Custom machine learning bidding that reduced costs by 54% and improved ROAS by 81%
Ilja Zonov
TL;DR Snapshot
Replaced a basic rule-based CPC bidding script with a custom machine learning model for a complex campaign structure. The ML system predicted optimal bids at granular levels, delivering seven-figure annual cost savings while maintaining performance.
-54%
Cost
81%
ROAS
Client
Unforgettable travel experiences
GetYourGuide is a leading global online marketplace to discover and book experiences worth traveling for. Travelers can use GetYourGuide to find things to do in more than 12,000 cities, including tours from local experts, exclusive access to must-see attractions as well as immersive bucket-list experiences through its Originals by GetYourGuide offering.
Goal
Improve efficiency through custom ML-based bid optimization
Timeline
A/B test implementation with performance validation
Challenge
Complex campaign structure made Google's Smart Bidding feature ineffective
Existing rule-based CPC script provided basic optimization only
Need to improve efficiency without relying on Google's automated solutions
Required granular bid predictions at the lowest unit level
What we did
Phase 1
Collaboration
Worked with internal data science team to develop predictive ML model
Phase 2
Test setup
Implemented controlled A/B test comparing ML model vs existing script
Phase 3
Implementation
Applied ML model to calculate the bids on most granular campaign structure level
Phase 4
Monitoring
Tracked cost, net revenue, and ROAS metrics during validation phase
Key results
-54%
Cost reduction
81%
ROAS increase
$1M+
Annual cost reduction
Maintained business-critical performance levels
Scalable, automated optimization system implemented