Why fragmented, unstructured data is limiting AI's value
When you purchase through links on our site, we may earn an affiliate commission. Here’s how it works.
Commerce in the information age has thus far been one long quest to gather as much information as possible. Customer data, sales data, inventory data, system data, operational data, technical data, every facet of business can and has been tracked and quantified, guided by the idea that knowing more means decisions can be made faster and more conclusively.
Business Development Manager at PropertyShark.
While sound in theory, that process has clearly hit a snag, as more and more companies abandon their AI projects after finding that the exponential explosion of data creation isn’t generating comparable value.
Has AI peaked in its utility? Or is the issue one of interoperability? Why exactly has this monumental trend hit such a stumbling block?
It’s entirely understandable to assume that feeding more information into an AI increases the accuracy and quality of the output. Models need to train on data; after all, that is how they establish reason.
But, like any computer system, input must be structured for it to be understood, and the issue many organizations now find themselves in has come as a result of them giving AI a decade's worth of unformatted, incomplete, non-standardized data and expecting it to read between the lines.
Every system a company uses, from CRMs and internal emails to performance tracking spreadsheets and PDFs, stores different data points and operates on vastly different logic. What they host, why they host it, and how they host it were all purposeful decisions based on their desired function.
While some support import and export options that translate material from one format into another, the developers behind these solutions never envisioned the need for a universal logic to tie everything together in a way that AI can easily parse.
Sign up to the TechRadar Pro newsletter to get all the top news, opinion, features and guidance your business needs to succeed!
Each document and dataset contains a fragment of the overall picture. While AI excels at processing data, it cannot derive meaning and struggles to understand and incorporate unstructured data into its output, regardless of how many times it is asked to.
We have largely solved the problem of data accessibility to the point that many public AI models have run out of new data and are now cannibalizing the output of other agents. The next step is in refining how AIs use the data they have.
Real estate is an excellent showcase of this process in motion. To appraise and list a property, agents need access to ownership records, zoning information, environmental information, and neighborhood demographics, just to name a few. None of these systems was designed to communicate or cooperate, which has meant AI has historically struggled to find its footing in the industry.
Newer approaches, however, prioritize the ability to find, interpret, and synthesize information from across different sources, aiding in the manual searches an analyst would otherwise do by hand.
