It’s Time to Bury the Document

Knowledge Management as we know it is not just being disrupted by vectorization, vector search and LLMs, but they may be the final nails in its coffin. I’ve previously gone into detail about how and why this is the case, so I’d like to look towards the future in which the document is no longer the basic building block of KM systems, but instead the “chunk” (as beautiful a name as it is). 

The document is such a foundational piece of both the analog and digital world that it goes completely unquestioned. Yet, the digital incarnation of the document simply transferred an analog concept and technology into a new form, without considering the differences in capability, format and delivery and the new possibilities stemming from those differences. 

Why Documents Suck

When we learned to write essays in school, we were taught to structure them clearly into sections, arguments and evidence. The idea that knowledge has to be organized and that it relates to other pieces of information is clear. But we’re still thinking about that in terms of a piece of paper. 

Consider how we consume information today, in tweets, FAQs, snippets and short posts. 

What’s the Purpose of a Document?

An essay or book is a coherent long-form piece of information meant to be read in its entirety from beginning to end. Customer support information such as a return policy, product technical specifications or compatibility chart is not. It’s meant to be queried against with a specific question that has a clear and usually short answer. 

Usability Limitations

Like mold in a humid environment, documents quickly grow and they both usually stink. With the ability to continuously write more, the length, readability and accuracy of digital documents suffer. When lengthy, they are difficult to navigate particularly in the context of finding the answer to a specific question. Yes, we have CTRL-F but what happens when you search with different words than are used in the document, despite it being the same idea? I can’t be the only one who has nightmares of trying to find a critical answer in a massive PDF! 

Documents are simply an inefficient means of storing data that should be actionable and is not meant to be consumed linearly or in its entirety. 

FAQs were one step towards this as are HTML formatted documents which add improved navigation options.

Consider your last Google search. Google moved from returning full documents (or web pages) to simply snippets which directly answered your question. Raise your hand if you are hoping to find a massive wall of text when you search for something. I won’t even wait for you to react because literally nobody wants that, especially not your customers and agents. 

Goodbye Documents, Hello Snippets & Chunks

So, documents suck, we’re supposed to use these so-called chunks. What are they exactly? A chunk is a unit of knowledge extracted from a knowledge source like a Word document, PDF, PowerPoint or similar source. Chunks are small, self-contained pieces of information can be processed and managed effectively. They’re designed to be consumed in small portions, perhaps along with another chunk or two, but not as an entire “document”. Think of documents like a massive main course and chunks like tapas. 

For instance, a chunk can represent a single paragraph, a sentence, or even a smaller unit of text from a document. By dividing the content into chunks, you gain better granularity, enabling that knowledge to be analyzed more easily and quickly and thus it can be identified and delivered to answer customer queries more efficiently. The same however goes for agent assistance scenarios where a human agent needs instant answers instead of “Please hold (while I switch to another tab and depressingly scroll through a Great Wall of text)”.

From a technical perspective, managing knowledge in chunks improves the system's ability to match the right information with customer questions, resulting in more accurate and contextually appropriate responses. Chunks can have associated metadata such as a title, topic tag, source and more. 

A document is to a chunk as a cow is to a steak. Which one do you want to be served at a restaurant?