Britain Has an Industrial Strategy for Industries It Cannot Measure

Britain has decided what its economic future should look like.

Artificial intelligence. Quantum technologies. Semiconductors. Advanced manufacturing. Clean energy. Life sciences. Defence.

The Government’s industrial strategy is built around eight sectors that it believes can generate a disproportionate share of future growth.

There is an awkward problem. Britain cannot properly measure several of them.

The Government says this itself. Its methodology for monitoring the industrial strategy says the eight priority sectors cannot be precisely mapped using existing official statistics. Clean energy is not captured at all in some of its principal measures of investment, output, employment and productivity. Defence and life sciences are only partly covered.

Artificial intelligence provides the clearest example.

On September 21, the Office for National Statistics published a methodology explaining how it intends to measure AI’s contribution to the British economy. It acknowledged that existing statistical frameworks do not provide enough detail or visibility to isolate the effect of AI reliably.

An economy inside old boxes

Every modern economy has to divide activity into categories. A restaurant belongs to hospitality. A car factory belongs to manufacturing. A bank belongs to financial services.

Britain does this principally through the Standard Industrial Classification system. Businesses are assigned according to their economic activity so statisticians can estimate employment, investment, output and productivity across industries.

The system is essential. It is also slow.

Until this year Britain was still relying on a classification designed in 2007. That becomes troublesome when economic activity changes faster than the categories used to describe it.

An AI company may look like a software company. Another may discover drugs. Another may design chips. A large business may employ thousands of people working on AI while most of the organisation belongs to older activities. The economic activity is real, but the statistical boundary around it may not be.

This matters when government starts spending money on the industries it wants to create.

Suppose ministers support new AI infrastructure and later want to know whether the policy increased productive investment. The computers are one part. The buildings are another. Electricity infrastructure sits elsewhere. Software may be produced in Britain or licensed from abroad. Intellectual property may belong to a foreign parent.

There is no single statistical drawer into which all of that activity fits.

Britain is building another map

The Department for Science, Innovation and Technology has already responded by constructing company level definitions rather than relying only on conventional industrial codes.

Its preferred definition of the digital and technology sector contains 107,082 companies, including 10,972 in six frontier technology areas: artificial intelligence, quantum technology, semiconductors, cyber security, engineering biology and advanced connectivity.

Using this approach, the Government estimates that the sector generated about £408 billion of turnover in 2024, employed 1.33 million people and produced £158 billion of gross value added. Estimated output per employee was £119,000, compared with about £78,000 across the wider economy.

AI required an even more specialised exercise. A Government study identified 5,862 companies engaged in AI activity and estimated AI related revenue at £23.9 billion in 2024, with about 86,000 people employed.

Those figures could not simply be lifted from ordinary industry statistics. Researchers assembled the sector from company records, commercial databases, surveys and other information, then estimated how much activity inside diversified businesses actually belonged to AI.

That is revealing. To measure one of the industries at the centre of Britain’s economic strategy, the state has had to construct a second statistical map alongside the conventional one.

The problem reaches beyond AI

The Government’s own methodology shows that AI is not an isolated case.

Clean energy cannot presently be represented adequately by the existing industrial classification for several of the strategy’s main indicators. Defence and life sciences are only partly represented. Advanced manufacturing and digital technologies are measured using proxies.

Britain therefore has a strategy built around eight growth sectors while some of the measures intended to judge its success cannot fully see all eight.

That does not make the strategy meaningless. It makes judging whether it works harder.

The statistical system is catching up

A new classification, SIC 2026, has now been completed. It gives more detailed treatment to software activities, including artificial intelligence, and is designed to reflect changes in the modern economy more accurately.

But statistical systems cannot simply switch overnight. Businesses must be recoded. Registers and surveys have to change. Historical data still need to be comparable with new data.

The ONS says the earliest anticipated use of SIC 2026 throughout the national accounts is the 2031 Blue Book.

By then, the artificial intelligence industry of 2026 may itself look old.

The ONS is therefore building an AI thematic account, which will cut across conventional industry boundaries to identify AI production, investment, infrastructure, imports and exports wherever they occur.

The need is becoming more urgent as adoption accelerates. ONS survey evidence suggests that the proportion of businesses with at least ten employees using some form of AI rose from about 12 per cent in 2023 to around 35 per cent by June 2026. Among businesses with more than 250 employees, the figure was 49 per cent.

The economy is changing while statisticians are constructing the instrument with which to measure it.

Not simply a British failure

There is an important qualification. Britain is not alone.

Artificial intelligence creates similar problems for statistical agencies elsewhere because it does not behave like a conventional industry. It can sit inside software, finance, medicine, manufacturing, advertising or professional services. International statistical systems were not designed around a technology that can spread across almost every sector at once.

Nor should economic classifications change every time a new technology appears. Statistics require stability because governments, businesses and economists need to compare one year with another.

There is a genuine trade off. Statistics need continuity. Industrial policy needs speed.

AI exposes the distance between them.

What cannot be measured is harder to manage

The deeper problem is not whether an AI company has been assigned the right code.

Governments now want to identify promising industries, subsidise research, reform planning, build electricity networks, encourage pension investment and attract foreign capital. They then need to know whether those interventions produced additional investment and productivity or merely rewarded activity that would have happened anyway.

That requires a credible baseline.

Without one, ministers can count announcements, grants, factories promised and jobs expected. Those numbers can be useful, but they are not the same as measuring what actually happened to the economy.

The industries Britain wants to build increasingly cross the boundaries on which its economic statistics were constructed. AI enters medicine. Quantum technology enters computing, defence and sensing. Clean energy reaches across electricity, transport, manufacturing and construction.

This is not an argument against an industrial strategy. It is an argument for recognising one of its least visible requirements.

Before a country can know whether it is building the economy it wants, it needs to know what that economy actually looks like.

Britain is now trying to build both at once.