Gusto aimed to enhance customer support efficiency by integrating an LLM-powered API into their Salesforce system. However, this integration posed a significant security risk, as customer support tickets often contained sensitive personally.
Challenge
Gusto aimed to enhance customer support efficiency by integrating an LLM-powered API into their Salesforce system. However, this integration posed a significant security risk, as customer support tickets often contained sensitive personally identifiable information (PII) and protected health information (PHI).
Solution
Gusto, prioritizing data security in its customer support operations, adopted the Formal HTTP Sidecar as an interface between their Salesforce instance and the LLM provider s API. Formal Sidecar plays a key role in Gusto s data protection strategy. The Formal Sidecar uses ML to identify in real-time and dynamically mask sensitive data within customer support tickets thanks to it s Open-Policy-Agent-powered engine, thereby avoiding sensitive data leakage. The ML model is continuously learning from hundred of thousand of requests processed.
“Formal Satellites simplified the process to ensure any PII or PHI entities transiting between Salesforce and our LLM provider would be automatically detected and masked, without needing detailed knowledge of the data s structure or content. wrote Ian Wardell, Data Privacy Lead at Gusto. Formal not only fortified our data privacy but also streamlined our compliance process, upholding the integrity and trust in our customer interactions”.
Gusto was able to implement the Formal platform, including proxy setup, configuration, policy creation, and instant masking within 5 days.
Results
Gusto’s implementation of Formal HTTP proxy successfully processed hundreds of thousands of requests, securely masking sensitive data in real-time while maintaining detailed audit logs. This pivotal enhancement enabled Gusto to confidently adopt advanced AI capabilities for customer support, ensuring top-tier data security without sacrificing efficiency.