AI Isn't Coming for Everything. But Where It Is Coming, It's Coming Fast.
Four functions where the change is structural, not incremental — and why leaders need to stop talking about AI in general.
The most common mistake leaders are making right now is treating AI as a single, uniform wave. "We need an AI strategy." "AI will change everything." "AI is going to disrupt our industry."
It won't. Or rather — not equally, not everywhere, and not at the same speed.
If you look past the LinkedIn theatre at where AI is actually changing how work gets done, the pattern is sharp. There are four functions where the shift is structural — meaning the work itself is being re-shaped, not just sped up. And there are dozens of functions where AI is helpful but not transformative.
Leaders who understand the difference are the ones building real advantage. The ones who don't are spreading AI investment across the org like jam across toast — thin, even, and ultimately tasteless.
What "structural change" actually means
I want to be careful here, because "transformation" is one of the most abused words in business writing. Including, sometimes, mine.
A productivity gain is when the same work gets done faster. A structural change is when the work itself — its shape, its inputs, its handoffs, its quality criteria — becomes something different. The role description changes. The team size changes. The skills required change.
The Excel spreadsheet was a productivity tool for accountants. The cloud-based BI platform was a structural change for the entire finance function. ChatGPT, for many functions, is a productivity tool. For four functions, it is something closer to the BI platform moment.
Those four: content production, sales, finance and back-office operations, and HR & people development.
Why these four
The pattern is not random. Functions get structurally re-shaped by AI when they have three properties at once.
High text and document density. The function lives in language. Briefs, proposals, contracts, emails, reports, transcripts, candidate CVs, customer support tickets. AI is, fundamentally, a language engine. It will reshape functions where language is the work, far more than functions where physical or judgement-heavy activity is the work.
Repeatable patterns under the surface. A finance close looks different every month, but the underlying logic is repeatable. A sales outreach feels bespoke, but the structure is patternable. A job description is unique, but the genre is well-understood. AI is brilliant at handling variations on themes. It is much weaker on genuinely one-off creative leaps.
Judgement that can be scaffolded, not replaced. This is the subtle one. AI is not going to make the final call on hiring a senior leader. It will, however, do 80% of the screening, summarising, comparison, and preparation work that goes into the final call. Functions where the senior judgement remains human but the scaffolding around it can be automated — those are the functions where AI delivers outsized value.
All four functions tick all three boxes.
Content production
This is the most visible one and the most over-claimed. Yes, AI can write a blog post. No, it cannot replace your senior content strategist. But the production economics of content have genuinely shifted.
A first draft of a long-form post that took four hours now takes thirty minutes plus an hour of editing. A bilingual newsletter that required a translator is now produced bilingual at source. SEO research that took half a day for one keyword cluster now takes twenty minutes for ten. A social calendar that used to need a freelancer is now drafted in a single morning.
What changes structurally is the ratio of strategic to executional work. The senior content people spend less time producing and more time directing. The junior production roles either evolve into editorial roles or disappear. Mid-sized organisations that could never afford a full content team can suddenly produce at the volume of one.
For DACH-region SMEs this is genuinely consequential. Most KMU have under-invested in content for a decade because the unit economics didn't work. They do now.
Sales
McKinsey estimates that around 20% of sales activities are already automatable with current tools. Salesforce's 2026 State of Sales reports that sellers using AI agents expect a 34% reduction in research time and a 36% reduction in email drafting time. Teams that used AI in the past year saw 83% revenue growth versus 66% for teams that did not.
But these numbers undersell what is actually happening. The structural change in sales is not about email automation. It is about who does the preparation, qualification, and personalisation work.
For decades, the SDR role existed because senior salespeople couldn't afford to spend their time on prospect research and outbound. That economic logic is now collapsing. Account research that took two hours per company now takes ten minutes. Outbound personalisation that required reading three press releases now requires one good prompt. The work doesn't disappear — but the seniority of the person doing it changes.
We will look back on the SDR-as-volume-machine model in five years the way we look back on the typing pool in 1985. The function will exist, but its shape will be completely different.
Finance and back-office operations
This is the function where the MIT data is most striking — and most ignored. Companies are pouring AI budgets into sales and marketing, where the visible-glamour value is highest. But MIT found the biggest measurable P&L return in back-office automation: eliminating outsourcing contracts, cutting external agency costs, streamlining month-end processes, and automating reporting.
Accounts payable that needed three FTEs now needs one and a half. Invoice matching that required manual review now runs as a pre-screened queue with humans handling only the exceptions. Compliance documentation that took two weeks to produce now drafts itself overnight and gets reviewed in two days.
In Swiss SMEs in particular — where finance functions are often lean and the cost of additional headcount is high — this is the place where AI most often pays for itself within a quarter. It is unsexy work. It also returns the highest IRR.
HR and people development
The structural change in HR is the most contested, and probably the most uncomfortable. Recruiting workflows are being completely re-engineered. CV screening that took an hour per candidate now takes minutes. Interview transcripts get summarised and scored automatically. Onboarding documentation that hasn't been updated in three years suddenly becomes maintainable.
Learning and development is in the middle of its own reckoning. Generic e-learning content has become a commodity overnight. What hasn't is applied, contextual, in-flow learning — the kind that helps someone solve the problem in front of them, in their own role, with their own data, today. That is a different L&D function, and it will require different skills.
People analytics is the other shift. The data has been there for years; the capacity to interrogate it intelligently has not. AI changes that. The HR business partner role of 2027 looks much more like a strategic data analyst with people skills than the policy-and-process role of 2018.
What this means for leaders
If you are running an organisation, the practical takeaway is this: stop talking about AI in general. The conversation that matters is not "how do we adopt AI?" The conversation that matters is "in which of these four functions do we have the most to gain, and what would it take to actually capture it?"
For most DACH SMEs, the answer is not all four at once. It is one. Pick the function where the workflow density is highest, where you have a leader willing to genuinely redesign the work, and where the business case writes itself. Win there first. Use that win to fund and credentialise the next one.
The companies that try to "do AI" across all functions simultaneously end up with twelve underwhelming pilots and no transformation. The companies that pick one function and rebuild it end up with a competitive advantage that lasts five years.
That is the choice that is actually in front of you. The AI is the easy part. The discipline of picking — and saying no to the rest — is the hard part. As it always is.