Case Study 06 ReturnKey
Product Manager Aug 2024 — Sep 2025 Jakarta, ID & Nashville, USA (Remote)

Enriching Product Metadata & Pricing Strategy with a Single Scan

Turning manual sorting through 3 stations and little insights to one scan enrichment that supports smart commercial decision.

EnterpriseAI/MLOperations
Artifacts
01 / 02
Problem

What we were actually solving

Returned and overstock inventory moved through two or three manual sorting stations before it could be re-listed. Even through the steps, we still know very little about the product to make meaningful commercial decision. Moreover, the process is expensive and especially tricky with items that are unlabeled, have obscure barcodes, near-identical variants, and mismatched information between manifest and real items.

Process

How we got there

  1. Step 01
    Enrichment Pipeline

    Co-designed lexical + semantic analysis, web scrapers, and LLM enrichment with CTO and VP of Data.

  2. Step 02
    Scan & Print Workflow

    Single-scan triggers a whole enrichment system that lets processor only need to label the product with the barcode the machine printed.

  3. Step 03
    Pricing Experimentation

    Built a system for markdowns, switchback tests, and A/B pricing experiments of the products based on the information we learn.

Solution

What shipped

One station, one operator: scan an item, the LLM recognizes and categorizes it, the workflow logs and prints. Pricing experiments run continuously underneath so the floor never stops learning what the inventory is worth.

Outcome

The numbers

−40%
processing cost
+10%
net margin