Gaining competitive advantage from non-textual information
Harmonizing textual and non-textual content
Historically, the most arduous part of gaining competitive advantage from non-textual content is the prerequisite for standardizing it alongside textual content. This vital undertaking often precedes, and can inhibit, describing and making this content searchable. In some instances, the type and format of the non-textual data determines successful searches.
Some platforms allow users to “import a spreadsheet, like a raw CSV file or Excel file, directly and write the code to extract the information from the underlying data format,” Brett Wujek, SAS head of project strategy for next-generation artificial intelligence, observed. “SAS did that so you can directly drag your spreadsheet in and start working on it.” For images, technologies such as OCR have advanced to the point in which “[y] ou can upload your W-2 image and it will pull the information out of that and put it into your online tax software directly,” Wujek commented.
In other cases, the the non-textual data’s format is irrelevant and can be readily ingested, transformed, and harmonized alongside any other enterprise knowledge an organization has. Knowledge graphs with semantic standards are particularly efficacious in this regard and are applicable for “pulling data in from textual sources,” Martin mentioned. “But it wouldn’t matter if those sources were images, video, or audio. The pipelines would work the same way. The annotators make sense of what’s in the unstructured, complex data and turn it into a graph ontology, growing that ontology as necessary. Then, that data ends up in graph form for analytics.”
Audio data is gaining credence as a way of monitoring customer service representatives, analyzing and making search available for internal meetings, and other use cases. Audio data often involves speech- to- text capabilities, so organizations can analyze the data via “techniques for understanding sentiment or extracting key topics and categorizations of text,” Wujek denoted. These capabilities are becoming increasingly democratized. “Speech-to-text or video-to-text transcriptions—we see it in video remote meeting tools like Teams or Zoom,” Nivala remarked.
Self-service non-textual data
Perhaps the most meaningful development impacting the use of non-textual information for competitive advantage is the marked reduction in technological faculty required to access and implement these sources. Martin reflected that, historically, the annotators he described for transforming any type of unstructured content into that consumable for knowledge graphs have entailed “relatively expensive and specialized software. Now, it’s getting to where anyone can write a prompt and say, ‘Tell me what’s in this stream of pictures and here’s the form I want it in.’”
Martin’s quote refers to projections, including those he said he’d read, for large language models like GPT-5 and Llama 3 to facilitate rich, in-depth descriptions of non-textual, visual content. What is of paramount importance—and what drastically decreases the technological savvy for incorporating such non-textual data into enterprise knowledge—is that the same conversational prompt can also spur the data engineering for transforming the data into a recognizable format for the consuming application. This previously specialized, difficult task has the potential to instantly be done by anyone, causing “the cost of the annotators to source the data from unstructured data to drop through the floor,” Martin predicted.