A leadership team that says "we're going to use AI" often doesn't mean the same thing by it. Half think of a language model that summarizes reports, the other half think of systems that partly handle customer contact themselves. Both require technology that can process that and capacity that can guide it. Those are two different kinds of readiness, and they are rarely tested separately from each other.
The cost of AI-readiness is not primarily in licenses. It lies in what needs to happen before a task can safely move from a system to AI: data that is findable and current, integrations that exist or need to be built, and people who know when to approve an outcome and when not to. What that costs depends on how many of those components are already in place and how many are still missing.
Within any ambition that assumes AI work, the underlying work falls into three categories. AI can take over the task, without a human still looking at it. A human maintains oversight and approves or rejects, for a reason. Or the work remains human work, because judgment, context, or responsibility requires it.
The technology needed differs per category. Full takeover requires reliable, well-structured data and a system that recognizes error signals before they cause damage. Oversight with approval requires an interface in which a human can quickly see why a system arrived at an outcome. Human work often requires no new technology, but does require clarity about who remains responsible for what — that touches on decision rights that need to be reassigned during a change.
Companies that already make this distinction notice the difference in how quickly a pilot moves from trial to production. Companies that don't make it keep testing without it becoming clear which part of the work actually shifts.
The difference between companies that already run this smoothly today and companies that aren't there yet usually doesn't lie in the choice of technology. It lies in three things that precede it.
First: whether the organization knows which tasks make up which process, separate from the job title attached to it. Without that inventory, any statement about "AI-readiness" is a guess.
Second: whether the systems that currently exist produce data that an AI application can use. Many organizations have data that is correct for a human reading it, but not structured enough for a system that needs to compute with it.
Third: whether there is capacity — in hours, not in good intentions — to guide the transition. Someone has to check the first hundred outcomes before a system is allowed to run without oversight. Those hours usually appear in no one's planning.
The first visible sign is not a cost saving, but a shift in where time goes. Work that used to consist of executing becomes work that consists of assessing. That requires different skills and sometimes a different role, not automatically fewer people.
The second sign is that questions that used to sit with IT now sit with leadership. Which decision may a system make on its own, and which must always be confirmed by a human? That is a decision right, not a technical setting, and it needs to be recorded somewhere before the incident occurs.
The third sign is in the budget. Hours freed up in one place don't automatically disappear as savings; they often reappear as extra capacity elsewhere, for example in oversight, in customer contact at moments where AI is not sufficient, or in maintaining the systems themselves. What that means net for the budget is tied to how financial resilience has been built up and what margin exists to absorb that shift before it yields a return.
This assessment says nothing about who an organization keeps employed or lets go. That is a decision with its own legal requirements, outside what is described here. What is described here is capacity: which hours become available, which hours are additionally needed, and which technology needs to be in place for that. The same applies to collaborations with other parties — what makes a partnership ready for AI partly determines whether an in-house shift actually works throughout the chain.
It also touches compliance. A system that takes over part of a process changes who is demonstrably responsible for what, and that deserves its own assessment of compliance and risk that need to be made AI-ready.
There is no percentage that applies to every organization. It depends on how much data is already usable, how many processes are already documented, and how much capacity there is to guide the initial period. Companies with a lot of unstructured data and little documentation start with more work than companies that already have that in order. That difference is exactly why an estimate made in advance says little without first looking at the work itself, task by task — that is what the [work scan from FTE TO AI](/) maps out per task.
To see where your organization is furthest along and least far along, there is a free readiness check: eight short questions, one per dimension, with a picture of where readiness already stands and where work still remains. The full ambition assessment, with the four layers and 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.