The impact of AI on jobs: what the macro numbers don't tell you

Every week brings a new study on AI and employment. One firm announces that 40% of administrative tasks will be automatable by 2030. Another talks about 300 million jobs “affected” worldwide. These numbers are true — and largely useless to you. None of them say whether it’s your customer service team, your accounting department, or your legal function that needs to transform first. None of them say on what timeline. Here’s how to answer both questions with data, not extrapolation.

The right question isn’t “will AI replace my job”

That’s the question everyone asks, and it’s the wrong one. A job isn’t a monolithic block that AI either replaces or doesn’t — it’s a bundle of dozens of tasks, each with a different automation potential. A financial controller who consolidates spreadsheets, writes summary reports, and arbitrates budget priorities is performing three activities with wildly different exposure. The first is already largely automatable. The third won’t be for a long time.

Asking the question at the job level gives you a binary, anxiety-inducing answer. Asking it at the task level gives you a plan: which hours to free up, which skills to strengthen, which roles to redesign — and in what order.

Three families of roles, three trajectories

AI exposure isn’t distributed evenly across an organization. It tracks fairly closely with the nature of the work itself:

• Customer relations and support — interactions with a strong procedural component (order status, FAQs, first-line support) shift quickly. Interactions with high emotional or commercial stakes hold up far better.
• Administrative and back-office functions — data entry, reconciliation, regulatory reporting: this is where automation potential is highest, and most underestimated internally.
• Support functions (HR, finance, legal) — AI absorbs preparation and first-pass analysis; decision-making, arbitration and the human relationship stay at the core of the role.
• Middle management — the job changes shape more than it disappears: less manual reporting, more arbitration on recommendations the machine has already drafted.

These trajectories aren’t sector generalities. They depend on your org chart, your current tooling, and the real digital maturity of each team — three variables no macro study knows anything about.

The risk isn’t that AI eliminates roles overnight. The risk is that you keep recruiting, training and organizing as if nothing had changed.

Why sector statistics don’t apply to your organization

A study announcing that 30% of accounting tasks are automatable is reasoning about an average job, in an average sector, with average tooling. Your accounting function isn’t average: it has its own level of ERP integration, its own volumes, its own history of partial automation. The national figure can be true on average and wrong for you — in either direction.

That’s the trap of sector benchmarks in an executive committee meeting: they create a feeling of command over the subject without giving you a single lever to act on. How many roles are affected, in which division, on what horizon — those questions only get answered by your own data.

What a role-by-role diagnostic actually measures

A serious diagnostic starts from your data, not a market average. The mechanics are simple in principle:

1. Decomposition — each role is broken down into elementary tasks, based on your job descriptions and real headcount volumes.
2. Scoring — each task is checked against a reference library (several hundred roles already pre-scored) to estimate its current automation potential.
3. Consolidation — scores are aggregated by role, then by division, to surface real priorities rather than intuitions.
4. Quantification — freed-up time is converted into full-time equivalents, cost, or redeployable capacity, based on your own volumes.

The output isn’t a single AI-exposure score for the whole company: it’s a map, division by division, role by role, of what moves first — and what can legitimately wait.

From diagnostic to action plan: three horizons

A diagnostic that sits on a shelf is worth nothing. The next question is always the same: where to start, and at what pace.

1. 0–6 months — quick wins. Tasks with high automation potential and low organizational risk are handled first; the time freed up funds the rest of the plan.
2. 6–18 months — redesigning roles. Job descriptions evolve to reflect the time actually available: fewer execution tasks, more value-added work reallocated to the role.
3. Beyond 18 months — workforce planning. Recruitment, training and internal mobility plans adjust to the trajectories identified role by role, rather than to a blanket cost-reduction target.

The regulatory framework no longer leaves room to wait

In France and across the EU, sitting on the sidelines is becoming a risk in itself. The EU AI Act imposes documentation obligations on high-risk uses starting August 2026 — December 2027 for entities covered by the Omnibus mechanism. The triennial GEPP already requires, for any company above 300 employees, a works-council (CSE) consultation whenever a transformation touches the organization of work. Without precise data, that consultation runs on estimates or fears — shaky ground for any leadership team.

Organizations that walk into that dialogue with a factual, role-by-role diagnostic turn a defensive negotiation into a discussion about means. Those that walk in with sector generalities end up improvising instead.

The question is no longer whether AI will transform your roles. It’s whether you steer that transformation with precise data — or discover it after the fact.