelligent cataloging as the foundation of AI readiness. They're using automated tools to understand what unstructured data they have, where it lives, who owns it, what sensitivity levels apply and which subsets of data are relevant for specific AI use cases.
This shift from "move everything" to "understand everything, move selectively" is changing how IT and data leaders evaluate platforms and vendors. The question is no longer just, "Can your platform store my data?" It's now, "Can your platform help me understand, classify and selectively access data where it already lives?"
The platform decision point
This creates a critical decision point for enterprises. Do you rely on built-in capabilities from existing storage platforms -- cloud providers, file systems and content management systems -- to add AI-readiness features? Or do you bring in independent data classification, cataloging and governance platforms that work across your heterogeneous environment?
The answer depends on your architecture, your data distribution and your tolerance for vendor lock-in. But one thing is clear: solving the unstructured data bottleneck is no longer optional. It's the difference between AI projects that scale and AI projects that stall.
Stephen Catanzano is a senior analyst at Omdia, where he covers data management and analytics. Omdia is a division of Informa TechTarget. Its analysts have business relationships with technology vendors.
