If you've spent your career in taxonomy management and knowledge management, you've probably heard some version of this by now: AI is coming for the classification work. The tagging, the extraction, the vocabulary reconciliation – it's all automatable. The role is shrinking.
It isn't. But it is changing. And if you've been doing this work for a while, the change is probably long overdue.
What the Job Has Actually Cost You
In many large organizations, there’s a growing backlog of content waiting to be classified and made usable. So with that in mind, take a moment to honestly assess just how much of your working week you've been spending on work that machines could have been doing all along.
Manual extraction. Gap-hunting. Vocabulary reconciliation across systems that were never designed to talk to each other. None of it is unimportant. All of it is relentless. And there's so much of it that the work requiring your actual expertise, from deciding what gets modeled and why, to governing the vocabulary an entire organization depends on, to connecting taxonomy decisions to outcomes leadership cares about, that all of that gets squeezed into whatever time is left.
At a recent Squirro webinar on automated content classification, veteran taxonomy practitioner Melinda Geist, who spent decades building knowledge management programs inside global technology companies, put it plainly: the work that consumes most of a taxonomist's time is work that machines should have been doing all along. The strategic thinking, she said, rarely gets the attention it deserves. Not because it isn't valued. Because there isn't enough room for it.
That's about to change.
Why the Extraction Problem Never Stayed Solved
Taxonomy drift is the dirty secret of enterprise knowledge management. And with growing awareness of the reliance of GenAI on taxonomies and ontologies, it is beginning to get the attention that it deserves. Every domain vocabulary – technology, finance, manufacturing – shifts constantly. New concepts emerge, terminology evolves, and the language your subject matter experts actually use in content quietly moves away from the controlled vocabulary built to describe it. It doesn't announce itself. It just widens, silently, until search stops working and nobody can explain why.
The only way to catch vocabulary drift manually is to read the content. A lot of it. Regularly. At enterprise scale, with hundreds of contributors across dozens of systems, that becomes a treadmill: you can run hard and still fall behind.
This is where LLMs change the game. Not because they replace taxonomist judgment – they don't – but because they can read at a scale no human team can match; surface the concepts living in a corpus; flag where the vocabulary has drifted; identify what's missing. And they can do it all in minutes, not months. Melinda described seeing this in a four-week trial of the Squirro classifier and thinking: this is it. After years of looking, the first tool that actually solved the problem rather than worked around it.
The Machine Surfaces. You Govern.
What automated document classification doesn't do is decide. It doesn't evaluate whether a candidate concept is the right level of granularity, whether it duplicates something already in the taxonomy under a different label, or whether it belongs at all. It doesn't know your taxonomy governance framework, your strategic priorities, or the downstream systems depending on the vocabulary being right.
That's your job. And in a well-structured human-in-the-loop workflow – which is how the Squirro Classifier is designed to operate – it stays your job. The LLM proposes: here are the concepts in this corpus, here are the gaps, here are context-grounded definitions for what's missing. You review, evaluate, and decide. What gets added, how it's defined, where it sits – all of it remains a human call.
The ratio just changes. Less time on taxonomy maintenance. More time on taxonomy governance. The highlighter goes away. The judgment work, which was always the hard part, and always the part that needed you, finally gets room to breathe.
And Then the Leadership Conversation Changes
Here's what happens when enterprise taxonomy management stops looking like an endless maintenance backlog and starts looking like a governed, measurable, strategic asset.
For most practitioners, the taxonomist role has been treated as invisible infrastructure – necessary, chronically underfunded, rarely understood by the people holding the budget. That's partly a perception problem, partly also a language problem. It's hard to make a strategic case for a function when most of what you do is invisible, and the value only becomes visible when something breaks.
Automated content classification changes that calculus. When you can show leadership that taxonomy drift cost the organization years of effort and millions of dollars on a single content integration and that the right tooling would have cut that to months, the room shifts. In the webinar, Melinda walked through exactly that kind of retrospective: a post-acquisition content integration, three years of work, two attempts, marginal results. Her estimate of what the Squirro classifier would have saved, had it been available at the time, is a number that is frankly too big to be overlooked.
The taxonomist who can govern that foundation and quantify its value across the full taxonomy lifecycle isn't invisible infrastructure. They're a strategic function. That's always been true. Now there's a way to prove it.
Hear It from Someone Who's Lived It
The webinar goes deeper than this article can. Melinda covers the full arc – the years of manual extraction, the failed attempts at scale, what the trial revealed, and what the role looks like once the extraction problem is off your plate. Panos Mitsias, Squirro's Semantic Graph Solution Specialist, covers the architecture: how the classifier connects your existing enterprise taxonomy to your unstructured content, and how the human-in-the-loop workflow keeps you in the driver's seat throughout.
If this is a shift you're navigating – or a case you need to make internally – the recording is worth an hour of your time. Register here to attend the on-demand webinar.
