AI-Powered Email Threat Detection: 20% Higher Malicious Detection & 20% Spam Reduction for a Global Email Security Firm
Industry: Cybersecurity, AI
Client
A global cybersecurity firm focused on email security
Goal
To transition from rules-based detection to advanced AI-driven systems that improve email threat detection, reduce false positives, and strengthen protection against phishing and impersonation attacks.
Challenges
- Inability of the rules-based system to scale with growing email volume and complexity.
- Difficulty detecting sophisticated phishing, impersonation, and evolving threat patterns.
- High false positives leading to poor user experience and inefficiency.
- Heavy reliance on manual intervention to maintain and update detection rules.
Solution
Transitioned Tessian’s detection system from rules-based to machine learning for improved scalability.
Implemented CatBoost models with Spark preprocessing to boost malicious email detection by 20%.
Fine-tuned Large Language Models (LLMs) for email topic classification, reducing spam by 20%.
Improved sales team efficiency by filtering irrelevant emails through advanced classification.
Built a Transformer-based model to detect senior executive impersonation attempts.
Deployed models on AWS SageMaker, scaling to handle 50,000 requests per minute.
Impact:
Achieved a 20% improvement in malicious email detection accuracy.
Reduced spam by 20%, enhancing overall email security.
Scaled the detection system to manage high request volumes efficiently.
Improved system robustness and response times for real-time threat handling.
Strengthened business case for LLM research and secured executive investment.
Enabled more advanced email body text analysis, enhancing future threat detection capabilities.
*Case studies reflect work undertaken by our Heads of AI either during their tenure with Head of AI or in prior roles before they were part of the Head of AI network; they are provided for illustrative purposes only and are based on conversations with our Heads of AI.