An administrative employee or management assistant who fills in declarations, contracts and letters daily based on a fixed template does work that at first glance looks very suitable for AI. It is repetitive, it follows a fixed pattern, and the input usually already comes from a system. Yet the answer is not simply yes. It is partial, and which part depends on how the template and source data are set up in a specific company.
Three axes determine the outcome here: structure, volume and error costs.
The structure is high. A template for an employment statement or a standard contract has fixed fields in fixed places: name, date, amount, position. This is exactly the kind of text work that AI can handle well today, because the structure of the document is already established and the task comes down to correctly filling in fields based on available data.
The volume is also favourable. Companies that produce many of the same documents, a staffing agency that draws up hundreds of employment contracts per month, a municipality that issues the same statement time and again, benefit more from streamlining this process than a company that draws up such a document a handful of times a year. At low volume, the time it takes to set up the process is often disproportionate to the time it ultimately saves.
The error costs are in the middle, and this is precisely why this is not a full takeover, but a task with human oversight. An incorrectly filled-in amount in a contract, a swapped name in a statement, an incorrect date, these are not cosmetic errors. They can lead to legal ambiguity or to a document that needs to be redone. That is why a check remains necessary: AI fills in, an employee approves or rejects, with reason.
The scope for judgement is low, which makes the task more suitable for automation: little interpretation is needed about what should be in the document, that is already fixed in the template and the source data. Creativity is virtually absent, and that too works in AI's favour: nothing is invented, it is filled in. Compliance plays a role as soon as it concerns contracts or official statements, which means that established approval steps remain relevant, regardless of who draws up the document.
A management assistant draws up ten to twenty standard letters per month: confirmations, statements, short contract amendments. The data is already in the personnel system. Here the step from data to document can be automated well: AI can draw up the letter based on the template and the source data, after which the assistant reads, checks and sends it. The time that is freed up does not go towards typing text but towards assessing it.
At another company, where every contract is negotiated on a case-by-case basis and every letter must differ slightly in tone and content, this is different. There the structure is lower, the scope for judgement higher, and the task shifts towards human work. The same type of document, employment contract, letter, statement, can therefore be largely automatable in one company and remain mainly manual work in another. The difference is not in the document type but in how standardised the template is and whether the source data is already in a system somewhere, or still needs to be looked up or requested.
The shift is not located at some future moment when this is "finished". It lies now already in the question of whether an organisation has standardised its templates and has its source data in order. A company where every employee keeps their own version of a contract, with their own formatting and their own variants, cannot simply hand this process over to AI, not even partly. A company with one established template per document type and data that is already kept in a structured way is in a different position.
This document-filling work is often connected to other administrative tasks that take place in the same office. Those looking at how incoming paper archives are digitised and made searchable often see the same question return: is the data already structured, or does it first need to be extracted from paper. The sorting and distribution of incoming post also touches on this subject, because many standard letters are a response to something that comes in. And the way a general email inbox is triaged often determines which documents need to be drawn up at all, and based on which data.
We make no statement about the personnel consequences of this shift: if freed-up hours give rise to a decision about personnel, separate statutory requirements apply to that, independent of this analysis.
Whether this affects a small or a substantial part of your organisation's administrative hours depends on how many of your documents already run on fixed templates and how much of the source data is already in a system rather than on paper or spread across separate files. This differs per company, and that is precisely why a generic statement does not suffice here.
The free quickscan of twelve questions, without an account, gives an indication of what part of the hours in this kind of document work can be taken over by AI today. The full work scan, which breaks down the work of an entire company into tasks and calculates each task in terms of FTE capacity, is still under construction.
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.