Many companies assess AI takeover of work as if it's only about the task itself: how repetitive is it, how much customer contact does it involve, how high are the error costs. That's correct, but there is a factor that cuts across all of that and is rarely named: whether the systems in which that work takes place talk to each other. A task that is on its own excellently suited for AI takeover can in practice remain human work because the data needed to carry out the task is scattered across a CRM, an accounting package, a separate Excel file and one employee's inbox.
The eight axes on which we assess work concern the nature of the task: how much decision-making room, how much variation, how much oversight is needed. System integration is not one of those axes, but it affects several of them at once. A task that would be fully takeable based on decision-making room and repeatability shifts, when integration is missing, into a category with human oversight, not because the assessment of the task changes, but because an extra step arises that no one had planned for: someone has to retype, export or verify data before AI can begin the actual task.
That changes the FTE calculation in a specific way. The time that is freed up is not the full duration of the task, but the duration minus the time that is already spent gathering data from various sources. For some tasks that is a matter of seconds, for example a single click to run an export. For other tasks, the retyping itself is a substantial part of the work, which means the gain from AI takeover turns out small as long as the integration is missing.
Not all work is equally sensitive to disconnected systems. Tasks that take place entirely within one system, such as drafting a text in a single word processor or answering a question based on a single knowledge base, experience little hindrance. Questions employees ask HR often fall into this category: the information needed sits in a single HR system or a single policy, and the need for integration with other systems is limited.
Tasks that are sensitive are those where data from multiple sources must be brought together before anything meaningful can be said or done. A quote that needs data from an inventory system, a price list and a customer history is an example of this, especially when the work involves a great deal of customization and each quote touches a different combination of sources. Work that involves a lot of customer contact is also sensitive, because during that contact an employee or an AI system often has to consult multiple systems to give a complete answer.
For work with high error costs, disconnection counts extra heavily, not because the task itself becomes riskier, but because an error caused by an incorrectly retyped or missed piece of data is harder to trace than an error within a single system. Anyone working with tasks where errors are costly would do well to look at the sourcing question separately from the execution question: is the problem the task, or is it the way the data is delivered.
Two companies with identical work arrive at different outcomes on this axis for a reason that has nothing to do with the task: the age and origin of the systems. A company that has grown in recent years through acquisitions often works with several systems that were never designed to work together. A company that has worked with a single integrated platform from the start does not have that problem, even though it performs exactly the same tasks.
A second reason lies with oversight. Work that falls under external oversight often requires a fixed path for data through systems, with a set sequence of approvals. Anyone doing work that falls under oversight cannot always resolve integration as they see fit as a result: the fixed path is sometimes itself the reason systems remain separate, even when integration would be technically possible.
A third reason is the extent to which employees are already working with AI tools themselves, and where they do or don't report this. Without a foundation of mandatory AI literacy, separate, individual solutions arise: one employee copies data by hand into an AI tool, another does not. That makes the outcome of the assessment dependent on who happens to be doing the work at that moment, not on the task itself.
The conclusion is not that disconnected systems make AI takeover impossible. The conclusion is that the systems question must be assessed separately from the task question, because otherwise two very different problems get confused with each other: a task that AI cannot handle, and a task that AI can handle as soon as the data is in one place. That distinction determines whether the solution lies with the work itself or with the infrastructure around it.
To see how this plays out for your own situation, a first indication can be obtained via the free quickscan: twelve questions, no account needed, with an indication of what share of the hours in your profile could be taken over by AI today. The full work scan, which breaks down tasks on all eight axes including the systems question, is still under construction. What is available today provides a first, provisional picture, not a definitive answer.
Vraag maar. Ik ken de kennisbank van deze site; wat ik niet weet, zeg ik erbij.
Answers come from this site’s knowledge base. Not tailored advice, and not a scan of your company.