An ambition like "AI lowers our cost per file" or "AI increases our revenue per employee" sounds logical, but it relies on something specific: that a portion of the work people currently do shifts to AI that takes it over, or to AI that prepares it with a human who approves or rejects it. The margin gate therefore asks a narrower question than "does AI pay off". It asks: which portion of the work in the designated tasks can actually shift, at what cost for that shift itself, and does enough margin remain afterward to justify the investment.
That is a different question from a business case with an expected return. A business case calculates forward. The margin gate tests whether the calculation assumptions behind that business case can already be substantiated with what the organization knows today about its own work.
In some companies this test is already routine work: there is an overview of tasks, and for each task it is known whether AI can take it over, partly take it over with oversight, or cannot take it over, and the cost per task is known well enough to calculate a shift. In other companies that overview does not exist, and the margin assumption is based on a demo, a vendor claim, or a comparable company that supposedly did something similar.
The difference is not in ambition level but in measurement work. Companies that already have time registration per task, or a clear cost breakdown per process, can test margin viability today. Companies where work is only described at the role level must first break down that work before there is anything to test. That breakdown is not a side issue: it is the precondition for making the distinction between the three categories -- taking over, taking over with oversight, remaining human work -- at the task level rather than on gut feeling.
The margin gate is satisfied when three things are on the table, not when there is a positive feeling about the business case.
One: an estimate of freed-up hours or FTE capacity per task, with a range rather than a single figure. Work that AI partly takes over still requires human control hours; you count those in, not as a correction afterward but as a fixed part of the calculation.
Two: an estimate of the costs that the shift itself brings with it -- setup, control setup, maintenance of the oversight that approves or rejects with reason. A task that AI takes over without oversight is cheaper to maintain than a task that permanently requires human control; that difference must be visible in the calculation.
Three: a statement about the time horizon within which the difference between costs and freed-up capacity pays for itself, with a range that indicates where the uncertainty lies -- in the adoption speed, in the quality of the data, in people's willingness to do the new work.
If these three points are presented not with ranges but with firm single figures, that is itself already a signal: someone has turned an assumption into a fact.
A margin gate that does not pass does not mean the ambition is called off. It means one of the underlying assumptions is not correct, and that can be traced. Three routes tend to recur.
The first: the freed-up-hours figure is too optimistic because the task is less AI-suitable than thought. Then the question is not "how do we push down costs" but whether the work in this company can genuinely be taken over by AI, mapped per task through the work scan -- because an incorrect task breakdown makes every margin calculation unreliable.
The second: the shift costs are higher than expected because people need more guidance than planned. That touches on whether your people can handle it, with a picture of what training and what replacing requires. Important: if a negative margin outcome raises the question of what this means for staff, a separate legal framework applies to that; this test provides no substantiation for dismissal decisions and makes no statement about that.
The third: the time horizon is off because the ambition was set up too large at once for the year in which it must yield returns. Then it helps to look at how you phase an ambition that is too large for one year, so that the margin test can be redone per phase rather than once for the entire trajectory.
An outcome you do not like is therefore not an endpoint. It is an indication of which assumption must first be corrected before the ambition goes through the gate again.
The margin gate does not stand on its own; it is connected to readiness on other dimensions, such as whether vendors can technically support the shift -- to be tested via whether your vendors can keep up with the requested shift -- and whether the organization can handle the pace of change, to be tested via how fast an organization can really change.
Anyone who wants to know where their own ambition currently stands can take the free readiness check: eight short questions, one per dimension, resulting in a picture of where the organization is furthest along and least far along. The full ambition test, with the four layers and the five confidence gates, is under construction.
Vertel wat u wilt bereiken, dan kijken we samen wat daarvoor moet staan.
Answers come from this site’s knowledge base. Not tailored advice, and not a scan of your company.