Supplier Sourcing & Approval Agent
A production n8n system that only researches a new supplier after a human approves it, runs a fully event-driven pipeline for that one candidate, and backs the whole queue with a scheduled recovery path that surfaces stalled or missed approvals instead of relying only on the primary execution path.
Human approval, not polling
A status change to approved is what starts research, invoked exactly once per candidate.
Approval to research to discovery
Approved candidates are researched immediately; a weekly agent proposes new ones to review.
One recovery job, no duplicates
A daily safety net recovers at most one missed run and only starts new work if nothing is already processing.
Manual supplier research falls behind, duplicates effort, or stalls silently when a step gets missed.
Replaces open-ended manual supplier research with a queue that only spends research effort on approved candidates, surfaces approvals a missed run would otherwise leave stalled, and keeps proposing new vetted candidates without duplicate work.
- Webhook-first approval sync: marking a candidate approved in the source-of-truth record triggers immediate, exact-match research for that one supplier
- Event-driven research agent invoked per approved candidate instead of continuously polling for changes
- Daily recovery orchestrator that recovers at most one missed approval and starts at most one new research job, and only if nothing is already mid-run
- Weekly discovery agent that learns from prior researched, contacted, and rejected records and proposes new deduplicated candidates above a minimum score
- Every stage writes back to the same tracked record so approval, research, and outreach status stay in sync end to end
From input to a reviewable next action.
- 01
Human approves a candidate
A person marks a sourced candidate's status as approved in the tracked record.
- 02
Event-triggered research
The approval webhook invokes research for that exact supplier, with no polling delay.
- 03
Record update
Research results are written back to the same tracked record for review.
- 04
Daily recovery check
Once a day, the system recovers at most one missed approval if the live path failed to fire, only when nothing is already in progress.
- 05
Weekly discovery
A separate agent reviews prior outcomes and proposes new deduplicated candidates above a minimum score.
Turning open-ended supplier research into an approval-gated queue
The starting problem wasn't a lack of candidate suppliers, it was that vetting them was unbounded manual work: every new name meant opening a browser, cross-checking claims, and manually noting the outcome somewhere that quickly went stale.
The automation only spends research effort on a candidate once a person has explicitly approved it, which keeps the system from ever doing speculative work nobody asked for. Research is invoked per candidate over a webhook rather than on a polling loop, so approvals are acted on immediately.
A bounded daily recovery job exists only to catch the rare missed event, not to duplicate the live path. It processes at most one missed approval and only starts new work if nothing is already mid-run, which was a deliberate choice to prevent two research jobs from ever colliding on the same candidate.
A separate weekly agent closes the loop by learning from what was already researched, contacted, or rejected, proposing a small, deduplicated batch of new candidates above a score threshold instead of an unbounded list.
Visible proof, without private operational data.
Select a step to see what happens there.
A person marks a sourced candidate's status as approved in the tracked record.
Visual documentation is being prepared. The case study remains complete and public-safe.
Approval trigger, research pipeline, recovery safety net, and discovery loop
The pipeline treats approval, research, recovery, and discovery as separate, independently inspectable responsibilities instead of one continuously polling process.
- 1Watch for a candidate's status changing to approved in the source-of-truth record.
- 2Trigger research for that exact candidate over a webhook, with no scheduled polling on the live path.
- 3Write research results back to the same tracked record for human review.
- 4Run a bounded daily recovery pass that checks for at most one missed approval and starts at most one new job, only if nothing is already processing.
- 5Run a weekly discovery pass that learns from researched, contacted, and rejected history and proposes new deduplicated candidates above a minimum score.
- 6Keep every write scoped to a single record so two stages can never overwrite each other's result.
How the approval-to-research pipeline is controlled
A public-safe view of the trigger, research pipeline, recovery safety net, and discovery loop.
Approval trigger
Research begins only when a human changes a candidate's status to approved. The exact record schema and product category are intentionally excluded from the public case study.
Event-driven research
Because the trigger is a webhook rather than a polling schedule, an approved candidate is researched immediately instead of waiting for the next scheduled pass.
Recovery safety net
A daily job at a fixed early-morning time recovers at most one missed approval and starts at most one new research job, and only when no job is already processing, so it can never duplicate live work.
Weekly discovery
A separate weekly agent reviews prior researched, contacted, and rejected records, then proposes a small number of new deduplicated candidates above a minimum preliminary score.
Human checkpoint
Every research result, discovered candidate, and outreach decision is written back for human review. The workflow does not contact a supplier on its own.
What this workflow demonstrates
Approval gates real cost
Research effort is only spent on candidates a human explicitly approved, not on every name that enters the pipeline.
A safety net that can't collide
The recovery job is deliberately bounded (max one recovery, max one new job, only when idle) so it backs up the live path without ever racing it.
Discovery avoids repeat work
The weekly agent dedupes against prior outcomes so the same candidate is never re-proposed after it was already researched or rejected.
Vetting suppliers or vendors manually?
Share how candidates get approved today and I will map where an event-driven, approval-gated workflow removes the manual research backlog.
