Why Classification Systems Are Broken

TL;DR: Classification systems were built independently by different organizations, in different decades, for different purposes. The result is global fragmentation that costs businesses millions in manual mapping, compliance risk, and missed insights.

The fragmentation

Your US supplier uses NAICS. Your EU partner reports in NACE. Your customs broker wants HS codes. Your HR team maps jobs to SOC domestically and ISCO internationally.

None of these systems were designed to interoperate.

graph TB
  subgraph US["United States"]
    NAICS["NAICS 2022\n2,125 codes"]
    SOC["SOC 2018\n1,447 codes"]
    HTS["HTS\nUS tariff"]
  end
  subgraph EU["European Union"]
    NACE["NACE Rev 2\n996 codes"]
    ESCO_EU["ESCO\n3,045 occupations"]
    CN["CN 2024\nEU tariff"]
  end
  subgraph Global["Global / UN"]
    ISIC["ISIC Rev 4\n766 codes"]
    ISCO["ISCO-08\n619 codes"]
    HS["HS 2022\n6,960 codes"]
  end
  NAICS -.->|"partial"| ISIC
  NACE -.->|"1:1"| ISIC
  SOC -.->|"lossy"| ISCO
  HTS -.->|"extends"| HS
  CN -.->|"extends"| HS

The dotted lines represent crosswalk mappings. Some are exact. Most are lossy, partial, or many-to-many. And many pairs have no official mapping at all.

How we got here

Each system was built for a specific purpose. NAICS exists for the US/Canada/Mexico census. NACE exists for EU statistical reporting. ISIC exists as a UN global reference. Conceptual overlap, structural divergence.

Systems evolve on different timelines.

System	Last major update	Update cycle
NAICS	2022	Every 5 years
NACE	2008 (Rev 2)	~15 years
SIC	1987	Frozen (still used)
ICD-10-CM	Annual updates	Yearly
HS	2022	Every 5 years

Granularity varies wildly. ISIC has 766 codes. NAICS has 2,125. ICD-10-CM has 97,606. Mapping between different granularity levels is inherently lossy.

No central authority coordinates mappings. The UN publishes some concordances. Eurostat publishes others. But these are static PDFs and spreadsheets - not queryable APIs.

The real cost

Every multinational that consolidates reporting across regions employs people whose primary job is maintaining crosswalk tables by hand.

Data engineering time. Hand-curated crosswalk tables that break whenever either system updates.

Compliance risk. Wrong industry code on a regulatory filing triggers audits. 20 countries = 20 different systems = enormous error surface.

Missed insights. If US and EU teams use different systems with no mapping, you cannot answer: "What is our total transportation revenue globally?"

AI blind spots. LLMs know classification systems exist but hallucinate specific codes. They lack structured access to hierarchies and crosswalk edges.

The fix is a graph, not a standard

The world does not need another standard. It needs a connector.

graph LR
  A["System A\n(tree of codes)"] -->|crosswalk edges| G["Knowledge Graph"]
  B["System B\n(tree of codes)"] -->|crosswalk edges| G
  C["System C\n(tree of codes)"] -->|crosswalk edges| G
  G -->|translate| A
  G -->|translate| B
  G -->|translate| C

Every classification system is a tree. Crosswalks are edges between trees. Collect enough trees and edges, and any code can be translated to any other system by traversing the graph.

That is WorldOfTaxonomy: 1,000 trees connected by 321,000 edges.

The systems will continue to exist independently. We connect them anyway.