Pushpay shares how it evaluates an agentic AI search built on Amazon Bedrock

AWS published a post today detailing how Pushpay built and continuously evaluates an agentic AI search experience on Amazon Bedrock.

• Pushpay serves churches and faith-based organizations with a digital giving and engagement platform, and added a natural-language AI search flow inside its existing app.
• The feature targets ministry staff who need quick answers from community data (for example, finding group members who haven’t donated this year or spotting people with low engagement).
• Early adopters reported cutting time-to-insight from minutes down to seconds, according to Pushpay.
• The agent design includes a fixed system prompt plus a dynamic prompt constructor that injects user- and church-specific context, sample queries, and application filters.
• The team used Amazon Bedrock to build a custom GenAI evaluation framework aimed at ongoing quality assurance and faster iteration via feedback loops on AWS.