Project snapshot
This case study tells the story of creating a performance advertising platform for the #1 European marketplace. The AI solution enabled the company to improve sales metrics for 128k merchants through fully autonomous campaign optimization and retraining mechanisms.
Client
The largest online marketplace in Central and Eastern Europe with 21 million customers, 70 million monthly item sales, and 128k merchants using their internal advertising solution.
Solution
Mass-model campaign optimization platform with fully automated retraining pipeline, featuring dynamic bid adjustment algorithms and real-time performance monitoring across hundreds of ML models.
Business function
Advertising and campaign optimization
Industry
Retail & E-commerce
Challenge
Create an internal marketplace advertising platform enabling sellers to promote offers with above-market performance while handling volatile traffic through automated model retraining mechanisms.
Result
Client background
The client is the largest online marketplace in Central and Eastern Europe, operating across 10+ countries with an integrated ecosystem serving millions of users. Their platform connects 128k merchants with 21 million customers, facilitating 70 million item sales monthly through comprehensive marketplace, advertising, and data analytics services.
Before implementing the AI-powered optimization platform, the client faced several critical advertising challenges:
The company needed an intelligent advertising platform that could automatically optimize campaigns in real-time, minimize costs, and scale efficiently across their massive merchant network.
The company sought an AI-powered advertising optimization platform to minimize cost-per-click and maximize campaign performance for 128k merchants while handling volatile traffic through fully automated model retraining mechanisms.
Potential threat: If the solution were not implemented, the client would continue facing rising advertising costs, declining merchant satisfaction, and competitive disadvantage in the marketplace advertising space.
Challenge
The client wanted to maximize the number of clicks or minimize CPC.
The target metric for optimization depends on the specific campaign goals, but the main focus is commonly on reducing effective cost-per-click.
The mass-model campaign optimization platform presented challenges in real-time bidding, automated model management, and performance optimization across diverse product categories. Xenoss team created solutions for dynamic pricing, automated retraining, and scalable ML deployment.
Click events were rare compared to non-clicks, creating training challenges for accurate conversion probability prediction models across different merchant campaigns.
Solution: We implemented resampling techniques including SMOTE (Synthetic Minority Oversampling Technique), class weighting in algorithms, and specialized loss functions to balance datasets through over-sampling and under-sampling strategies.
Predictions must be made in milliseconds to participate in RTB auctions, requiring optimization between model accuracy and inference speed for thousands of concurrent campaigns.
Solution: We prioritized latency over accuracy in model selection and tuning, implementing lightweight models that meet strict timing requirements while maintaining sufficient predictive performance for CPC optimization goals.
User consumption behavior varies vastly across different product categories – people buy dog food, consumer electronics, or luxury perfume differently, making a single model ineffective for all campaigns.
Solution: We developed separate models for every user behavior class, classifying all product categories into logical classes by consumption behavior patterns. Classification was based on marketplace taxonomy (Electronics, Baby Items, Mobile devices) and brand taxonomy, later implementing more complex statistically-defined similarity classification.
Categorical variables like user demographics and ad IDs led to high-dimensional feature spaces, creating computational and storage challenges for real-time bidding systems.
Solution: We employed dimensionality reduction techniques like PCA and special encoding methods including target encoding and binary encoding that don’t increase training data dimensionality while preserving predictive power.
User behavior patterns, ad creatives, and competition landscape change over time, causing model performance degradation as training data distributions shift from current market conditions.
Solution: We built automated monitoring for data drift with active learning and periodic retraining processes, including repeatable drift identification, threshold-based alerting, and proactive model replacement mechanisms.
Most campaigns start on merchant requests without manual optimization, requiring fully automated pipeline covering model selection to continuous improvement across hundreds of diverse campaigns with varying durations.
Solution: We created a comprehensive automated pipeline with pre-trained model repository, permanent performance monitoring, gradual model switching (5%, 10%, 50% up to 95-98%), and unlimited lightweight model deployment capability handling thousands of models without infrastructure impact.
Technical implementation
A very high-level performance optimization engine diagram.
The approach is iterative. For example, initially, decision-making rules can be cautious, but as we discover more, we make them more granular and complex.
The solution consists of several logical components:
Model repository
A logical storage of pre-trained, verified models mapped to specific product categories. Only stable and well-performing models will be a permanent part of the repo. We use those models as a starting point for a campaign CPC minimization approach.
Automated model training pipeline
A pipeline automatically trains campaign-specific, ephemeral models. For training, we employ several techniques, including, but not limited to, finetuning the base model, training the model from scratch, cross-validation, hyperparameter search, and so on.
A/B testing block
Allows splitting traffic between multiple ML models at a constant rate.
Monitoring block
Calculates pre-defined metrics, including but not limited to AUC, MSE, and RMSE.
Decision block
Encodes business rules to act based on the performance of specific models as calculated by the monitoring block.
Pre-campaign model
Model trained for a specific campaign.
Main architectural considerations
Permanent model performance monitoring
We permanently and automatically evaluate the model’s performance within each campaign. Once the degradation signs are detected, the re-training or replacement process is automatically triggered. Also, model replacement can be initiated according to the schedule.
Gradual switch to the new model
We switch to the new model gradually (e.g., 5%, 10%, 50%, and up to 95-98%). We always leave some campaign traffic without going through the ML CPC optimization model.
Degradation criteria
It is worth mentioning that model performance is a complicated criterion. It was initially set into empirical values but tuned throughout the system lifecycle per actual performance.
An unlimited number of models
We use lightweight models, so training and deployment are never a bottleneck.
Furthermore, our proposed infrastructure can handle as many models as needed,
allowing us to create thousands of models in fully automated pipelines without any
infrastructure performance impact.
27% CPC reduction
The AI-powered bidding strategy reduced CPC by 27% compared to initial rule-based bidding strategies. It enabled advertisers to acquire more clicks within the same budget.
45% reduction in operational cost
Optimization automation and a fully automated model maintenance cycle significantly reduced the number of campaign management tasks. On average, the new system reduced manual effort by 45%, enabling marketing teams to focus on strategy rather than execution.
18% CTR lift and 9% CR lift
Another effect of historical user behavior analysis is the system identified high-intent audiences and optimized ad placements accordingly. As a result, CTR increased by 18%, ensuring that ads reached users most likely to engage. Furthermore, the AI-powered conversion prediction model improved targeting efficiency, leading to a 9% increase in overall conversion rates.
Faster experimentation &
continuous optimization
The AI optimization solution substantially accelerated A/B testing cycles, allowing for rapid iteration and continuous campaign improvement.
Utilize first-party data
to improve targeting possibilities
With an in-house optimization engine, the marketplace could activate their 1st party audience segments without risk of data leakage.
Want to build your own solution?
Contact us