Ideathon · Fall 2026 · 2nd place

Salesforce Ideathon

Ethan Skinner and I took 2nd place, and $150 each, at the University of Utah’s Salesforce Ideathon with a support system for Pronto, a made-up food-delivery company, built overnight in Service Cloud and Agentforce.

The idea

Most teams built a better customer chatbot. We treated the restaurant as a customer too. Merchants get their own support site where a store manager can report a problem (“out of fries tonight”) in a few seconds. When customers start complaining about missing fries, the AI agent already knows: it checks for open reports at that storefront, tells the customer the restaurant flagged it, and offers a refund without asking for proof.

When the agent can’t help, it doesn’t loop. It logs a case summarizing the conversation, gives the customer the case number and offers a human, and the support lead gets an alert. A dashboard then shows which restaurants customers complain about but who rarely report problems themselves, which is who Merchant Ops should call first.

I built it with Ethan Skinner; I was the support lead and org admin.

The merchant reports a problem

Pronto merchant support site home page with a Report a problem tonight form
A support site for restaurants, separate from the customer one
The report form filled in: storefront Urban Table Downtown, message Out of fries tonight
No login or app: an email, a storefront and a message
Salesforce case Out of fries tonight with type, reason Out of Stock, storefront, contact and Merchant Ops Queue filled in
It lands as a case, auto-classified as Out of Stock and routed to Merchant Ops

The customer complains

Chat: the customer says their fries were missing and the agent asks for email and last name
Verify identity
Chat: the agent says Urban Table Downtown already reported being out of fries and offers a $14.20 refund
The agent already knows
Chat: the customer accepts and the agent issues refund REF-0007
Refund, once confirmed
Chat: the customer replies vaguely and the agent asks one clarifying question
One clarifying question
Chat: after a second vague reply the agent logs case 00001131 and offers a live agent
Then a case, not a loop

Behind the scenes

Agent failure case with a notification to the support lead: Agent failure, Clarification Loop (Medium)
The agent writes the failure case and the support lead is alerted
Pronto Service Health dashboard: issues by storefront and source, issue mix, agent failures by type, open follow-up work
Service health: customer vs. merchant reports per storefront (sample data)
Salesforce Flow that matches the contact, finds the storefront, classifies the report and routes it
The flow that fills in merchant reports
Agentforce Builder showing a router and five subagents
The agent: a router and five subagents

What’s next

The judges’ note was that a merchant report should change what customers can order: if the restaurant is out of fries, the app should stop selling fries rather than refund them afterward. The menu data was already in the org, so it’s one more step in the flow.

Service Cloud Agentforce Flow Apex LWC Experience Cloud With Ethan Skinner