The problem wasn’t screening — it was losing candidates
The client came to us with what looked like a screening problem: too many applications, not enough time to read them. But when we mapped the actual workflow, the real issue was simpler and worse — resumes were arriving across three channels (a job board, a careers page form, and direct email forwards), landing in different inboxes, and nobody owned the job of making sure every one of them got looked at.
Some candidates were being screened within a day. Others sat for two weeks and got a rejection nobody could explain, because the resume had landed in a folder someone only checked on Fridays. The fix wasn’t a smarter AI model — it was a workflow that guaranteed every resume hit the same pipeline, no matter where it came from.
What we built
The n8n workflow normalizes intake first: job board webhook, form submission, and forwarded email all get parsed into the same shape — candidate, role applied for, resume file, cover note if any — before anything else happens. That normalization step is what actually solved the “lost candidate” problem; scoring only matters once nothing can fall through a channel-specific gap.
From there, each resume is parsed and scored against the specific job requirements for the role it was submitted against, not a generic rubric. The score, a short rationale, and the original resume all land in one place — a lightweight internal queue, not a new tool the hiring team has to learn — with a visible status so nobody has to ask “did anyone see this yet.”
Where the AI actually helps, and where it doesn’t
The scoring step is the only part of this build that uses an LLM. It reads the resume against the role’s stated requirements and produces a score with a short written reason — not a black-box number. That rationale is what made hiring managers trust it enough to actually use it as a first pass instead of re-reading everything themselves.
Everything else — intake normalization, routing, status tracking, notifying the hiring manager when a strong candidate comes in — is plain n8n logic. No AI needed, and no AI wanted; those steps need to be predictable and fast, not judgment calls.
What it replaced
Before this build, screening a role meant one person manually checking three inboxes, roughly, whenever they had time. After: every application lands in one queue within minutes of arriving, pre-scored, with nothing dependent on someone remembering to check a folder.
Total build time was just under two weeks, including a short revision pass after the hiring team saw the first version and asked for the rationale text to be shown up front instead of behind a click.