How we arrive at a range

Why not a single figure

A statement like "70% of this task can be taken over by AI" looks precise, but that precision is usually an illusion. Behind such a figure lies an estimate of structuredness, repetition, cost of error and a number of other characteristics that differ per organisation. If we do not show that uncertainty, we create a false sense of exactness. That is why we work with a range and a certainty label instead of a single figure.

That label has direct consequences for how you read an outcome. A task that comes out at "40-55%, reasonably certain" is quite different from a task that comes out at "20-70%, low certainty". In the second case, the outcome mainly says that the task, as it is currently described, is too loosely defined to build a judgement on. That is information in itself.

The eight axes that determine the range

We assess every task along eight axes:

Structuredness of the input — is the input predictable and unambiguous, or does the form keep changing?
Scope for judgement — does it require weighing things up, or is there a fixed answer?
Sensitivity of customer contact — how directly does the task affect the relationship with the customer or employee?
Physical component — is an action in the physical world required?
Creativity — does the task require coming up with something new, or applying something existing?
Cost of error — what does a mistake cost, in money, time or trust?
Compliance sensitivity — is the task laid down in laws or regulations that must be followed precisely?
Volume/repetition — how often does the task recur in the same form?

Each axis is given a score based on the way you describe the task. The more unambiguous the input on all axes, the narrower the range can be. The more axes that score doubtfully, the wider the margin and the lower the certainty label. This is also why can AI take over keeping track of leave balances shows a narrower range than a task with many exceptions: keeping track of leave balances scores high on structuredness and volume, and low on scope for judgement.

Why the same task still turns out differently

The eight axes provide a framework, but the input is up to you. Two organisations that both submit "recording hours" as a task can get very different outcomes. At one organisation the recording runs through a fixed system with fixed categories, at the other through loose notes that are interpreted afterwards. See also can AI take over recording hours and the explanation of why the same work turns out differently for you. The range is therefore not a property of the task name, but of the task as it is carried out at your organisation.

What low certainty means — and what it does not

A wide range or a low certainty label is not a fault in the method. It is the answer to a question that has not itself been posed sharply enough yet. It often means that the task description combines several types of work that would separately give a very different outcome. Consider can AI take over answering employees' HR questions: under one task name fall both simple, factual questions and questions that require personal customisation and judgement. As long as these two are not described separately, the range remains wide, and that is then a correct outcome, not an inaccurate one.

Low certainty says nothing about whether AI could never take over the task. It says that the current description offers too few points of reference to say anything meaningful about it right now. More detail in the input — for example about what proportion of the work consists of exceptions — often leads to a narrower margin.

What a range is not

A range is not a personnel decision and not advice on what you should do with a role or a team. It is a description of a task at a given moment, based on the data entered at that moment. If an outcome touches on decisions about roles or staffing levels, separate statutory requirements apply that this method does not provide for.

The outcome is also not a fixed fact for the future. Tasks change, systems change, and with that the score on the eight axes changes. What the range was at the time is not what it will be a year from now. On what a re-measurement shows you can read how a repeated measurement shows the difference.

What you can do now

The free quickscan from ftetoai consists of twelve questions, without an account, and gives an indication of what proportion of the hours in your profile can be taken over by AI today, with the corresponding range and certainty. The full work scan, which examines tasks at a detailed level along the eight axes, is still under construction. We therefore do not yet offer it; what is available today is the quickscan as a first, provisional indication.