Skip to main content

Using Reinforcement Learning to Optimize Campaign Management

· 3 min read
Jochen Luithardt
Jochen Luithardt
Co-Founder @ pi-optimal

This case study examines how Stellwerk3, a German advertising agency, revolutionized their campaign management through model-based reinforcement learning. By developing Adpilot, a custom middleware solution integrated with pi_optimal's platform, they achieved 24% CPC and 31% CPM cost reductions while automating 60+ campaigns, demonstrating AI's practical impact in digital advertising.

Key Metrics​

  • CPC (Cost per Click): Reduced by 24%
  • CPM (Cost per Mille): Reduced by 31%
  • Automated Campaigns: Over 60

Client Background​

Stellwerk3, established in Reutlingen, Germany, brings 18 years of digital advertising expertise. Their portfolio includes managing over 12,000 campaigns for 1,500+ clients, specializing in targeted, performance-driven advertising solutions.

The Challenge​

Modern digital advertising faces several complex challenges:

  • Demand for continuous real-time campaign optimization
  • Heavy dependence on expert knowledge and manual intervention
  • Increasing workload pressure on campaign managers
  • Limited scalability of traditional manual processes
  • Rising costs in an increasingly competitive market

Technical Solution​

Stellwerk3's solution comprised two key components:

1. pi_optimal Platform​

  • Implements advanced model-based reinforcement learning
  • Processes historical campaign data for pattern recognition
  • Generates hourly optimal settings (e.g. bidding, frequency-capping, whitelists, ... ) based on learned patterns

2. Adpilot Middleware​

Custom-developed solution providing:

  • Multi-platform data orchestration
  • Automated data preprocessing and refinement
  • Real-time campaign adjustment implementation
  • Integration layer between ad platforms and ML systems

Implementation Architecture​

Quantifiable Results​

Performance Metrics​

  • Cost Efficiency: 24% reduction in CPC, 31% reduction in CPM
  • Scale: Successfully automated 60+ campaigns
  • Success Rate: 91% automation success rate

Operational Benefits​

  1. Enhanced Campaign Personalization

    • Granular control over campaign settings
    • Real-time monitoring and adjustments
    • Data-driven decision making

  2. Resource Optimization

    • Improved bid strategy efficiency
    • Better budget allocation
    • Reduced manual intervention

  3. Operational Excellence

    • Streamlined workflow processes
    • Reduced human error
    • Increased campaign manager productivity

Implementation Challenges and Solutions​

  • Initial data quality issues addressed through robust preprocessing
  • Integration complexity managed through phased rollout
  • Team adaptation supported through comprehensive training

Future Outlook​

  • Planned expansion to additional advertising platforms
  • Development of advanced ML features
  • Integration of predictive analytics

Key Learnings​

  1. Start with clear success metrics
  2. Invest in data quality early
  3. Focus on team training and adaptation
  4. Implement changes gradually
  5. Monitor and adjust continuously

Conclusion​

The integration of reinforcement learning through pi_optimal and Adpilot has set new standards in digital advertising automation. This case study demonstrates that AI implementation, when properly executed, can deliver significant improvements in both campaign performance and operational efficiency.