Enterprise-scale mapping and search platform handling billions of requests per day across multiple regions and languages.
Industry
Mapsearch Technology
Year
2024
Services
Information Retrieval
Search Relevance
Machine Learning
Experimentation
Improving Global Search Relevance
The Challenge
The client faced challenges related to search relevance and quality across multiple regions and languages.
Improving relevance required more than incremental model updates. The existing ML workflows made it difficult to debug model behaviour effectively, evaluate improvements consistently, and safely deploy changes across regions.
Massive scale of the system handling billions of search requests per day
Difficulty debugging model behaviour effectively
Inconsistent evaluation of improvements
High risk in deploying changes across regions
Fragmented system knowledge and incomplete documentation
Ownership spread across multiple teams and time zones
SK Good Tech AB's Contribution
SK Good Tech AB worked hands-on to improve the end-to-end ML workflow, initially targeting European markets and later expanding to global locales.
Enhanced the model training pipeline to support faster and more reliable iteration
Improved evaluation and debugging tooling to better understand model performance and failure modes
Developed and refined NLP models for query understanding
Improved ranking and recommendation models to increase search relevance
Enabled robust A/B testing and experimentation to validate changes before deployment
Supported multilingual and region-specific search behaviour
Working Within a Large Enterprise
The engagement took place inside a large, distributed organisation where system knowledge was fragmented, documentation incomplete, and ownership spread across teams and time zones.
SK Good Tech AB navigated organisational complexity to identify system ownership and technical dependencies, using strong communication and stakeholder management skills to maintain delivery momentum.
Collaborated effectively with teams across Europe and North America
Navigated complex enterprise environment to unblock progress
Identified system ownership and technical dependencies across distributed teams
Scale & Impact
Improvements were rolled out across major global markets, achieving sustained search quality improvements validated through controlled experiments within systems handling billions of requests per day.
Improvements rolled out across 8 major global markets
Achieved ~2% sustained improvement in search quality over time
Impacted millions of users worldwide
All changes validated through controlled experiments
Deployed within systems handling billions of requests per day
