From XLOOKUP to Semantic Matching

> The Client

A global language technology company with 10,000+ target accounts across DACH, Benelux, and Nordics.

> The Challenge

The field marketing team ran 8-12 events per quarter across Europe. After each event, an analyst spent 4+ hours matching attendee lists against a Salesforce target account list of 10,000+ companies. The process used XLOOKUP in Excel, which failed on spelling variations ("Deutsche Telekom" vs "Dt. Telekom AG"), legal form differences ("GmbH" vs "Inc" vs "Ltd"), and regional naming conventions. Match accuracy hovered around 70%, which meant 30% of high-value attendees were either missed or incorrectly matched. The team had accepted this as the cost of doing business.

> The Approach

Built a semantic matching engine powered by Claude AI (Sonnet). The system accepts two CSV files, lets the user map columns through a visual interface, then runs parallel batch matching. Instead of exact string comparison, the AI understands that "SAP SE" and "SAP Deutschland" are the same company, that "Dt. Bahn" is "Deutsche Bahn AG," and that "McKinsey" matches "McKinsey & Company, Inc." The matching runs in batches of 40 across 3 parallel streams for throughput. Results export as matched-only CSV with BOM encoding for German umlauts. Deployed to Kubernetes for the field marketing team to use independently.

> Methodology Phases Applied

> The Results

Processing time dropped from 4+ hours to under 90 seconds per event. Match accuracy increased from approximately 70% to 99.2%. The field marketing team reclaimed 20+ hours per month. The system handles edge cases that no amount of XLOOKUP formulas could cover: abbreviations, legal entity suffixes, language variations, and regional naming. The analyst who previously owned this process now focuses on campaign strategy instead of spreadsheet wrestling.