Scaling Behavioral Science Research to 100,000+ Students
3 Years, 4 Iterations, 30+ Universities
Custom Applications
Applied AI
Education
Challenge
Researchers at the Behavior Change for Good (BCFG), an initiative within the School of Arts and Sciences and The Wharton School at the University of Pennsylvania, wanted to answer a critical question: Which behavioral interventions most effectively improve student outcomes for first-year college students?
BCFG needed a technical solution that was scalable, reliable, and extensible; that provided an engaging user experience for the target audience; and that could be solidly built, thoroughly tested, and smoothly rolled out on an ambitious timeline.
All of this needed to meet the standards for peer-reviewed behavioral science research: complete audit trails, precise delivery timing, and rigorous data integrity.
BCFG needed a partner who could build a platform capable of delivering interventions at scale while maintaining research integrity.
Partnership
Pod worked with BCFG over a 3-year-period—each year bringing new requirements, more interventions, more universities and participants, new integrations, and evolved research designs.
This partnership included 3 separate studies involving:
- 10+ interventions
- 100,000+ participants across 30+ institutions
- Program messaging managed at scale with message delivery on specific days and times for each student, accounting for different interventions, academic calendars, and time zones across every institution
How It Works
We built a multi-tenant research delivery platform that enables BCFG to run large-scale behavioral studies across university partners nationwide.
MULTI-TENANT ARCHITECTURE
Within the platform, each participating university operates as an independent “school” with its own branding, academic calendar, authentication via existing SSO, and reporting. BCFG manages studies across all institutions from a central dashboard.
PLUGGABLE INTERVENTION SYSTEM
Critical to the system design was allowing researchers to deploy new intervention types without engineering bottlenecks. We built a modular system where interventions are configured, not coded.
ROBUST DELIVERY INFRASTRUCTURE
The system handles scheduled delivery, two-way conversations with automated responses, group messaging, MMS support for rich media, and carrier compliance requirements. A sophisticated state machine tracks each participant’s progress through multi-step interactions such as conditional follow-ups, input validation, automated incentive distribution. Among the multiple interventions, schools, and time zones, there were hundreds of possible message schedules to track.
THE “TIMING PROBLEM”
Here’s the challenge that mattered most: when you’re running a research study, timing isn’t flexible. Messages scheduled for “11am EST on Tuesday” need to arrive at the specified time and date for all participants. With tens of thousands of participants in different time zones, that’s a distributed systems problem.
We implemented a scaling system to spin up resources when they were needed and intelligently manage bottlenecks in the distribution process. Messages arrive at the precise scheduled time, even when the system is processing thousands of deliveries simultaneously. Complete audit trails enable post-hoc analysis—every message, every delivery event, every response, logged and timestamped.
Technologies Used
Django with PostgreSQL, Celery and Redis for distributed task scheduling, AWS ECS infrastructure, integrations with Twilio, SendGrid, OpenAI, Box, and university SSO systems. Three-tier deployment with automated CI/CD. The kind of foundation that scales without manual intervention.
The Results
We built a robust platform that handles the entire research lifecycle—participant enrollment, intervention delivery, response tracking, data export for analysis.
More importantly, since behavioral science research requires precision and auditability, this research wouldn’t have been practical at this scale without purpose-built infrastructure. The platform delivers both while remaining flexible enough to evolve with changing study designs.
