Verified conversions, never guesses.
Converting a code between classification systems is easy to do badly — many sites match by number or by fuzzy text and silently mislabel businesses. Every mapping on IndustryCodes carries an evidence source (hover any exact/approximate/inferred badge to see it) and is labeled by its weakest hop: verified correspondences rank first, then verified chains, then clearly-labeled inferred matches through a standardized ISIC bucket. Only verified evidence can ever be called exact — and when even an inferred match would be dishonest, we say so instead of guessing.
Official correspondence tables
UN Statistics Division (ISIC Rev 4 ↔ Rev 5), US Census Bureau (ISIC ↔ NAICS 2022) and the Australian Bureau of Statistics (ANZSIC ↔ ISIC) publish authoritative mappings. We compose them conservatively: a result is only "exact" when every hop is exact.
Exact code identity
NAICS is maintained trilaterally: where the US, Canadian and Mexican editions share an identical code, the match is exact by construction. UK SIC 2026 is built directly on NACE Rev 2.1 — identical codes with identical titles are exact. ATECO 2025 (Italy), NAF 2025 (France) and WZ 2025 (Germany) adopt NACE Rev 2.1 wholesale at their first four levels — we verified the code sets are identical level by level, so those mappings are exact by construction too.
Verified title identity
Systems that renumber ISIC (UK SIC, NACE, NIC, SSIC, KSIC) are never matched by code alone — that produces wrong answers (UK SIC 01.42 is cattle; ISIC 0142 is horses). We only link classes whose titles are identical after normalizing boilerplate ("other than" ≡ "except"). KSIC Rev 11 and HSIC V2.0 are built on ISIC Revision 4, so their title-verified classes are chained onward through the official UN ISIC 4→5 correspondence, with the weakest hop deciding the label.
Derived evidence
Broader mappings (a whole section or division) are derived from the class-level evidence beneath them, weighted by how many verified class pairs support each candidate — and labeled exact only when the evidence is ≥90% mutually exclusive in both directions (e.g. NAICS 31-33 ↔ ISIC C).
Inferred matches (the standardized bucket)
When no verified path exists between two systems, we fall back to a standardized bucket: every code in every system is assigned to the international ISIC class that best contains it — from its own verified mappings where they exist, from its parent otherwise, and by careful semantic assignment for the remainder (each assignment carries its provenance and a confidence). Codes sharing a bucket are offered as candidates, ranked by how closely their official descriptions match. These results wear a distinct "inferred" badge, are never presented as exact, and always show the shared ISIC bucket so you can judge the evidence yourself.
Ranking within the evidence
When one code fans out to several candidates with equal evidence (a NAICS industry landing on a whole family of NIC subclasses, say), we order them by how closely their official titles match — including cross-dialect vocabulary (a US "truck" finds an Indian "lorry", "soybean" finds "soyabean") — and rank residual "other / n.e.c." catch-alls below specific classes. A broader parent class that merely contains the specific matches is labeled "broader" and listed after them, never presented as a peer answer.