You have probably lived through this moment. Two of your systems need to share data, and someone discovers that what System A calls “Facilities Maintenance”, System B calls “Building Services”. What your ERP calls a Tier 2 support ticket, your help desk software calls a “Standard Incident”. The information is the same. The labels are not. And suddenly a project that was supposed to take six weeks turns into a six-month argument about definitions that nobody budgeted for.
This is the taxonomy problem. A taxonomy is just a system of categories; a way of sorting things into buckets. Your chart of accounts is a taxonomy. Your product categories are a taxonomy. Your HR job classifications, your IT service tiers, your customer segments; all taxonomies. And the headache is that every software vendor, every industry group, and every company has its own.
For decades, the only solution was to force everyone onto the same system of categories. Pick an industry standard. Make all your vendors adopt it. Spend months building cross-reference tables that translate Category A in one system to Category B in another. It was slow, expensive, and brittle; but it was the only game in town.
That game is over. AI has quietly made the entire problem optional.
What Changed
In the old world, translating between two different category systems meant building a giant lookup table by hand. Your team would sit down, look at every category in System A, and decide which category in System B it matched. This is what IT people call a “crosswalk”; a line-by-line mapping from one set of labels to another.
These lookup tables were painful to build and even more painful to maintain. They broke constantly, because real data does not fit into neat boxes. What happens when System A has five categories for something that System B handles with three? What about the items that could go in two categories depending on context? Every edge case required a human judgment call, and edge cases were the norm, not the exception.
AI does something fundamentally different. It does not just match labels. It understands what the labels mean. When an AI sees “Facilities Maintenance” in one system and “Building Services” in another, it recognizes that both refer to the same real-world activity; keeping the building running. It gets this right even when the two systems organize their categories in completely different ways, with different levels of detail and different grouping logic.
This is not a small improvement over the old lookup table. It is a different kind of capability entirely. And it is available today, not in some future product roadmap.
Why This Matters More Than It Sounds
The deeper impact is that you no longer have to decide how to categorize your data before you store it. Think about what that means.
Today, when you enter data into most systems, you have to pick a category up front. Is this expense “IT” or “Operations”? Is this customer “Enterprise” or “Mid-Market”? Once you pick, that choice follows the data everywhere. If you need to see things organized differently six months later; say, for a board presentation or a regulatory report; someone has to go back and re-sort everything. That is a project.
With AI, you can store detailed, plain-language descriptions and then sort them into whatever categories you need at the moment you need them. Need your IT spending organized one way for the auditors and a different way for the strategy offsite? Same data, two different category structures, applied on demand. No migration required.
For the CEO who just acquired a competitor and is staring at two completely incompatible sets of product categories, customer segments, and cost centers, this is transformative. Under the old model, someone spends months arguing about whose categories are “right” and manually re-sorting years of historical data. Under the new model, both systems keep their categories. The AI translates between them on demand. You get consolidated reporting in weeks instead of quarters.
Where This Breaks Down
Now for the honest part, because this capability does have real limits.
It gets less reliable at very large volumes. AI-driven translation works well for hundreds or thousands of records. When you scale to millions of records with lots of edge cases, errors start to add up. One wrong classification in your supply chain data can cascade into a purchasing mistake, a compliance problem, or a financial reporting error. The accuracy that is fine for a strategy discussion is not fine for a regulatory filing.
Regulators want to see the math. In heavily regulated industries like healthcare and finance, it is not enough for your data to be correctly categorized. You need to prove how it was categorized, show that the process is repeatable, and point to a responsible human when something goes wrong. “The AI figured it out” is not an answer that satisfies an auditor. You can solve this by using AI to build the translation tables and then having humans review and certify them; but that is an important constraint.
Shared definitions keep people honest. This is the subtle risk that is easy to miss. Categories do not just organize information; they force people to be precise. When an entire industry agrees that there are exactly seven types of cybersecurity risk, that shared language means everyone is actually talking about the same things. Remove the shared categories and you get richer descriptions, but you also get drift. Two companies using AI to translate freely between their own private category systems might think they understand each other perfectly, while their definitions have quietly moved apart. The AI makes the translation look smooth. The real understanding underneath might not be.
Standards Aren’t Dead · They Just Changed Jobs
None of this means you should throw out industry standards. It means you should understand what standards are actually for.
Standards still matter when they create shared accountability. When a hospital codes a procedure, that code is not just a label; it triggers payment, feeds into public health research, and creates a legal record. The value is not that the coding system is elegant. The value is that every hospital, insurance company, and regulator has agreed to play by the same rules. AI cannot replace that agreement.
Standards still matter when you need to compare data across many organizations over long periods of time. National economic data, disease tracking, environmental reporting, financial disclosures; these depend on thousands of organizations categorizing things the same way for decades. Even small inconsistencies, multiplied across that scale, create serious problems.
But standards are no longer necessary just to get two systems to exchange data. That is the breakthrough. The rigid formatting rules that dictated exactly how a purchase order or insurance claim had to be structured for electronic transmission; those are the standards that AI makes obsolete. When an AI can read a purchase order in any reasonable format and understand what it says, the value of forcing everyone into the same rigid template drops dramatically.
Think of it this way: standards for meaning are still essential. Standards for formatting are becoming optional. The first kind creates shared accountability. The second kind was always just a workaround for the fact that computers were too dumb to figure out what the data meant without rigid instructions.
What You Should Do About This
If you have a system integration, a post-acquisition consolidation, or a vendor migration that has stalled because the data categories do not line up; stop waiting for perfect alignment. Use AI to bridge the gap now. Validate the results at the scale that matters for your business. Save your standardization energy for the areas where shared definitions create genuine accountability.
Do not be reckless about it. Build checks. Have your team spot-test the AI’s work. Keep human oversight on high-stakes classifications. But stop letting a labeling disagreement hold up a project that your business needs to move forward.
The 90% of translation work that was always just tedious, mechanical matching can now happen in hours. That frees your people to focus on the 10% that actually requires expertise and judgment.
The companies that figure this out will integrate faster, spend less on data plumbing, and make better decisions; because they will actually use their data instead of endlessly preparing it. The companies that keep waiting for the whole industry to agree on categories will keep waiting for a very long time.
Taxonomy freedom is not a future capability. It is here now. The question is how much longer you are willing to let a labeling disagreement hold your business hostage.
