CS / 02 · AI Internal Tool

RAG + HITL workflow.

Helped uncover $1M+ in procurement savings opportunities through AI workflow capture across ERP-adjacent purchasing processes, then operationalized a SharePoint-backed RAG + HITL workflow that reduced manual work 60% & generated $85K in verified savings.

Role
Sr. PM / AI Lead
Scope
Procurement · ERP-adjacent
Timeline
2024 — 2025
$1M+
Savings opportunities
−60%
Manual work cut
$85K
Verified savings
435+
Queries processed
01 / Context

What was broken.

Problem

The starting state.

  • Buyers spent hours per day chasing live lead times, pricing, and inventory across 4+ disconnected systems.
  • Critical procurement decisions made on stale spreadsheets — corrected days later when reality caught up.
  • ERP updates lagged real-world supplier conversations by 2-5 business days, distorting planning downstream.
Approach

What I built.

  • Indexed authoritative procurement data into a SharePoint-backed RAG knowledge base — single query surface for buyers and planners.
  • Built a human-in-the-loop approval workflow: AI drafts the ERP-ready update, a buyer approves or edits, the system commits.
  • Instrumented department-level adoption from day one — Engineering led, Manufacturing and Product followed.
  • Tight feedback loop: every HITL correction trained the next iteration of prompts and retrieval scoring.
02 / Mechanism

How it actually worked.

RAG UI · WORKDONE

The tool teams actually use.

WORKDONE Corporate Memory CM Copilot interface

Single query surface for buyers and planners. Authoritative SharePoint docs, embedded and retrieved on demand.

Adoption · 13 departments

Where adoption landed.

Engineering
213
Manufact.
78
Product
52
Operations
34
Marketing
12

435+ queries processed. Engineering led, Manufacturing and Product followed.

03 / Outcome

The numbers that mattered.

Headline result
$1M+

Procurement savings opportunities surfaced through AI workflow capture across ERP-adjacent processes. $85K already verified · 60% manual work cut.

04 / Takeaway

The fastest AI ROI is boring infrastructure work.

No model fine-tuning. No agent orchestration. Just clean retrieval over the right docs and a human gate on the write path — $1M+ in savings opportunities surfaced, $85K verified, 60% of manual work gone.