Forge Confluence Macro
Forge Analysis Output
Sample Website to test Statsig Experiments
ISO Workflow Chart

ISO — Smarter Product Decisions Inside Atlassian

Inspiration

Every Product Manager faces the same impossible question: "Which feature should we build next, and will it actually make an impact?"

The truth is, over 65% of product teams struggle because critical data is fragmented / scattered across Confluence docs, Jira tickets, and external experiment tools like Statsig. This forces PMs to make decisions based on intuition instead of evidence.

ISO (Intelligent Smarter Optimization) was built to solve this. Our vision was to prove that a PM could go from idea → prediction → experiment → validation in minutes, all within their existing Atlassian workflow. We built a full-stack AI agent on Atlassian Forge that connects Jira, Confluence, and Statsig to make every decision data-driven and validated.

What it does

ISO is an intelligent agent that serves as the unified decision layer for product managers. It combines AI reasoning, workflow automation, and live experimentation into a seamless, three-click process inside Jira and Confluence:

AI Prediction & RICE Scoring: The agent analyzes your Confluence PRD content and historical data from Jira (velocity, past feature success). It then uses an LLM to generate an immediate, explainable RICE (Reach, Impact, Confidence, Effort) prediction, complete with rationales, right on the page.
Automated Experiment Orchestration: Using the Atlassian Rovo Agent framework, ISO automatically triggers the creation and deployment of a live feature experiment in Statsig, eliminating manual setup.
Closed-Loop Validation: The Forge agent fetches the final results from Statsig. It visualizes the Predicted Lift vs. Actual Lift directly on a Confluence dashboard and automatically creates a final Jira ticket summarizing the validated implementation.

How we built it

We built ISO as a coherent, four-layer system to operate natively within the Atlassian ecosystem.We built the Atlassian Forge App (React/TypeScript) to read content from both Jira and Confluence, forming the core user interface. We utilized the Atlassian Rovo Agent APIs to handle the deep reasoning and autonomously invoke the experimentation layer. Crucially, we built a secure Serverless Helper on Vercel to securely hold Statsig server keys, handling the experiment creation and streaming the results back to Forge. This allowed us to orchestrate end-to-end experiments using the Statsig API and a simulated demo site for user traffic generation.

ISO Architecture Diagram showing Forge, Rovo, Statsig, and Vercel connections

The Hardest Challenges We Faced

Building ISO required blending an AI reasoning layer with enterprise-grade toolchains, presenting several key technical hurdles:

Forge Deployment & Cross-App Synchronization: We had to master the steep learning curve of scaffolding and debugging a single Forge app that runs seamlessly and securely across both Confluence and Jira environments while ensuring cross-app synchronization throughout the experiment lifecycle.
LLM Consistency & Schema Enforcement: Prompt engineering the Rovo Agent was difficult. It took many iterations to consistently generate valid, context-aware RICE predictions in a strict JSON format across highly variable inputs from different documents.
Secure API Integration: We faced significant challenges debugging the authentication and precise data flow handoff between the Atlassian Forge environment and our secure Vercel serverless worker to successfully orchestrate experiments via the Statsig MCP (Model Control Plane).

Accomplishments that we're proud of

We built a fully functional intelligent agent that solves the product manager's core problem: moving from fragmented data to validated decisions in minutes. We proved that Atlassian Forge can host complex, multi-tool agents powered by emerging technologies like the Rovo Agent. We successfully integrated three distinct platforms, Jira, Confluence, and Statsig, into a single, cohesive workflow, eliminating context switching and providing PMs with immediate, data-backed confidence in their prioritization.

What we learned

Building on Forge taught us that delivering a native, secure experience within the Atlassian ecosystem has a steep but rewarding learning curve. From the AI perspective, we learned that the power of an LLM comes not just from its reasoning, but from its ability to orchestrate external tools (Rovo Agent) and handle schema enforcement for enterprise-ready, structured data. Most importantly, we learned that blending AI reasoning with experimentation is the key to automating decision-making at scale.