Supplier Discovery AI Agent
Turned a 4-hour process into 10 minutes for 100+ procurement teams.
The Problem
Sourcing was broken. Buyers and category managers lost 4+ hours per sourcing cycle searching through fragmented supplier data, manually scoring candidates, and writing personalized outreach emails. Slower cycles. Higher maverick spend. Frustrated teams doing work that felt stuck in the past.
What I Built
An LLM-powered system that searches across fragmented data sources at once. Supplier databases, web data, risk platforms, ESG registries. It reads a category brief, scores candidates against risk and ESG signals, ranks them, generates contextualized outreach emails tailored to each supplier, and delivers a shortlist with reasoning behind every recommendation. Buyers stay in control. Nothing ships without review. The technical foundation: multi-source retrieval pipelines, guardrails, human-in-the-loop checkpoints, escalation logic that actually works.
The Impact
97% time reduction. From 4 hours to under 10 minutes per sourcing cycle. Started as a 3-team pilot. Expanded to 100+ procurement users who made it their default tool. With 100+ users saving 3.5+ hours per cycle, that's roughly 350+ hours of procurement time freed up per sourcing round. Buyer satisfaction: 4.6/5 across the cohort.
My Role
Product lead. User research, PRD, evaluation framework, pilot management, full rollout.
What I Learned
Model choice didn't matter as much as I thought. Evaluation frameworks and clear escalation paths drove adoption. Opinionated defaults beat flexibility every time. Buyers wanted a tool that just worked, not a blank chat to figure out.
Built With
- OpenAI
- Copilot Studio
- Power Automate
- N8N
Try the agent
Chat with a live version of the Supplier Discovery agent, wired to the real n8n workflow behind the case study.