Proving the Value of Knowledge Management

These classifications are influential when applying data access controls, which can include masking, encryption, redactions, tokenization, and outright denials of requests for content. As such, KM is instrumental for ensuring data security, enhancing data privacy measures, and sparing organizations hefty fines, churn, and loss of reputation from noncompliance to regulations such as GDPR. With this approach, “When data is accessed, the relationship between user, context, and content is much clearer to enable or restrict that data access,” Vaidyanathan mentioned. “This makes the content even more useful and compliant.” It’s noteworthy that these access controls are also applicable to what many have termed “AI governance,” which includes not only governing prompts to RAG-based systems, but applying controls to the responses of models too.

Real-Time KM

The timeliness of the access control enforcement use case—for employing KM to fortify data governance and regulatory compliance—typifies the contemporary value of KM. Top data governance solutions can use KM constructs to deny or provide access to content in real time. This is particularly germane for implementing access controls for RAG, prompts to models, and outputs from models. There’s a similar temporal application of KM when simply enlarging the context for language models to provide thorough, accurate, and relevant responses for organizations. According to Smith, “People realize AI needs knowledge to learn, to ground responses, and to be controlled. Good data, good context, and good content are vital.”

The ad hoc question-answering of current language models, their modern approach to text analytics, and their facilitation of semantic search exploit the context furnished by KM. That it does so almost instantaneously supplies further proof of the increasing recognition of the value of KM.

Moreover, there’s a dichotomy between the customer-facing and internal applicability of this dimension of KM’s value that’s beyond dispute. Take a RAG system supporting a customer help desk or customer support line. “There’s the people that are taking those phone calls, the customer service agents or employees … making sure they have access to the right content and knowledge and are able to find it quickly and easily,” Schuerman said. “And then there’s the customer use case, because, increasingly, customers don’t even want to get on the phone. They just want to be able to go and find the answer to the question on their own.” The context KM delivers for language model deployments supports both of these paradigms.

Simplifying the Skills Requirements

Another demonstration of KM’s enduring value to the enterprise pertains to its penchant for reducing the skill requirements necessary to meaningfully engage with content. Proving KM’s worth to organizations is perhaps inextricably bound to its symbiotic relationship with generative models that simplify skills requirements. Now that the foundational constructs of KM (including its vocabularies, business rules, best practices, etc.) can be accessed via natural language questions and search methods, it’s easier than ever to utilize these resources. Additionally, when these constructs effectively ground language models for such deployments, as Smith said, they make model outputs better, reducing the complexity and necessity of prompt engineering and aspects of fine-tuning. In this respect, “The timeless value of explicit knowledge still drives core outcomes for organizations,” Smith added.