Following a strategic acquisition of a cross-border company, a leading nordic retailer partnered with Apex Digital to integrate the new product catalog. We developed an intelligent, multi-step AI process that accurately mapped over 74,000 different language products to the client’s category structure, achieving over 90% high-confidence matches and accelerating post-merger integration.

Unifying systems after a cross-border acquisition

Following the acquisition of a cross-border competitor, our client, a major nordic retailer with a large item catalog, faced a significant integration hurdle. To realize the full value of the acquisition and operate under a single, unified system, they needed to map the entire acquired product catalog (74,000 products) to their existing category structure. The challenge was twofold: the sheer volume of products made manual mapping infeasible, and the language barrier between acquired product data and the client’s category tree added immense complexity.

A small portion of products could be matched using EAN codes, but the vast majority required a deep, contextual understanding of the product descriptions to place them correctly. The client needed a solution that was not only fast and accurate but also cost-effective, ensuring the merger’s synergies could be realized without a long, expensive data integration project.

A Multi-tiered AI mapping engine

Apex Digital designed and implemented a sophisticated AI process built on Large Language Models (LLMs) and a vector database. Our approach was not one-size-fits-all; it was an intelligent, multi-tiered strategy designed for maximum accuracy and efficiency. Before full deployment, we validated and fine-tuned our models using the EAN-matched products as a ground-truth dataset, ensuring the highest possible performance.

The final, automated workflow involved three key stages:

1. High-speed initial pass: We first processed the entire product list using a fast and cost-effective LLM to map the majority of items quickly.
2. High-power re-analysis: Products that received a low confidence score from the initial pass were automatically re-processed by a more powerful, advanced LLM for deeper contextual analysis.
3. Group consistency check: Finally, the AI re-analyzed products in their newly assigned category groups to identify and correct any remaining inconsistencies, further improving the overall accuracy.

This tiered approach ensured that we used the most powerful (and more expensive) AI resources only where they were needed most, delivering an optimal balance of speed, accuracy, and cost.

Accelerated integration and unprecedented accuracy

The AI-powered mapping solution delivered exceptional results, allowing the client to move forward with their system integration far ahead of schedule. The project was a clear success, turning a potential months-long manual task into a swift, automated process.

The key performance indicators highlight the value delivered:

• Accuracy: 90% of products were categorized with high confidence, a score over 90%.
• Reduced manual work: Only about 2% of products had a confidence score below 50%, meaning the team’s manual review effort was reduced to a tiny, manageable fraction of the catalog.
• Cost and time efficiency: The automated, multi-tiered process was completed at a low cost and in a fraction of the time required for traditional methods, directly accelerating the business benefits of the acquisition.

The client was extremely pleased with the outcome, which not only solved the immediate integration challenge but also demonstrated a powerful new capability for managing complex, multi-lingual data in the future.