Gaining competitive advantage from non-textual information

The most daunting task for contemporary knowledge managers, knowledge workers, and organizations as a whole is the management of complex, non-textual, unstructured content.

Estimates that up to 80% of data within an organization is unstructured have long existed. With advancements in generative foundation models and AI, this percentage is surely increasing. Enterprise knowledge is no longer neatly confined to textual documents, taxonomies, and business glossaries. Instead, it seems to be randomly dispersed throughout any assortment of videos, images, webpages, data visualizations, spreadsheets, and other sources that may involve text, but ultimately express knowledge in non-textual forms.

The objective of enterprises is to obtain this knowledge, store it, and render it discoverable for any number of vertical- and application-specific processes. The last of these concerns, discoverability, is perhaps the essence of KM—although it hinges upon the success of the other two.

“Non-textual information often requires a specific skill set to change an image, video, or email design,” acknowledged Andrea Carrillo, director of brand and creative at marketing technology company Acoustic. “Having a tool that allows everyone to change this part of the email template, or restructure something a little bit, allows for larger enterprises to make those edits more easily and more scalable.”

With the right approaches, tools, and self-service facilities, it is possible for users possessing any degree of technical aptitude to quickly find and avail themselves of non-textual content. More significantly, they can also align it with knowledge from traditional textual forms.

Subsequently, non-textual information becomes as accessible and viable to KM processes as textual information is, once organizations “turn complex and unstructured data into a structure, so it can be married with everything else,” revealed Sean Martin, Altair’s VP of software engineering.

Information models

Underscoring the closeness of the relationship between textual and non-textual information is the fact that the enterprise applicability of the latter is based on a textual information model. Encompassing various facets of ontologies, taxonomies, business glossaries, and schema, this information model provides the concepts and definitions of what the content means. This fact is true for textual and non-textual content. Organizations can establish an information model by assembling their own definitions and taxonomies, implementing those in KM solutions, or generating this necessity with language models.

“We store a fairly robust taxonomy; we also have a business glossary,” revealed Vishal Kirpalani, VP of design at Acoustic. Such vendor resources are often tailored for specific use cases, such as marketing campaigns or customer interactions. Regardless of how they are devised, these semantics are integral for tagging non-textual content to heighten its discoverability. These are some of the most influential techniques for classifying content in this way:

• Prefabricated tags. Some KM systems supply tags in which “the pre-populated tags come from the information model,” Kirpalani explained. With this approach, the tags are already created, and the user applies them.

• Auto-tagging. In other systems, these metadata descriptions are automatically rendered by machine learning “for simple things like generating keywords, or a little bit more complex things like, ‘What is this document?’” mentioned Antti Nivala, M-Files CEO. “Is this a contract, a meeting memo, an audio or a video recording? Which organizations are related to this?”

• Lengthier descriptions. Some tools are endowed with capabilities for delivering lengthier descriptions about content. These features are particularly effective for static images, infographics, dynamic data visualizations, and videos. According to Kirpalani, these “larger descriptions allow us to actually talk about what that graphic actually signifies. For example, if I say we are at 85% campaign effectiveness, what does that actually mean? How is that derived? What came out of it?”