Why SIC codes are not what you think

A SIC code is a company telling you what it does. It is not anybody verifying what it does. That distinction is the single most common way analysis built on the UK company register goes wrong.

How a company gets its code

At incorporation, someone fills in a form and picks up to four SIC codes from a list of roughly 730. Often that someone is a formation agent, or an accountant, or the founder skimming a dropdown at ten at night.

Nobody checks the choice. There is no verification step, no audit, no correction process. And while a company can update its codes on its confirmation statement, most never do, because nothing forces them to and nothing happens if they do not.

So a code records what somebody believed, once, at the moment the company was created.

What that means in practice

Your filter will miss companies that belong in it. A property developer might sit under 41100 development of building projects, or 68100 buying and selling of own real estate, or 70229 management consultancy because that is what the accountant put. All three are defensible. Only one shows up in your query.

Your filter will include companies that have moved on. A business that incorporated in 2015 as a web design agency and now sells physical products is still coded 62012. The register has no idea.

The generic codes are enormous. 82990 other business support service activities and 70229 management consultancy activities absorb a large number of companies that could not find a better fit. So does 64209 activities of other holding companies. Any of those appearing near the top of a count usually means "unclassified" rather than a real sector.

The rule of thumb

SIC is strong in aggregate and weak per company.

"How many companies incorporated in construction last year" is a question SIC answers well. The misclassification exists but it is roughly consistent year to year, so the trend is real even if the level is approximate.

"Is this specific company a builder" is a question SIC answers badly. For that you need to look at the company, not its code.

How to work with it anyway

• Group at division level, the first two digits, rather than the full five. It absorbs a lot of the noise between adjacent codes.
• Use several codes per sector, not one. Real estate is not 68209, it is all of 68 and arguably some of 41.
• Treat the generic codes as unknown, and say so, rather than counting them as a sector.
• Combine with something else when you need per-company accuracy. Company name is a surprisingly good second signal.
• Watch what people picked, not just what they are. A company that chose 62012 was thinking of itself as a software business on the day it formed. For some purposes that intent is the useful bit.

The honest summary

SIC codes are the best free classification of UK companies that exists, and they are wrong often enough that you should never present a count from them without saying where it came from.

Use them. Just do not trust any single row.