Blog · Research · 12 July 2026 · 8 min read

Which Jobs Will AI Replace First? An Evidence-Based Analysis

The direct answer, from documented deployment evidence rather than prediction: AI is replacing high-volume, rules-based information work inside large organizations first — finance and accounting processing, payroll, HR service delivery, customer service, and routine IT operations, concentrated in shared-services and back-office units. That is where our evidence base — 1,822 documented entries covering 506 companies across 21 industries — shows displacement actually underway, not merely forecast.

What the exposure data shows by sector

Average AI exposure score (0–100; higher means more of the tracked work is demonstrably automatable with tools deployed today) across the verticals in the dataset:

Two things stand out. First, the top of the table is institutional: financial services and technology-sector operations carry the highest average exposure. Second — and this is the part the doom headlines skip — the largest group in the dataset,business services & sme (455 companies, mostly small and mid-sized firms), averages just 16/100. Only 35 of the506 tracked companies sit in the high or critical exposure tiers. The displacement frontier is real, but it is narrow: it runs through large-organization back offices, not through the whole economy at once.

The documented signals behind the scores

Exposure scores aggregate evidence entries — public announcements, deployment reports, and workforce statements. The highest-scoring companies in the dataset right now, with the signals that put them there:

• IBM Hungary ISSC (90/100): CEO: "30% of back-office replaced by AI"; AskHR handles 94% of HR queries; $3.5B productivity gains
• Citi Solutions Center (90/100): 20,000 global cuts by end-2026; CFO: AI will allow "running middle-office with fewer employees"; Fraser's Jan 2026 memo
• Allianz Technology (90/100): 1,500-1,800 AI-driven cuts; Anthropic partnership for "agentic AI operations automation"; call center roles explicitly targeted
• Morgan Stanley (80/100): 30% back-office efficiency gains; AskResearchGPT deployed; Debrief at 98% adoption; some staff let go "because AI automated their roles"
• BT Group (80/100): 55,000 cuts by decade-end; 10,000 explicitly AI-attributed; CEO: AI will "deepen already significant cuts"
• Accenture (80/100): 22,000 cut in 6 months; $865M restructuring; CEO: "exiting" people who can't reskill

Read those again as a pattern: back office, middle office, HR queries, call centers, reskilling ultimatums. Nobody on that list announced "we replaced our salespeople" or "we automated the plumbers." The work being absorbed first is desk work with high volume and low exception rates.

Which roles that maps to

In role terms, the highest near-term exposure in the evidence base concentrates in:

• Transaction processing: accounts payable/receivable, payroll, billing, reconciliation — the core of the shared-services model.
• Tier-1 service delivery: HR helpdesks, customer service, IT service desks — anywhere the first response follows a playbook.
• Reporting and summarization: status reports, meeting summaries, standard research briefs — the connective tissue of middle-office work.
• Routine review: first-pass document review, compliance checks, quality screening at volume.

And the mirror image — lowest current exposure: work anchored in physical presence, relationships, exception handling, and accountability. Real estate scores near the bottom of the vertical table; so do generalist roles in small firms, where one person's job is twenty intertwined tasks no single tool covers.

The honest caveats

This is deployment evidence, not prophecy. Three limits worth stating plainly: the dataset over-represents organizations that announce things publicly; exposure describes tasks, not the political decision of whom to lay off; and a low industry average does not protect an individual whose personal task mix is 80% reporting and summarization. Sector averages are the water temperature — the full table is public — but your own exposure depends on your specific tasks, tools, and role. That is a per-person question, which is exactly why the free assessment scores your task mix rather than your job title.