Every task we assess gets one of three outcomes: AI can take over the task, AI can do part of it under human oversight that approves or rejects with reason, or the task remains human work. That outcome does not come from a loose estimate. We lay every task along eight axes. Together they determine how suitable a task is to hand over to AI, and with what degree of certainty we say so.
Structuredness of the input. Does the information come in in a fixed format, such as a structured file or a form, or must meaning first be extracted from free text, speech or a photo? The more unstructured the input, the more a task leans on human interpretation.
Judgment scope. Is there one correct outcome that can be derived from rules, or must a trade-off be made each time between interests that may conflict? A lot of judgment scope usually means no full transfer, but at best the middle category.
Customer-contact sensitivity. Does the customer or relationship notice anything of how the task is carried out, and how great is the damage if it goes wrong? Internal administrative processing weighs differently here than something shared directly with a customer.
Physical component. Does the task require an action in the physical world, such as lifting, installing or inspecting something on site? Purely digital tasks score low here, and that increases the chance of transfer to AI.
Creativity. Does the task require coming up with something new without a clearly best answer, or variation within a recognizable pattern? Purely generative work without a testable criterion is difficult to fully transfer.
Cost of error. What does a mistake cost: a small correction, or a financial, legal or reputational problem that is difficult to reverse? High cost of error pushes a task more often toward the middle category, with oversight that approves or rejects.
Compliance sensitivity. Does the task fall under laws or regulations, internal control requirements or a mandatory audit trail? This can be seen, for example, with keeping an audit log of financial changes, where documentation and accountability weigh more heavily than speed.
Volume and repetition. Does the task recur often in a similar form, or is every case different? High volume with little variation, as seen with managing periodic subscription billing, makes transfer more attractive, because the investment in a good setup pays off more often.
Two companies may have set up a task with the same name in very different ways. One company works with fixed templates, a limited number of customer types and a system that already provides much structure. The other company does the same task, but with custom arrangements per customer, exceptions handled by phone and a history of incidents that calls for extra control. The eight axes then score differently, and so the outcome differs. With exchanging digital invoices via e-invoicing this plays out strongly: the structured format of e-invoicing pushes toward transfer, but deviating contract terms per customer can pull that same task back into the middle category.
The outcome therefore says something about a task as it is carried out at your company, not about the task name in general. An indication that fits another company in the same sector may not apply to you.
The estimate is only as good as the information we have about the task. For tasks with a lot of judgment scope, varying customer situations or thin or unclear regulation, the range is wide because the actual spread between companies is large, not because we calculate carelessly. A wide range is then a fair outcome, not a shortcoming. More on how you should and should not read such an outcome can be found at what a range does and does not say.
The method also measures tasks, not people or roles. We calculate in freed-up hours and FTE capacity, and explain why on the page why we calculate in hours and not in people. An outcome per task says nothing about who carries out that task today or should carry it out. What an employer does with that information falls outside our method. If an outcome is used in the context of a reorganization or dismissal, separate legal requirements apply for that, which this page does not provide for.
Finally: an outcome is a snapshot. Regulations change, systems are adjusted, and what is human work today may no longer be so in a year. The eight axes stay the same, but how a task scores on them shifts over time.
If you want to know how this plays out for your situation, the free quickscan from ftetoai is a first step: twelve questions, no account needed, with an indication of what part of the hours in your profile can be taken over by AI today. The full work scan, which lays tasks out in more detail along the eight axes, is still under construction. We therefore do not yet offer it, and would rather say so honestly than promise something that does not yet exist.
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.