E-commerce and retail
From 12 supplier formats to one automated data set enriched with AI
Danish e-commerce company
Format: One-day hackathon
In short
- 50,000 products are enriched automatically every day with no manual work
- Supplier data in more than 12 formats is normalised into one shared master data set
- AI enrichment: brand copy, images and translations into four languages
- Three full-time roles freed up for strategic product work
During a one-day hackathon the team built an automated pipeline in n8n that accepts supplier data in any format, normalises it into one shared master data set and enriches every product line with AI: search-optimised brand copy, automatic image improvements and translations into four languages. At the end, the pipeline generates exports tailored to each customer. One distribution partner receives a CSV with 12 fields, another an XML file with 47 fields, a third a JSON export straight into its product database. 50,000 products now run through the pipeline every day without a single manual click.
Results in numbers
The challenge
An established Danish online shop with 50,000 unique products received product data every day from more than 12 suppliers, in everything from Excel sheets and XML files to PDF price lists. Three full-time employees spent their entire working day copying, normalising and enriching product data with brand copy, images and translations before each product could go out to distribution partners, who all had very different requirements for format, language and fields. The company could not keep pace with its growth plans, because every new supplier meant yet another manual process.
The solution
During a one-day hackathon the team built an automated pipeline in n8n that accepts supplier data in any format, normalises it into one shared master data set and enriches every product line with AI: search-optimised brand copy, automatic image improvements and translations into four languages. At the end, the pipeline generates exports tailored to each customer. One distribution partner receives a CSV with 12 fields, another an XML file with 47 fields, a third a JSON export straight into its product database. 50,000 products now run through the pipeline every day without a single manual click.
How we built it
- 1
Mapping sources and formats
The team mapped every supplier and the data fields each one delivered. AI helped spot which fields were consistently missing and which variations actually meant the same thing.
- 2
Normalisation in n8n
We built an n8n workflow that takes input by email, FTP and incoming webhooks and converts everything into one shared JSON format with validation rules. When a supplier changes its format, only one step needs adjusting.
- 3
AI enrichment per product
Claude and ChatGPT generate search-optimised copy, category tags and variant descriptions from the supplier's raw data and the brand guidelines. Every enrichment is logged, so its quality can be spot-checked afterwards.
- 4
The language layer
Automatic translations into Danish, English, German and Swedish, with a consistent brand voice set by fixed templates. Native speakers on the team checked the first 200 products before the full roll-out.
- 5
Export generator per customer
A configurable template engine produces the exact format each distribution partner expects, sends it out on a fixed schedule and raises an alert if a field is missing.
What we learned
- →The biggest gain was not the AI copy but the normalisation step. Every data problem could be traced back to the varying supplier formats.
- →Keep the AI enrichment modular. When a supplier changes its format, you only adjust one step, not the whole pipeline.
- →Spend an extra 30 minutes on logging and alerts. A workflow you can see into is far easier to maintain.
- →Distribution partners love configurable exports. That is often what wins the deal, not the price.
“We used to have three full-time employees maintaining product data. Now it runs automatically, and the quality is higher than ever, because people make mistakes and a well-built workflow does not.”