Putting enterprise search in perspective at KMWorld 2023

If you’re aware of KM, or even the internet more widely, you’ve likely heard of AI, generative AI (GenAI), and machine learning (ML). The massive popularity that AI and GenAI has seen in the past year alone has greatly impacted the KM industry and its search capacity, though not entirely for the better.

The proliferation of AI and large language models (LLMs) has complicated and blurred the purpose of enterprise search, making it generally less accessible through AI’s latest air of grandiosity.

Sebastian Klatt, senior consultant, Raytion GmbH, led KMWorld 2023’s Enterprise Search and Discovery session, “Putting Enterprise Search in Perspective,” to examine the principles of enterprise search, bringing back into focus its true value and use cases.

Klatt reiterated the confusion and fear rampant around the ideas of AI, search, and LLMs, aiming to alleviate these sorts of anxieties from attendees.

He further defined search, where it makes “‘documents’ findable by searching its content and metadata.” It generates a “lexical” index by using linguistic methods to provide a robust result. In most cases, it provides links to the documents that contain the query terms.

Narrowing the scope, enterprise search incorporates these qualities: security, scalability, retrievability, and the indexing of all relevant content sources.

Klatt clarified the use cases for search, separating it into three different areas:

• Question answering, i.e., “What’s on the cantina’s menu tomorrow?’
• Document research, i.e., “I need the presentation from college x about topic y.”
• Topic research, i.e., “What information do we have about product a or topic b?”

The introduction of vector search also necessitates some clarification. Klatt explained that vector search converts words into numbers (high dimensional vectors) to then be ingested by machine learning algorithms. Vector search, however, isn't new, nor is it a replacement for lexical search (though it can support it).

“Vector search is really good as a foundation for large language models by nature,” said Klatt. “If someone recommends vector search, they should prove to you that it’s better than enterprise search.”

Klatt then transitioned the discussion toward GenAI, providing insight into what search can do for GenAI.

“When you think of a generative AI solution in an enterprise, you basically have a pre-trained model (in most of the cases) ... the main problem, [though,] is that it is not trained on enterprise data,” said Klatt. “An orchestrator can combine the pre-trained model with enterprise search to provide context for the LLM.”

As a result, enterprises provide organization-specific content to the GenAI without the need to train the model. Ultimately, this will reduce hallucinations since the LLM is then based on enterprise documents.

LLMs are capable of injecting context into enterprise search—an extremely beneficial proposition for any modern organization, according to Klatt. It can enable:

• Immediate answers if the query is a question
• Synonym generation based on the given query for better result sets
• Result summarizations that give a management abstract of the top results for a query
• Intent recognition for understanding the intent of a search
• Expanded search by extracting the key phrases out of a result set/single result
• Metadata generation during indexation for relevancy optimization

On the bright side, “normally GenAI initiatives are ranked very high [in enterprise priority], and normally, the higher you are in the food chain, the higher budget you have,” said Klatt.

“It’s really about positioning yourself as the content supplier for such initiatives,” he added.

Klatt concluded that “neither vector search nor generative AI will replace enterprise search. However, all can support each other.”

KMWorld returned to the J.W. Marriott in Washington D.C. on November 6-9, with pre-conference workshops held on November 6.

KMWorld 2023 is a part of a unique program of five co-located conferences, which also includes Enterprise Search & Discovery, Enterprise AI World, Taxonomy Boot Camp, and Text Analytics Forum.