Palo Alto Networks: Infusing AI Into Business Operations

Palo Alto Networks is a leading cybersecurity company, committed to making each day safer than the last by offering a comprehensive enterprise cybersecurity platform. This includes network security, cloud security, endpoint protection, and a set of cloud-delivered security services. 

Key AI Use Cases at Palo Alto Networks

In a 2024 Everyday AI San Francisco session, “Sticking the Landing: Making the Leap to AI and Surviving to Tell the Tale,” Lionel Some, a principal data scientist at Palo Alto Networks, highlighted several AI use cases that his team is working on:

3. Customer Retention Models: These models monitor customer health to enhance retention efforts, predicting potential churn and enabling proactive interventions.
4. Escalation Management Platform: Developed during a company hackathon, this platform predicts the likelihood of support tickets escalating, thereby helping the support team prioritize tasks and reduce resolution times, ultimately improving customer satisfaction.
5. Content Generation for Marketing: A Retrieval Augmented Generation (RAG)-based application designed for the marketing team that automates the generation of content such as blog posts, press releases, and articles, streamlining the content creation process.

Prioritizing AI Use Cases

Prioritizing AI use cases at Palo Alto Networks is a structured process that involves multiple teams. First, business leaders submit requests through a business intake form. These requests are then evaluated by an AI model team, which includes data science, information security, and legal experts. Each team assesses the feasibility, risks, and value of the project from their unique perspectives, ensuring a holistic evaluation.

Today, the Palo Alto support team uses an app built with Dataiku by internal teams, the Escalation Management Platform mentioned earlier. Born out of a hackathon hosted by the Palo Alto IT organization, this platform helps predict the likelihood that a support ticket will escalate. This initiative was fast-tracked due to its feasibility and the business impact it could deliver. This project exemplifies how Dataiku was instrumental in rapidly moving from data preparation to model building and integration, allowing the team to focus on the application’s presentation and functionality.

Ensuring Transparency and Explainability

Lionel’s team, composed of nine data scientists, is focused on assisting customer support, sales, and marketing teams through various AI-driven projects. In all of these projects, there is a significant focus on ensuring transparency and explainability of AI models. Making stakeholders understand how AI models work, what data they use, and the constraints they operate under is necessary to bring projects to production and ensure everyone is on the same page. 

For example, when presenting their RAG-based content generation tool to the architecture team, Lionel’s team received a request to increase visibility in what was happening behind the scenes. The team thus added features to the UI that allow users to see the source chunks retrieved from a vector database, fostering trust and enabling validation of the tool’s functionality as well as rapid feedback. 

For me, transparency is about explaining what goes into the model, how the model works, and also what constraints we have to work with. Explainability, on the other hand, is about providing a description of what the model outputs or how the model output came to be. Lionel Some Principal Data Scientist, Palo Alto Networks

Explainability was all the more important for Palo Alto as they were replacing existing rules-based systems, which are usually easily interpretable, by more complex machine learning models that are not as interpretable. For these new models, Lionel’s team provided explanations using models like SHAP to clarify how the model outputs were derived. 

This transparency and explainability are critical in maintaining stakeholder trust and ensuring the successful adoption of AI tools through feedback loops and continuous improvements.