Improving margin while AI takes over work: what the organization must be capable of

The ambition as it comes to the table

In the boardroom it often sounds like a single sentence: margin needs to go up, and AI helps with that. Everyone nods, but one person thinks of lower procurement costs, another of faster turnaround times, and a third of less manual work in the back office. That is not a problem with the ambition, but with the translation. Improving margin with AI is not a cost-cutting measure, it is a shift in who does the work and who checks whether it has been done correctly.

That shift runs along three tracks that overlap. Part of the work can be taken over by AI itself: standard reports, reconciliations, initial analyses of deviations. Another part remains a collaboration, where AI makes a proposal and an employee approves or rejects it, with a reason attached. And a third part remains human work, because it requires judgment, negotiation, or context that is not contained in data. Improving margin means you must be able to identify these three tracks in your own company, per process, not in general terms.

Why it already works at one company and not at another

The difference is rarely in the technology. Two comparable companies with the same AI tools achieve different results, because one has arranged who is allowed to approve the outcome of an AI proposal, who monitors the quality of the data on which those proposals run, and who is responsible if a margin improvement causes a problem elsewhere, for example in delivery reliability, while the other has not.

Companies where this is already working have made that decision right explicit before the work shifted. They knew which role assesses the output of an AI model on a production schedule, and which role only monitors the process. Companies where it is not yet working only discover this when the first deviation occurs and no one knows who is entitled to have a say.

What the organization must be capable of

Improving margin under these circumstances requires readiness on eight dimensions, not one. A few that in practice most quickly slow down an ambition:

Data reliability. An AI proposal about procurement margin is only as good as the underlying data. If invoicing, inventory, and contracts have not been recorded consistently, every proposal becomes a guess with a confident appearance.
Decision rights. Who approves an AI-generated proposal, who is allowed to reject it, and on what grounds. Without that answer, oversight remains a formality.
Process ownership. Margin often sits between departments: procurement, production, sales. An improvement that generates margin in one place can cause costs elsewhere. Someone must monitor that overview.
Ability to read AI outcomes. An employee who must approve or reject an AI proposal must understand what the proposal is based on. Without that skill, approval becomes a formality in the other direction.

The ambition assessment captures this ambition in four layers: the vision as articulated by leadership, the vision state that describes what the work looks like after the shift, the target with a concrete goal, and the target state that describes what capacity is freed up as a result. These four layers are tested against eight dimensions and five confidence gates, and translated into roles and decision rights that must already be filled in today, not only once something goes wrong.

Roles that come with it

Improving margin with AI does not automatically require new functions, but it does require new responsibilities within existing roles. A controller who now compiles reports becomes the role that assesses AI-generated deviation analyses. An operational manager who now approves schedules becomes the role that explains why an AI proposal was rejected, so that the model learns from it. That calls for a different kind of time allocation: fewer hours spent compiling the overview, more hours spent assessing it.

This shift touches on the question of how much FTE capacity is freed up and where that capacity is subsequently deployed. That is a question about work, not about personnel; what an employer does with that falls under its own legal requirements and is up to the employer.

What you will notice about it after a year

No leadership team needs to wait for an annual report to see whether this is working. The signals are visible earlier: a team that approves or, with reasons, rejects AI proposals within days, instead of letting them sit for weeks. A controller who can point to which part of the margin improvement is attributable to an AI proposal and which part to their own intervention. An organization that no longer discusses whether AI can handle the work, but which deviation deserves attention. More about what these early signals precisely are is described at what are early signals that a change is taking hold.

This ambition rarely stands alone. Improving margin often touches the same decision rights as what capabilities faster delivery requires when AI takes over work, and in a merger or acquisition the question returns in accelerated form, as described at what capabilities integrating an acquisition requires when AI takes over work. Anyone who notices that a strategic plan remains on paper despite this ambition will find an explanation at how to prevent a strategy from remaining a document.

What to do now

The underlying question is usually not whether AI can improve margin, but which work in this specific company can actually be taken over. That question is answered per task with the work scan from FTE TO AI.

Anyone who first wants to know where the organization stands can take the free readiness check: eight short questions, one per dimension, giving a picture of where you are furthest along and where you are least far along. The full ambition assessment, with the four layers and the five confidence gates, is under construction.