CampaignAI

Ad Bid Optimization: 45% CPA Reduction

The Problem

Manual bid optimization was too slow. Real-time ad auctions changed every minute, but bids were updated daily. Clients overspent by 35-45% to acquire customers. CPA was rising as Facebook/Google algorithms changed.

The Solution

Kurai deployed a real-time bid optimization system using deep reinforcement learning (DQN) and XGBoost. The system:

• Predicts conversion probability per impression
• Calculates optimal bid based on budget, CPA target, auction dynamics
• Adjusts bids every 5 minutes across 3 platforms
• Learns from auction outcomes in real-time

Model Architecture:

• Conversion model: XGBoost (94% AUC)
• Bid optimization: Deep Q-Network (DQN) RL agent
• Features: 180 features (user intent, creative, time, platform, competition)
• Training: Reinforcement learning with simulated auctions

The Results

• CPA reduction: $150 → $82.50 (-45%)
• Conversion rate: 2.8% → 3.7% (+32%)
• Bid update frequency: Daily → every 5 minutes (288x faster)
• Annual ad spend savings: $112M

Technology Stack

XGBoost, TensorFlow (Keras), Ray RLlib, Kafka (real-time streaming), PostgreSQL, Redis, Facebook/Google/TikTok Marketing APIs

Key Metrics

Metric	Before	After	Improvement
CPA	$150	$82.50	-45%
Conversion rate	2.8%	3.7%	+32%
Bid updates	Daily	Every 5 min	288x faster
ROAS	280%	410%	+46%

CampaignAI saves clients $112M annually in ad spend while increasing conversion rates by 32%.