The Nederlandse Arbeidsinspectie's Monitor Arbeidsongevallen 2024, published on 2 October 2025, contains a number worth sitting with. In 18 percent of the investigated accidents the victim was a temp worker, while temp workers make up roughly 4 percent of the working population. In industry it is worse: nearly 25 percent of victims, against an estimated 8 percent of that sector's workforce. The underlying dataset is 1,990 investigations, 2,001 victims and 58 fatal accidents.
Two honest qualifications. These are investigated accidents, not all accidents. And the comparison is against a workforce share, which is an estimate. Neither qualification closes the distance between 18 and 4.
The instinctive explanation, that temp workers take more risks, does not survive contact with the floor. The people in that statistic are new, on a site they have never seen, often working in a language that is not the language of the manual, and expected to produce from the first hour. What failed is the onboarding system around them. It was built for someone who starts on a Monday and stays five years. Nobody redesigned it for someone who starts on a Tuesday afternoon and may be gone in three weeks. This guide is about that redesign.
Diagnosis one: the induction assumes a cohort
Most site inductions were designed around an intake: ten people, a room, a set date, a trainer, a couple of hours. That format is not wrong. It simply describes a group that no longer arrives.
What arrives instead is one person on Tuesday at 13:40, two more on Thursday for a peak lasting nine days, someone covering a sick colleague from 06:00 tomorrow. The induction still exists, so it moves to the next scheduled date, which may be after the worker has left. Meanwhile the person is on the floor, and the gap between arriving and being instructed is the window the accident statistics describe.
The tell is easy: if your induction is a calendar event rather than a state a person is in, you have this problem.
Diagnosis two: two employers, one worker, and an assumption
The second failure is that nobody is quite sure whose job it is. Both countries answer clearly, and in both the answer is split.
In Belgium the instruction on the floor is the gebruiker's job, not the agency office's. Codex art. X.2-11 puts it on the user, or the designated member of the hierarchical line, to give every temp worker the risks of the workpost, the safety instructions specific to it, and training in line with art. I.2-21. The werkpostfiche (workstation record sheet) is drawn up by the user under art. X.2-3, and only for a workpost or function subject to mandatory health surveillance, not for every assignment. The agency, which remains the legal employer, completes sections C and D, keeps a copy and hands one to the worker, under art. X.2-4 and X.2-5. Art. X.2-17 then binds both ends: where a fiche is required, user and agency are both forbidden from deploying a temp worker at that workpost without one and without having informed the worker. The consequence to quote internally is not a fine, it is that the person may not start.
In the Netherlands it is a relay. Arbowet art. 5 lid 5, the doorgeleidingsplicht, runs inlener to uitlener to worker: the inlener supplies the RI&E description of hazards and risk limiting measures in good time before the work starts, so the uitlener can pass it on. Waadi art. 11 obliges the agency to do exactly that before the assignment begins. The CAO voor Uitzendkrachten 2026-2028, article 19, is the plainest wording of all: the agency must inform the temp worker, before the work at the client starts, about the required qualifications, the possible safety and health risks, and how to deal with them.
Read those together and the failure mode is obvious. Every duty is real, but each depends on the other party having already done something. The client assumes the agency briefed them. The agency assumes the client will instruct on arrival, which is legally correct in Belgium but only happens if someone on site is named for it. Nobody is wrong, and the worker starts uninstructed.
Diagnosis three: the window is minutes
The third problem is arithmetic. The realistic window between a temp worker arriving and being needed on the line is fifteen or twenty minutes, at the entrance, with a shift leader who is already short-handed. Onboarding designed for longer than that window will not be delivered late. It will not be delivered at all. Designing for the window you have is not a compromise, it is the only version that reaches everybody.
The playbook
Six moves, none of which needs a training department.
Separate site-generic from workstation-specific. Two contents, two lifespans. Site-generic is the traffic plan, the gates, the alarm, the assembly point, the PPE rules, who to report to: it changes rarely and applies to everyone at that address. Workstation-specific is the machine, the substances, the guard that must stay closed, the manual handling in that cell. It changes with every assignment and it is the part that prevents the accident. Most inductions merge the two, which makes them too long to deliver and too generic to be useful.
Decide who owns each piece, and write it down. One page per client site: who delivers the site-generic block, who delivers the workstation block, who holds the record, who checks before the badge is issued. In Belgium name the member of the hierarchical line, because X.2-11 points at a person, not a department. In the Netherlands name who prepares the art. 5 lid 5 description and who sends it. An unwritten split is the assumption above waiting to happen.
Get the workstation risk description to the agency before the worker arrives. The highest-value change on this list, and it costs a process, not money. If the RI&E extract, or the werkpostfiche content where a fiche is required, reaches the agency at booking rather than on the first morning, the agency can deliver its article 19 and Waadi art. 11 duty properly, and the worker turns up with the risks already in their head. Day-one instruction then confirms and localises rather than starting from zero.
Make the instruction short enough to happen, and build it per site rather than per person. Ten minutes of site-generic and five minutes of workstation content, on a phone at the entrance, gets done. Ninety minutes in a room does not. Build it once per site: the same block serves every temp worker who ever goes to that address, which turns onboarding from a repeated effort into an asset you maintain.
Record per person and per site, not per session. This is where clients and agencies are exposed. A worker doing three assignments at three client sites in a month needs three records, because what they were instructed on differs each time, and a session-based record cannot express that. Dutch law prescribes no form of evidence, but without a per-person record you cannot demonstrate the art. 8 duties were met. In Belgium it is firmer: art. I.2-11 requires a member of the hierarchical line to sign a document per starting worker confirming the necessary information and instructions were given.
Build week two. Almost all temp onboarding stops at the gate. The first days are when someone copies what the person next to them does; the second week is when they work unsupervised at speed. A short follow-up in week two, on what went wrong or nearly went wrong, is cheap and is the piece nobody plans.
What a certificate does and does not tell you
If you are a client relying on an agency's paperwork, be precise about what it covers. The SNA keurmerk and the NEN 4400-1 behind it are fiscal and administrative. They say nothing about safety instruction, and treating them as safety assurance is a common and expensive misreading.
VCU is different. It is the certification scheme for agencies placing people into higher-risk work, not a legal obligation but a client requirement, and it audits the office's procedures for selection, instruction, follow-up and evaluation of the workers it places. That is genuinely relevant. A VCU certificate is evidence the agency has a process. It is still not evidence that this worker was instructed for this workstation.
Where Aristotl fits
Aristotl builds both blocks from documents you already have: the site rules, the RI&E extract, the werkpostfiche, the machine instruction. Modules run three to five minutes on the worker's own phone, no laptop and no app installation, which is what makes delivery at the entrance realistic. Courses are assigned per client site, so someone going to plant A gets plant A, and each is translated per worker, usually the difference between reading it and nodding at it. Knowledge checks show it landed. Completion is registered per person, per site and per version, and exports, so the worker with three assignments in a month has three records without anyone keeping a spreadsheet.
Bring one client site's rules and one werkpostfiche to a demo, and we will show what a temp worker starting tomorrow at 13:40 would receive, and what you would hold afterwards.