Snap Seller Cockpit
Autonomous AI agents that call prospects and book meetings, while sellers supervise.
Snapchat sales org
The problem
Sellers drown in dashboards but have no clear next action. Account intelligence is scattered across systems, and outreach (calls and emails) is manual and slow, so the same high-value advertisers never get worked consistently. Most "AI" in sales only suggests; it does not execute.
What it's good for
- B2B sales orgs working large books of accounts
- Teams that want AI to execute outreach, not just recommend it
- Sales ops standardizing a repeatable hit-list to action motion
- Managers who need pipeline visibility and coaching signals in one place
Overview
Snap Seller Cockpit goes beyond a dashboard: it introduces autonomous AI agents that run real sales workflows while human sellers supervise and approve. An AI Voice Calling Agent (Gemini Live) makes human-like phone calls from the browser to verify accounts, follow up, and book meetings. A signal-triggered AI Email Agent watches account signals (spend drops, campaign launches, dormancy, renewals, competitive moves) and drafts personalized outreach with calendar links, gated by approval workflows. The same Hit List Engine that human reps use can be pointed at an AI agent instead, so a list becomes action the moment it is built. Roles span individual sellers, managers (insights + weekly business reviews), and admins (agent and signal configuration).
Key features
AI Voice Calling Agent
In-browser, human-like phone calls via Gemini Live, with a live transcript, briefing cards, and post-call summaries. Trained on seller call transcripts and the Snap product portfolio.
Signal-Triggered Email Agent
Monitors account signals and auto-drafts personalized emails with calendar links. Approval modes: full-auto, review-and-send, or draft-only.
Hit List Engine
Build prospect lists with smart filters (spend, industry, product adoption, contract status), preview the KPI impact, then assign to a human or an AI agent.
AI Agents Monitor
A real-time console of agents working campaigns: call outcomes, email performance, meetings booked.
Manager Insights + WBR
Pipeline health, team performance, signal trends, coaching alerts, and an AI-generated Weekly Business Review.
Admin & Agent Config
Configure agents, calibrate signal thresholds, set email approvals, and wire integrations.
How it works
- 1Admin or seller builds a Hit List with filters and previews the projected impact.
- 2The list is assigned to a human rep or to an AI agent.
- 3The Voice Agent dials prospects from the browser (Gemini Live for the conversation, Telnyx for the PSTN leg), streaming a live transcript and briefing.
- 4On a successful call it books the meeting; the Email Agent, meanwhile, watches signals and drafts outreach for approval.
- 5Managers see it all roll up into pipeline, performance, and an auto-generated WBR.
How to make it live
- Frontend: `npm run build` produces a static Vite bundle in dist/. Host it on Vercel, Netlify, or Firebase Hosting.
- Backend (optional, for live calls): containerize server/ with the included Dockerfile and deploy to GCP Cloud Run.
- Set environment: VITE_GEMINI_API_KEY, and for live calls TELNYX_API_KEY, TELNYX_PHONE_NUMBER, TELNYX_CALL_CONTROL_APP_ID, PUBLIC_WEBHOOK_BASE_URL.
- Point VITE_API_BASE_URL at the deployed Cloud Run URL, and register the Telnyx Call Control app + webhook to that host.
- No keys? The cockpit UI runs fully in demo mode without the voice backend.
cd demos/snap-seller-cockpit npm install cp .env.local.example .env.local # add your keys (optional) npm run dev # http://localhost:5173
Tech stack
Why it matters
- Models the program that reported $30M+ incremental revenue and +18% win rate
- Agents that execute (call + email), not just advise
- One hit-list motion shared by humans and AI
Snapchat is named with permission as a public partner. Prospect data shown is illustrative sample data.
Want a walkthrough of this build?
Happy to demo it live and talk through the architecture.