Deutsche Bank has compared thirty economies on how far artificial intelligence could reach into their workforces, and Britain came out first.
The interesting part is not the ranking. It is the second of the two reasons behind it, which moves the argument out of the labour market entirely.
Two Reasons, Not One
The first driver is familiar: an unusually high share of domestic jobs sit in roles the technology touches. Britain’s large service sector means a comparatively high proportion of employees work in tasks AI could automate or substantially change.
The second is the one worth sitting with. The UK’s export base is weighted towards precisely the services most likely to be affected, according to the analysis.
Its author, Deutsche analyst Shreyas Gopal, was direct about the asymmetry. Britain sells the world a great deal of services vulnerable to automation while buying in goods and other services. That turns a question about jobs into a question about the balance of payments.
A Different Question From the Usual One
Most AI analysis asks how many roles disappear. This one asks what happens to a country whose principal export is the very thing being automated.
Those are not the same question and they do not have the same answer. A domestic job lost to automation is, at the national level, partly offset: the output still happens, more cheaply, somewhere in the same economy. An export lost to automation performed abroad is not offset at all. The revenue simply stops arriving.
That is why the framing matters for a country running a goods deficit and a services surplus. The services surplus is the thing under discussion.
It also changes who the competitor is. In the domestic version of this story a British worker is displaced by a British firm’s software, and the value stays in the country. In the export version, a client in New York or Frankfurt stops buying an hour of British analysis because a model does it in-house. There is no domestic offset in that transaction at all, and no British firm captures the saving.
Exposure Is Not Displacement
The distinction the report itself draws is the one most likely to be lost in summary, so it is worth stating plainly.
Exposure describes how much of an economy sits in the technology’s path. It does not describe how much gets swept away. A high exposure score could equally precede a substantial productivity gain, provided the workforce moves and firms invest in the transition.
Deutsche’s report measures the first and does not claim to settle the second. Treating an exposure ranking as a forecast of job losses is a misreading of what was actually published.
The distinction has a practical consequence for how the ranking should be used. Exposure is a measure of concentration, and concentration cuts both ways: an economy heavily weighted towards the work a technology improves stands to gain disproportionately if it adopts well, and to lose disproportionately if it does not. First place on this list is a statement about variance, not about direction.
The Week It Landed In
The finding did not arrive in isolation, and the surrounding evidence points the same way without proving the case.
Work Foundation research reported six in ten large UK firms attributing fewer junior roles to AI, and separate figures put graduate vacancies at their weakest in at least a decade. The OECD has already flagged London’s exposure specifically. Bank of England governor Andrew Bailey warned earlier this year that the technology could destroy jobs faster than expected.
The underlying labour market is soft in a way that is consistent with all of this without being caused by it. ONS figures show payrolled employees down 86,000 over the year to April to June 2026. That is a slow contraction, not a collapse, and no single cause can be assigned to it.
The Counter-Case
Not everyone accepts the framing, and the dissent comes from a serious quarter.
AstraZeneca chief executive Pascal Soriot has argued the effect is people doing their existing work better rather than losing it. That is the augmentation case, and it has the advantage of describing what has actually happened with most previous workplace technologies over a long enough horizon.
Soriot’s position also has a specific advantage in his own sector. Pharmaceutical research is a field where the constraint is the difficulty of the problem rather than the volume of routine output, and technologies that accelerate analysis there tend to expand what gets attempted rather than reduce who is needed. Whether that generalises to a mid-sized professional services firm billing by the hour is precisely the open question.
The honest position is that Deutsche’s report does not adjudicate between the two. It measures how much of the economy is in scope. Whether being in scope means being replaced or being made more productive is a separate argument that this data cannot settle.
Why Sequencing Matters More Here
What the ranking does establish is that Britain has less room than its peers to be relaxed about the order in which things happen.
An economy with more of its output in goods can watch how this plays out elsewhere and adapt with a lag. On this reading the UK cannot, because the exposed sector is the one paying for the imports.
That is an argument about timing rather than severity. It does not say the outcome will be worse in Britain. It says Britain will find out sooner, and has less capacity to absorb being wrong about the pace.
Timing arguments are also the hardest to act on, because they require a decision before the evidence is conclusive. By the time the effect on services exports is visible in trade data it will have been underway for some years, and the investment decisions that would have positioned firms for it will already have been taken or missed. That is an uncomfortable place for policy and for boards, and it is the actual implication of ranking first.
What a Business Should Take From It
For an individual firm the national ranking is close to useless as a planning input, and the underlying observation is not.
The question worth asking is narrow: what proportion of the work sold to customers is the kind that AI is getting good at, and is any of it sold abroad? A UK professional services firm billing international clients for analysis, drafting or processing is precisely the profile the report describes. A UK manufacturer selling physical product is not, whatever the national score says. The ranking is an average across an economy, and almost no individual firm sits at the average.
The gap between adopting the technology and being changed by it also remains wide. Adoption in this country is still shallow, with AI adoption at 35% of UK firms and just 1.6 tools deep, which is not the profile of an economy that has already restructured around the technology.
Exposure, in other words, is a description of the starting position. It leaves almost everything about the outcome still to be decided, which is both the reassuring reading and the demanding one. This is the same split running through most of the summer’s data, where every dataset divides the economy along the same line.
About The Author
More Stories
UK House Prices Rose 2% and London Fell for a Tenth Month
Trusted Payments Holds 10% in Escrow Until You Sign Off
UK Manufacturing Hired at a Two-Year High While Output Slowed