What company formation data can — and can't — tell you about the UK economy

Company formation data is one of the few economic signals that is exact, free and published daily. It reliably tells you where new commercial activity is being set up — by sector, by place and over time. It does not tell you how many businesses are trading, how much they earn or how many people they employ. Most bad readings of the data come from treating the first kind of fact as the second.

This guide sets out what the Companies House register can support, using figures from our own research editions, and the four traps that catch most people who quote it.

What the register measures

Every UK limited company, LLP and PLC is created by registration at Companies House. The register records the date, the registered office, up to four SIC codes describing the intended activity, and later the company's status — including when it is dissolved. Because registration is a legal requirement, the count is a census, not a survey: nobody is sampled, and the figure for a period is exact.

Our H1 2026 edition measured 400,176 incorporations in the first half of 2026, down 5.9% on the 425,088 in the first half of 2025. That is a precise statement about company registrations. Whether it is a statement about the economy is the question this article is about.

What formation data is good for

1. Sector momentum

Changes in what people register are a fast read on where founders think the opportunity is. In H1 2026 the clearest movement was in software:

• Software development (SIC 62012): 20,765 registrations, up 71.4% year on year.
• Data processing and hosting (SIC 63110): 3,810, up 56.4%.

Over the same period buying and selling own real estate (SIC 68100) fell 14.0% and online retail (47910) fell 7.1%, although online retail remained the single most-registered activity. Those are large, simultaneous movements in opposite directions — the kind of shift that shows up in the register months before it appears in output statistics.

What the register cannot tell you is why. A surge in software registrations is consistent with a wave of AI-related start-ups, with contractors re-describing their activity, or with formation agents changing the default codes they file. The register alone doesn't separate those. What it does tell you, unambiguously, is that a lot more companies are now declaring software as their business — which is exactly what a supplier selling to software companies needs to know.

2. New demand

Every new company is a buyer before it is anything else. In its first months it opens a bank account, appoints an accountant, buys insurance, registers a domain and often builds a website. Formation counts are therefore a direct measure of fresh demand for those services — by sector and by place. This is the most practical use of the data and the least controversial, because it doesn't depend on any of the companies succeeding.

3. Where activity is being registered

Formation counts by place show where companies choose to register, and how that is changing. London took 27.9% of UK registrations in Q2 2026, up from 24.2% in Q2 2025 (counts re-run on 8 October 2026). That rise is real in the register — but read the next section before treating it as a fact about London's economy. We look at it in detail in London's growing share of new UK companies.

Four ways formation data misleads

A registration is not a trading business

A company can be registered and never trade. Property-holding vehicles, holding companies, dormant companies kept to protect a name, and companies formed for a single contract all count as incorporations. Some SIC codes — holding companies (64209) and property letting (68209) especially — reflect structuring as much as new economic activity. Formation counts measure intent and legal set-up, not activity.

A registered office is not where the business operates

The location on the register is the registered office: a legal address for correspondence. Many companies use their accountant's or a formation agent's address, which concentrates registrations in city centres and in London above all. A city's formation count is where companies are registered, which is only a proxy for where they work.

SIC codes overlap and are self-declared

A company can list up to four SIC codes, so code counts double-count and shares don't sum to 100%. The codes are chosen by the person filing, aren't verified, and are rarely updated as a business changes. Use them for direction and scale, not as an industry census.

Dissolution is not failure

When companies leave the register, the cause is usually voluntary strike-off by their own directors, not insolvency. Our company survival study found that of companies formed in 2016, 71.6% have since been dissolved — but fewer than 1% of that cohort is currently in liquidation. Treating dissolution counts as business failure rates overstates failure by a very wide margin.

Register artefacts to watch for

The register also contains administrative events that look like economic ones:

• The Companies House default address. When a registered office is disputed or removed, Companies House moves the company to its own default address, which is a Cardiff postcode. In Q2 2025 this put 8,196 companies into Cardiff's raw count. Uncorrected, Cardiff's formation figures show a collapse that never happened. Our city edition measures and removes them.
• The 2020 strike-off pause. Companies House paused striking companies off the register during the pandemic. Companies formed in 2019 therefore show an unusually low first-year dissolution rate (9.0%, against 16–26% for every other cohort we measured) — a timing effect, not a healthier generation of companies.
• Late registrations. Counts for a recent period creep up for a few weeks as late filings are processed. Our H1 2026 total read 400,150 in early September and 400,176 a month later. We hold every period back three weeks before reporting it for this reason.

How to use the data well

1. Compare like with like. Same period length, same filter, same retrieval method. Year-on-year comparisons of the same quarter remove most seasonality.
2. Use exact counts, not samples. Companies House advanced search returns an exact hit count for a filter. A figure from a sampled list is an estimate and should be labelled as one.
3. Separate the signal you need from the one you don't. For finding new customers, a registration is the event that matters, whether or not the company succeeds. For judging economic health, combine formation data with survival, employment and output data.
4. State your caveats next to your number. The figure travels further than the footnote.

From pattern to prospects

The point of reading formation data is usually to act on it. If software development registrations are up 71.4%, the useful next step is the list of companies behind that number. You can open software development companies formed in the last 12 months on the live register, or browse every SIC code we track. Free new-company alerts send new registrations in a sector or place each week.

Figures are exact Companies House counts from CompaniesIQ research editions and queries re-run on the dates stated, under the Open Government Licence v3.0. See our data sources.