Responsible AI in Action: How We're Using LLMs to Protect Privacy and Understand Users

Like many organisations, we’re constantly working to better understand what our users need from us. We hold a wealth of unstructured data - from free-text notes entered by Macmillan professionals in our CRM, to insights from fundraising activities. The challenge is that this data often contains personally identifiable information (PII), making it difficult to report on or analyse safely. Yet, qualitative data like this can hold vital insights. We’ve begun exploring how AI can help us unlock this value responsibly. By using large language models (LLMs), we’re investigating how to learn from user interactions - improving how we surface relevant information, understanding sentiment, and safeguarding privacy through automated PII obfuscation. This session shares our early-stage experiments and proof-of-concepts, designed to responsibly integrate AI into our data workflows. We’ll discuss how we’re preparing proposals for governance approval, keeping humans in the loop, and learning from user feedback - all while building towards scalable, trustworthy AI-powered services that better support people living with and affected by cancer.