By Sergiy Troitskyi in Solutions —

AI Product onboarding

Delhaize faced challenges onboarding product data from suppliers' CSV files due to inconsistent formats and missing information. Initially using rule-based scripts, we transitioned to an AI solution that normalized data, identified columns, and enriched product info.

Problem Statement

Delhaize, a major retailer, faced challenges in onboarding product data from numerous suppliers. Suppliers provided product information in CSV spreadsheets containing details such as prices, images, quantities, delivery dates, and other attributes. The key issues were:

  • The data often did not conform to the schema of Delhaize's warehouse database.
  • Missing or incomplete data needed to be cleaned and normalized.
  • As the number of suppliers grew, maintaining manual scripts for data transformation became increasingly cumbersome.

Initial Solution: Rule-Based Scripts

To address the initial problem, the team developed rule-based scripts that relied on if-else conditions. These scripts:

  • Analyzed the content of the CSV tables.
  • Transformed and normalized the data according to predefined rules.
  • Identified missing data and filled it where possible using basic logic or default values.

While effective in the short term, this approach became unsustainable as the number of suppliers grew. Updating and maintaining a growing set of complex conditional rules proved labor-intensive and error-prone.

AI-Powered Solution

To overcome the limitations of rule-based scripts, the team implemented an AI-driven solution with two key components:

  1. AI for Column Identification and Data Normalization:
    • Developed an algorithm capable of automatically analyzing CSV files to infer column meanings (e.g., identifying "price," "quantity," and "delivery date" columns based on their patterns and content).
    • The AI dynamically mapped these columns to Delhaize's warehouse database schema without requiring manual intervention.
    • It applied advanced cleaning techniques to normalize inconsistent formats (e.g., standardizing date formats or currency values).
  2. AI for Image Processing and Enrichment:
    • Leveraged computer vision models to analyze images attached to CSV files.
    • Extracted additional metadata from these images, such as tags for colors, clothing types, or other product attributes.
    • Integrated this enriched data into the database schema to enhance product descriptions and improve inventory management.

Impact

The AI solution significantly streamlined the onboarding process by:

  • Reducing reliance on manual updates for transformation rules.
  • Improving accuracy in data normalization and cleaning.
  • Enabling richer product information through automated image analysis.
  • Scale efficiently as the number of suppliers increases.

This transformation not only improved operational efficiency but also enhanced Delhaize's ability to manage its warehouse inventory with higher-quality data.