A process consists of steps, and each step has a different relationship with AI. Assessing an application consists of checking data, applying criteria, recognizing an exception, and recording a decision. Of those four steps, AI can often take over the first, the second partly with a human who approves or rejects with reason, and the third and fourth generally remain human work. The process itself does not change category. The steps do, and that is where the money and time lie: not in labeling "this process is AI-suitable", but in breaking apart each step and redesigning what remains afterward.
The costs involved rarely have a fixed amount. They depend on how many processes there are, how consistently they are currently carried out, and how much of the knowledge about that execution resides in people rather than in documentation.
Readiness on this dimension requires three kinds of work, and they are unevenly distributed across organizations.
The first is process and task inventory: recording what actually happens now, step by step, rather than what the process document claims. In many companies practice deviates from the diagram, and that deviation is precisely where AI application gets stuck.
The second is redesigning the handover between AI and human. When part of a task shifts to oversight rather than execution, the employee's role changes: from doing to assessing. That requires different skills, a different allocation of time, and often a different kind of responsibility. This design work is often underestimated, because it does not feel like "introducing AI" but like "rethinking the process".
The third is setting up the feedback loop: what happens when oversight issues a rejection. Without a defined route for rejection, the human ends up redoing everything in practice, and the benefit of the AI step disappears. This is one of the most underestimated cost items, because it only becomes visible after the first step has already been taken over.
Some companies already have this operational: a claim is submitted digitally, a system assesses the standard cases, an employee only receives the doubtful cases along with a rationale for why the system is uncertain. Other companies with a comparable process do not have this, and the difference rarely lies in the available technology. It lies in whether the process was ever split into steps that were separately assessed for what AI can do, what requires oversight, and what remains human work.
Companies that have already done this have generally also invested in the two steps before it: clear criteria for what counts as an exception, and a defined way to handle that exception. Companies that have not yet done this often only discover, at the first attempt at AI application, that those criteria were never written down, existing only in the minds of experienced employees.
The first sign is confusion about what "making the process ready" concretely means. Some on the management team mean purchasing a system, others mean rewriting work instructions, and still others mean training employees for a new role. All these meanings are part of readiness, but they are different work packages with different timelines and different owners.
The second sign is a mismatch between ambition and current governance. If the vision state assumes that a process will largely be handled by AI, while the current decision rights for that step still lie entirely with an employee, that is not a technical problem but a governance question: who will be allowed to override a rejection from the AI step, and on what authority. That question touches on how work is divided and assessed within the organization; specific legal requirements apply to that, and it is up to the organization itself to determine, not something this page makes claims about.
Process readiness does not stand alone. When a task shifts from execution to oversight, what needs to be measured also changes: no longer how many tasks were completed, but how many rejections there were and why, which relates to how performance measurement must change as AI takes over part of the work. It also touches the systems in which those steps are recorded, which in turn connects to what technology and capacity are needed before a process can actually shift. And because employees are the ones filling the new oversight role, it works better when they are involved early in the design, as described in how the workforce is involved in a strategy without it taking years.
The underlying question — which work in this specific company, in these specific processes, can genuinely be taken over by AI — is not answered by a general estimate but by the FTE TO AI work scan, which investigates this per task.
The free readiness check consists of eight short questions, one per dimension, and produces a picture of where your organization is furthest along and least far along, including this dimension. The full ambition assessment, with the four layers and the five confidence gates, is under construction. Anyone who already doubts whether the ambition in the vision state matches what has been set up in practice in terms of decision rights and oversight will find an initial step in how a vague vision is translated into something concrete enough to test.
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.