AI That Turns More Donors into Donations

A fundraising platform serving nonprofits of all sizes identified a persistent market gap: while enterprise organizations routinely optimize donation experiences through rigorous testing, smaller nonprofits rarely have the expertise or bandwidth to do the same. The opportunity was clear — bring AI-driven form optimization to every customer on the platform and increase donor conversion rates across the entire portfolio. With major industry conferences and a critical giving season approaching, the organization partnered with Protagona to turn that vision into a demonstrable proof of concept.

The goal was to transform an existing A/B testing capability from an underused feature into an intelligent, AI-powered tool that automatically generates donation form variants grounded in nonprofit best practices and organization-specific context. The immediate objective was a working, demonstrable POC ready for the AWS Imagine Conference and the NEO Summit, with a clear path to production deployment before the November giving season — all within a fixed five-week engagement window.

The core challenge was not simply building an AI feature — it was making AI recommendations trustworthy and actionable within a highly constrained operating environment. Nonprofit customers had access to A/B testing tools but lacked the knowledge to use them effectively, creating an adoption gap that no amount of UI improvement alone could close. The solution needed to encode nonprofit fundraising expertise directly into the generation process, applying guardrails derived from internal best practices and third-party consulting research so that every generated variant was both technically valid and strategically sound.

‍

On the technical side, the team needed to integrate a large language model into existing infrastructure without requiring an overhaul of backend systems. That meant working within established data storage patterns, version control tooling, and database schemas rather than imposing a greenfield architecture. Layered on top was a compressed five-week timeline with fixed, non-negotiable demo deadlines tied to two major industry conferences — leaving no room for extended discovery or mid-project architectural pivots. Every decision had to balance innovation with immediate operability.

Protagona designed a serverless, event-driven pipeline that accepts donation form configurations as structured JSON and uses Amazon Bedrock to generate optimized B-variant forms. The system applies a multi-layered validation framework at every stage — encoding fundraising best practices, enforcing schema compliance, and tailoring recommendations to each nonprofit's audience, industry, and brand guidelines. Every inference is logged with full prompt, parameter, and latency data, providing complete visibility into how variants are produced.

‍

A key architectural decision was fitting the solution into existing infrastructure rather than introducing new dependencies. Form templates and validation rules live in PostgreSQL, prompt configurations are managed through Bitbucket, and inference logs are captured in DynamoDB. This reduced integration friction and allowed the internal team to take ownership immediately after handoff.

‍

The solution is exposed through a REST API via Amazon API Gateway, allowing the existing frontend to consume AI-generated variants without backend changes. All infrastructure is defined in Terraform for full reproducibility, and comprehensive documentation was delivered alongside the working POC.