The starting point
The trading firm opened junior roles that attract thousands of applicants. A single role could bring in more than a thousand applications in a month. The recruiter went through them by hand: CV, form answers, LinkedIn. She could not keep up, so further roles stayed closed.
The problem
- Volume. Plenty of applications, one recruiter.
- The same round on repeat. The same criteria, the same steps, moving people through the system by hand.
- No AI judging people. The EU AI Act classes recruitment systems as high-risk. If a model decided about candidates, the firm could not explain why someone dropped out.
The solution: a script that AI wrote
We used AI to write the code. AI plays no part in processing the candidates themselves. The script is deterministic: the same input always gives the same result.
It runs once a day in Google Cloud. It connects to the applicant tracking system (ATS) through its API and goes through about 1,500 active candidates. For each one it checks the rules the firm wrote:
- Does the candidate meet the basic conditions? If someone needs a visa the firm cannot arrange, for example, the script sends a rejection.
- Should they move on? If they meet the criteria the hiring manager set, the script invites them into the process and sends them a test.
- Did they complete the test? The script waits one to two weeks. Those who submitted it go to the next round. Those who did not get a rejection.
The team does the interviews. The script prepares the queue and sends the emails the rules call for.
We tried n8n and Make first. Now that AI writes the code, those tools added little. Through the API you can do anything with a candidate that the ATS allows: add a note, change the stage, send an email, schedule an interview.
What changed
- Candidates move forward every day. The recruiter no longer works through the queue by hand.
- More junior roles. The team can open them, because applications no longer sit waiting.
- Every decision can be explained. Behind it is a rule the firm wrote, not a model’s guess.
What transfers to your company
- AI does not have to sit inside a process to automate it. It writes the code and steps out. In hiring, that is a safer route than a model that judges people.
- Rules before code. Until the hiring manager writes down what they are looking for, there is nothing to automate.
- Junior roles often fail on volume, not appetite. Once a script works through the queue, the firm can open them again.
How to bring this into your company
If you want something similar in your company, let us start with a 15-minute call. We go through whether it makes sense for you and which of our workshops is the best starting point. More examples are in case studies.