When an AI agent acts on a nonprofit's behalf, it's only as trustworthy as the data underneath it. At Blackbaud, we build software for the social impact sector, and nonprofits have relied on our data for decades. But validating the data that drives insights has been slow, manual work, and agentic use cases are moving faster than people can check. Keeping up meant rethinking our data platform.

Our original strategy was simple: build it and the product teams will come. They did not come. At least not in the way we hoped. What we got instead was a lakehouse full of disconnected datasets without a deep sense of ownership, and product teams each solving the same problems in their own corners. 

This talk walks through our ongoing move from collecting datasets to curating data products: the cultural and technical challenges of getting producing teams to truly own their data, the standards and tooling that make ownership practical, and why this shift is necessary for building a trusted AI engine. You'll leave with patterns for moving an existing platform toward data products that people and AI can trust.