Vendor Master Dedupe & Merge Assistant

The problem this kills

Your vendor master is full of the same supplier three times: "ACME", "Acme Inc", and "ACME INC." Each one collects its own purchase orders and invoices, so your spend is split across rows that should be one. Reporting undercounts how much you actually buy from a vendor, you miss volume discounts you've already earned, and payments go to the wrong record - sometimes to a stale bank account on a duplicate nobody cleaned up.

The usual "fix" is one analyst eyeballing a giant spreadsheet, sorting by name, and merging by hand. It's slow, it's error-prone, and the scariest part is that a wrong merge on an active, high-spend vendor can break open purchase orders and payment runs. So nobody wants to touch it, and the mess grows.

What you'll build

A small, login-protected web app that does the boring matching for you and then gets out of the way so a human can decide. You import a vendor-master export, and the tool groups likely duplicates into clusters, each with a confidence score and the total spend at stake. Strong signals like a shared tax ID or bank account are weighted far above a name that merely looks similar.

A data owner opens each cluster, sees the candidates side by side, picks the golden record (the one that survives) and which fields to keep, and approves. Nothing merges until a person clicks approve - and the tool refuses to quietly merge active or high-spend vendors without an explicit extra confirmation. When you're done, it produces a cleaned vendor master CSV and a remap file (old vendor ID to golden ID) you can use to repoint transactions in your real system.

What's inside the Implementation Plan

• It starts by interviewing you about your business. Before writing a line of code, the plan has the AI agent ask about your current dedupe process, your ERP and the exact column names in your export, your vendor-ID and tax-ID formats, your typical and peak vendor counts, your match rules, and your messy edge cases. It reflects a short tailored spec back to you and waits for your thumbs-up - so you get a tool shaped to your data, not a generic template.
• A step-by-step build you complete by pasting ready-made prompts into an AI coding agent.
• A matching engine that clusters on a cluster signature (normalized name + tax ID) and scores confidence by signal strength.
• A reviewer screen that shows spend at stake and flags active-PO / high-spend vendors.
• CSV import (the no-API fallback is the whole point here - pure CSV in, CSV out) and CSV export of the cleaned master plus a transaction remap file.
• Login, row-level security, an approval gate, duplicate guards, and a complete audit log built in.

The governance it includes (this is the point)

• Login so only your team can open the tool.
• Row-level security so each organization only ever sees its own vendor data.
• A human-in-the-loop approval gate: the AI proposes clusters and a golden record; a person reviews, edits, and approves. Nothing is merged automatically.
• A complete audit trail: who imported what, which clusters were approved, which record survived, and exactly when.
• Duplicate guards: each vendor lands in at most one approved cluster, and the same import can't be processed twice.
• Safety rails on risky merges: high-spend and active-PO vendors require an explicit extra confirmation before they can be merged.

Who it's for

Procurement and master-data analysts who own the vendor list and are tired of split spend, payment errors, and "is this the same ACME or a different one?" You don't need to be a developer. If you can export a spreadsheet and follow instructions, you can build this.

You've got this - paste the first prompt and let the interview tailor it to your vendor master.