A role is a collection of tasks, and those tasks vary widely in whether AI can take them over. An employee who processes invoices, has customer contact and assesses exceptions does three things that each score differently. If you calculate in FTEs, you smear those differences over one job title. If you calculate in hours per task, it remains visible which part of the work is structured and repetitive, and which part requires judgment, contact or exception handling. That is why ftetoai works per task, not per role or department.
When we speak of freed-up capacity, we mean the number of hours that no longer need to be spent on carrying out a given task itself, because AI takes over or prepares that part. That is not a statement about what an organization does with those hours. Freed-up capacity can be deployed for other work, for quality, for backlogs, or for something else. What an employer decides to do with it falls under its own legal requirements and own judgment; our method delivers a factual estimate of the task, not a justification for personnel decisions.
A task does not score on one axis, but on eight: structuredness of the input, room for judgment, customer-contact sensitivity, physical component, creativity, cost of errors, compliance sensitivity, and volume or repetition. Each axis pulls the outcome in a different direction. A task with highly structured input and high volume, but also high cost of errors, produces a different outcome than a task with the same structure and volume but low cost of errors. Because these axes interact with each other, a precise percentage is misleading. A bandwidth shows within which limits the estimate is realistic, and where the uncertainty comes from. More about what such a bandwidth does and does not say can be found at what a bandwidth says and what it doesn't.
The wider the bandwidth, the less unambiguously the eight axes point in one direction. This happens particularly with tasks that have varying costs of errors, with tasks that appear formally structured but in practice involve many exceptions, or with tasks where compliance sensitivity and volume contradict each other. A wide bandwidth is then not a weakness of the method, but an honest representation of a task that does not easily fit into one of the three categories: fully taken over, partially with oversight, or human work. For example, this is at play in managing the document signing process: the process is partly structured, but the legal and compliance component rarely makes full takeover a viable option, and the oversight that remains must approve or reject with reason.
The outcome of our estimate says little to nothing without the context of your own organization. The same task — think of screening customers and partners against sanctions lists — can run largely structured with a limited number of exceptions at one organization, and be full of unpredictable cases that each require judgment at another. Volume, cost of errors and compliance pressure differ by sector, by customer base and by risk appetite. That is exactly why the same work turns out differently for you than for a comparable organization. A generic outcome is a starting point for the conversation, not a final judgment about your situation.
A bandwidth is useful for seeing where in a process there is the most room, and where there is not. If a task lies predominantly in category one, full takeover is a realistic question to develop further. If a task lies in category two, the conversation is about the form of human oversight: who approves or rejects, and on what grounds. If a task lies in category three, the estimate mainly provides clarity about why AI does not play a role there, for example due to high cost of errors or a physical component that cannot be automated. Even with tasks that at first glance seem routine, such as digitizing and making a paper archive accessible, the eight axes together determine which category the task falls into, and that is rarely a matter of one dominant characteristic.
If you would like an initial indication for your own situation, you can fill in the free quickscan from ftetoai: twelve questions, without an account, resulting in an indication of which part of the hours in your profile can be taken over by AI today. The full work scan, which goes deeper into individual tasks and processes, is still under construction. The quickscan gives a direction, not a definitive answer; for that, more context per task is needed than twelve questions can offer.
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.