ActionFlow: Tableau-to-AI Bridge

Turn passive dashboards into active solutions. Detect revenue risks in Tableau and trigger an autonomous AI agent that executes remediation plans and sends instant Telegram alerts with one click.

Comment

💡 Inspiration

We live in an era of "Data Overload" but "Action Paralysis." While working with business dashboards, we noticed a critical gap: executives see a problem (e.g., sales dropping, churn risk increasing), but fixing it requires switching context—opening CRMs, sending emails, or calling managers. This "Insight-to-Action" latency costs companies millions.

We asked ourselves: What if the dashboard wasn't just a mirror, but a steering wheel? What if you could fix a business problem instantly, right from the chart?

This inspired Revenue Guardian AI—a bridge between visual insights and autonomous resolution.

🤖 What it does

Revenue Guardian AI turns passive Tableau dashboards into active command centers.

  1. Detection: An executive spots a risk (e.g., low Win Rate or a specific Opportunity at risk) on the Tableau Dashboard inside Salesforce.
  2. Trigger: They click a single button: "Analyze & Remediate".
  3. Orchestration: Salesforce Flow captures the context (Opportunity ID, metrics) and securely tunnels it to our external AI Agent.
  4. AI Resolution: Our Java Spring Boot backend analyzes the data, determines the best course of action (e.g., "Apply 20% Retention Discount"), and executes the logic.
  5. Instant Notification: The system instantly triggers a Telegram Bot alert to the regional manager's phone with the analysis and the action taken.

⚙️ How we built it

We architected a multi-layer integration connecting three distinct ecosystems:

  • The Frontend: We used Tableau embedded in Salesforce to visualize the data and Salesforce Flows to handle the user interaction.
  • The Bridge: We utilized Salesforce External Services and Named Credentials to create a secure, authenticated pipe between the CRM and our custom backend.
  • The Brain (Backend): We built a robust Java Spring Boot application. It serves as the "Agent" that receives the webhook, processes the business logic, and creates the remediation plan.
  • The Notification Layer: We integrated the Telegram Bot API directly into our Java service to deliver real-time, mobile-first alerts to stakeholders.

🚧 Challenges we ran into

The biggest challenge was the security handshake between Salesforce and the external world.

  • Authentication Nightmares: We faced multiple HTTP 403 Forbidden errors. Configuring the correct Principal Access in Salesforce Permission Sets and fine-tuning the Named Credential headers took significant debugging.
  • Data Transport: ensuring the JSON payload correctly passed dynamic variables (like OpportunityID) from the dashboard visual through the Flow and into the Java controller required precise mapping.

🏅 Accomplishments that we're proud of

  • End-to-End Latency: We achieved a near-instantaneous response. From the moment the button is clicked to the phone vibrating with a Telegram notification, it takes less than 1 second.
  • Clean Architecture: We implemented a clean Controller -> Service -> DTO pattern in Java, making the agent scalable for future AI model integration.
  • User Experience: We successfully hid all the complex complexity behind a single, simple button.

🧠 What we learned

  • We mastered Salesforce Flow Builder and its ability to act as an API gateway.
  • We deepened our understanding of Spring Boot REST controllers and handling cross-origin (CORS) requests from enterprise platforms.
  • We learned how to integrate ChatOps (Telegram) into business workflows effectively.

🚀 What's next for Revenue Guardian AI

  • LLM Integration: Connecting the Java backend to Gemini Pro to generate personalized emails for clients based on the specific sales data.
  • Multi-Channel Support: Adding Slack and Microsoft Teams as notification targets.
  • Auto-Pilot Mode: Allowing the agent to monitor the dashboard 24/7 and trigger fixes without human intervention when metrics hit critical thresholds.
Share this project:

Updates