AI and Employment: What Jobs Are At Risk Across Different Sectors?
AI is not replacing work in one clean sweep. It is redesigning tasks, changing workflows and forcing employers to make sharper choices about productivity, skills, governance and trust. That may be the most important sentence in the AI employment debate.
The popular version of the story is simple: AI arrives, jobs disappear, everyone panics, and somewhere a robot with suspiciously good posture takes your desk. The real story is more useful, and more uncomfortable. AI is already changing employment, although not evenly and not always visibly. In some roles it removes repetitive work. In others it lifts productivity. In others it creates new risk because the task may look easy to automate, while the judgement behind it remains deeply human.
As the research summarises, the real issue for employers is not whether AI will be used, but which tasks will be automated, which will be augmented and which will need tighter human oversight. That framing matters. It moves us away from theatrical predictions and towards practical workforce strategy.
“The question is no longer whether AI will change your job. The question is which parts of your job were never really the job in the first place.”
What does the latest evidence say about AI and jobs?
The strongest current UK evidence points towards task exposure rather than immediate mass replacement. The UK Government’s 2026 labour market assessment says the evidence still does not provide clear answers to many of the questions that matter most for policy, even as AI capabilities are improving quickly in areas such as coding, cybersecurity and research. That uncertainty is important. It should stop lazy certainty on both sides.
AI optimism can become careless when it treats every productivity gain as an automatic improvement to working life. AI pessimism can become equally careless when it assumes every exposed task becomes an eliminated role. The labour market is more complicated than that, because work is made up of tasks, relationships, responsibilities, rules, judgement and context.
The UK is particularly exposed because it is a service-heavy economy. Government assessment cites estimates that around 70% of UK workers are in occupations containing tasks that AI could potentially perform or enhance. Around half of exposed workers are in high-complementarity roles, where AI is more likely to improve output than replace the worker, while the other half are in lower-complementarity roles where substitution risk is higher. That split gives us the useful lens.
Exposure is not the same as replacement. Exposure means AI can touch the work. What happens next depends on the role, the sector, the organisation and the quality of management.
Which jobs are most exposed to AI?
Jobs with high levels of digital, text-based, analytical or repeatable cognitive work are most exposed to AI. That includes tasks in marketing, HR, finance, legal services, administration, customer service, software development, research, consulting and public-sector case handling.
The Department for Education’s 2023 report specifically examined AI’s potential impact across occupations, sectors and geographic areas, including the qualifications and training routes linked to highly impacted jobs. Exposure varies sharply by sector, role and organisational choice. This is why headlines about “jobs at risk” can mislead.
A marketing executive, HR officer, finance analyst and legal associate might all use the same AI tool, but the impact will be completely different. In one team, AI may speed up research and first drafts. In another, it may reduce the need for junior administrative support. In a third, it may create a compliance headache because nobody can explain how a recommendation was produced. The difference is not the technology alone. The difference is workflow design.
Is marketing at risk from AI?
Marketing is highly exposed to AI, but much of the near-term change is augmentation rather than total replacement. AI can accelerate research, content drafting, campaign variation, image generation, video editing, SEO analysis, reporting and customer insight work. It can also make mediocre marketing frighteningly scalable, which is less a productivity miracle and more a warning label.
Government assessment cites evidence of productivity gains in writing-heavy and consultancy-style work, with writing tasks among the areas where AI has shown strong speed improvements in experimental settings. Research also identifies marketing as one of the clearest examples of augmentation, where tools such as ChatGPT, Microsoft Copilot and Adobe Firefly can support research, copy variants and asset production while increasing the need for human editors, brand guardians and legal review. For marketers, the risk is not simply that AI writes copy. The risk is that organisations mistake content production for marketing capability.
A junior marketer who used to learn through research, drafting, editing, testing, reporting and client feedback may find the early stages compressed by AI. That can be good if it removes drudge work and creates more time for strategy, measurement and customer understanding. It can be damaging if it removes the very tasks through which judgement is built.
Marketing leaders therefore need to redesign entry-level work carefully. Do not just hand the first draft to AI and ask the junior person to tidy it up. Involve them in prompt design, source checking, audience interpretation, brand judgement, testing hypotheses, performance analysis and ethical review. That is where the future marketer becomes stronger, not just faster.
“AI can produce the first draft. It cannot take responsibility for whether the work should exist, who it serves, what it risks, or whether it is true.”
How will AI affect HR and recruitment?
HR is one of the most sensitive areas of AI adoption because it combines efficiency potential with high personal and legal risk. AI can help draft job descriptions, summarise interview notes, screen large volumes of applications and identify patterns in workforce data. Used well, that can reduce administrative pressure. Used badly, it can automate unfairness at scale.
HR uses raise issues around bias, explainability, privacy, selection criteria, retention rules and audit trails. This is where “human in the loop” needs to mean more than a person glancing at a spreadsheet five minutes before a decision is made. If AI influences who gets shortlisted, interviewed, promoted, managed or exited, the organisation needs clear rules. What data was used? What criteria were applied? Who checked the result? How was bias tested? What happens if a candidate challenges the decision?
HR teams should be especially cautious about outsourcing judgement to tools that appear objective because they produce neat outputs. A beautifully formatted recommendation can still be wrong, biased or based on criteria the organisation would never defend in public.
What happens to finance, legal and professional services jobs?
Finance and legal roles are exposed because they contain substantial amounts of document-heavy, rules-based and analytical work. AI can help with reconciliation, invoice coding, anomaly detection, management reporting, document review, contract comparison and legal drafting. That does not mean finance directors, accountants, solicitors or compliance specialists become optional. In fact, the more AI is used in these sectors, the more valuable review discipline becomes.
Finance functions need human sign-off on material estimates, assumptions and regulatory submissions, while legal services need supervision because hallucinated citations, missed clauses and confidentiality breaches create direct professional and commercial risk. This is the central lesson for professional services.
AI can reduce the time taken to produce a draft, comparison or summary. It does not remove accountability. Someone still needs to know what good looks like. Someone still needs to understand the client’s context. Someone still needs to notice when the answer is elegant, plausible and entirely wrong.
The professional advantage will belong to people who combine domain expertise with AI literacy. Not people who simply “use AI”, because that will soon be a very low bar. The advantage will sit with people who can supervise AI outputs, challenge assumptions, protect confidentiality and connect technical output to commercial consequence.
Will AI replace healthcare and public-sector workers?
In healthcare and public services, AI is more likely to reshape administrative and decision-support tasks than replace whole roles at scale. The opportunity is significant, especially in scheduling, coding, documentation, call handling, triage and case processing. The risks are also higher because errors can affect patient safety, equality of access, benefits, appeals, public trust and democratic legitimacy.
Research describes healthcare as a constrained augmentation environment, where the consequences of mistakes are higher and human review is essential. There are still public-sector opportunities in call handling, document processing and case triage, alongside risks around unequal treatment, transparency and appeals. ONS evidence cited in its 2025 FOI response shows AI adoption within the public sector was higher in central government, health boards and local government than in some other public bodies in 2023. This is not just an efficiency conversation. It is a legitimacy conversation.
When a customer receives a poor AI-generated product recommendation, they may be irritated. When a resident receives an automated decision affecting housing, benefits, care or access to services, the stakes are different. Public-sector AI must be understandable, challengeable and accountable.
If people cannot understand how a decision was reached, they will not trust the decision, even if the model was statistically impressive.
How will AI affect retail and manufacturing?
Retail and manufacturing are often discussed as if AI only matters to office workers. That underestimates the change. The impact may simply show up in different places.
In retail, AI can improve demand forecasting, product recommendations, pricing support, stock management, customer service and contact-centre operations. The immediate pressure may fall more heavily on back-office administration and customer support than on shop-floor roles. The bigger question is whether employers use efficiency gains to improve service quality and scheduling, or simply reduce headcount and increase pressure on remaining staff.
In manufacturing, AI is likely to affect quality control, predictive maintenance, robotics, process monitoring and safety systems. That can reduce repetitive inspection work while creating demand for technicians, process engineers, data-literate supervisors and safety oversight.
ONS evidence suggests a more nuanced picture than “replace everyone”. In late September 2025, nearly a quarter of UK businesses reported using some form of AI technology, up from 9% when the question was introduced in September 2023. Among businesses using or unsure whether they were using AI, 33% said they were training or retraining existing staff, compared with 10% automating or replacing roles.
That does not remove the risk but it does show that workforce adaptation is already part of the story.
What should employers do now?
Employers should stop asking whether AI will affect jobs and start mapping which tasks, workflows and decisions are affected.
A practical AI workforce review should look at five areas:
- First, identify which tasks are repetitive, digital, text-based or rules-based. These are usually the first candidates for AI support
- Second, separate automation from augmentation. Automation removes or reduces human work in a task. Augmentation helps a person perform the task better, faster or with more range. Confusing the two leads to poor decisions
- Third, set human review thresholds. If AI influences content, people, payments, legal advice, clinical information, public services or financial decisions, human responsibility must be explicit
- Fourth, redesign junior roles. If AI absorbs routine production tasks, early-career employees still need structured ways to build judgement. That means more coaching, more review, more exposure to decisions and better explanation of why work is changed
- Fifth, measure outcomes beyond speed. Productivity is useful, but not sufficient. Employers should also monitor quality, error rates, staff confidence, customer experience, bias risk, rework, complaints, retention and skill development.
The organisations that get this right will treat AI as managed operating change. The organisations that get it wrong will treat AI as a software subscription with a motivational launch email. We have all seen how that film ends.
What should workers do now?
Workers should move from task ownership to workflow ownership. That means understanding how value is created before, during and after the task AI can now perform. If AI can draft the report, your value moves towards framing the question, checking the evidence, interpreting the output, applying judgement and explaining the recommendation. If AI can generate campaign variants, your value moves towards strategy, audience understanding, measurement and creative direction. If AI can summarise a document, your value moves towards knowing what was missed, what matters and what action follows.
The most resilient roles will be those where people can supervise outputs, challenge assumptions and connect AI-produced work to business context. That is the career lesson. Do not compete with AI at the task it does cheaply. Become better at the judgement around the task.
Is AI good or bad for employment?
AI is neither automatically good nor automatically bad for employment. Its impact depends on adoption, management, governance, skills and the quality of organisational choices. The UK Government’s 2026 assessment states that AI can deliver substantial productivity gains, while also emphasising that exposure does not equal adoption and that outcomes depend on how organisations integrate tools and train staff. That is the most balanced conclusion available.
AI can remove boring work. It can also remove learning opportunities. It can improve service quality. It can also create invisible unfairness. It can help smaller teams produce more. It can also flood markets with average work and make human expertise harder to identify. The difference sits in the choices employers make now.
FAQ: AI and employment
What jobs are most at risk from AI?
Jobs with routine, digital, text-based, administrative or repeatable analytical tasks are most exposed. This includes parts of marketing, HR, customer service, finance, legal support, software development, administration and data analysis. Whole-role replacement is less predictable than task-level change.
Will AI replace marketing jobs?
AI will replace some marketing tasks, especially first drafts, content variations, reporting summaries, research support and simple asset production. Strong marketers will become more valuable where they can add strategy, judgement, audience insight, brand control, measurement and ethical review.
Is AI more likely to replace junior jobs?
Junior roles may change first because they often contain routine production and administrative tasks. Employers need to redesign early-career pathways so junior staff still develop judgement, rather than simply checking AI outputs.
Which sectors are most exposed to AI?
Service-heavy, knowledge-intensive sectors are highly exposed, including professional services, finance, marketing, HR, legal, technology and parts of the public sector. Retail and manufacturing are also affected through forecasting, automation, customer service, quality control and predictive maintenance.
What is the best way for workers to protect their careers from AI?
The best career protection is to build AI literacy alongside domain expertise. Learn how to use AI, but also learn how to brief it, check it, challenge it, explain it and apply its outputs in a real business context.
Final thought: the future of work is a management choice
AI will not hit every sector in the same way. It will not remove every exposed job. It will not leave work untouched either. The near-term reality is more uneven and more practical. Some tasks will be automated, but many will be augmented. Some roles will shrink and others will grow around supervision, exception handling, quality control, interpretation and trust.
For marketers, business leaders and educators, the lesson is clear. AI strategy and workforce strategy can no longer sit in separate meetings. Every adoption decision is also a skills decision, a governance decision and a culture decision.
The winners will not be the organisations that use AI the most loudly, they will be the organisations that know where AI helps, where humans matter more, and where the handover between the two must be designed with care. That is where the real future of work is being written.
Contact Neil Wilkins, to discuss how he can help you with marketing resource planning, AI/Human hybrid working and AI process automation.
