Skip to content

Latest commit

 

History

History
450 lines (299 loc) · 10.3 KB

DEMO_GUIDE.md

File metadata and controls

450 lines (299 loc) · 10.3 KB

Ad Campaign Agent - Demo Guide

A complete walkthrough for demonstrating the Ad Campaign Agent platform.

Executive Journey

Executive Journey

Overview

This demo showcases an AI-powered in-store retail media platform built with Google's Agent Development Kit (ADK). The system demonstrates how multi-agent AI can transform retail advertising from creative ideation to performance optimization.

Key Value Propositions:

  • 10x faster creative production - Generate video variations in minutes, not weeks
  • Human-in-the-Loop (HITL) control - AI assists, humans approve
  • Data-driven optimization - Apply winning formulas across campaigns
  • Geographic intelligence - Store-level performance with Google Maps integration

Demo Duration

Version Time Notes
Full Demo 20-25 min All 6 acts
Core Demo 15 min Skip optional scenes
Quick Overview 5 min Acts 1 + 4 only

Audience: Retail Media Executives, Marketing Leaders, Technical Decision Makers


The Story

"You're the Director of Retail Media for a fashion retailer with stores across the US. Holiday season is approaching, and you need to launch personalized video campaigns for each store location - but your creative team is overwhelmed. Let's see how AI agents can help..."


Pre-Demo Checklist

  • ADK Web UI running (adk web app or Cloud Run URL)
  • Product images available in GCS at product-images/
  • GOOGLE_MAPS_API_KEY set (for static maps)
  • Test one video generation to warm up models
  • Clear browser cache for artifact viewing

Act 1: Understanding the Platform (2 min)

Scene 1.1: Meet the Multi-Agent System

Query:

What agents are available and what are our current campaigns?

What to Highlight:

  • Coordinator routes to 4 specialized agents
  • Product-centric model: Each campaign = 1 product + 1 store
  • Pre-loaded with 4 demo campaigns

Scene 1.2: Browse the Product Catalog

Query:

Show me all available products with their image links

What to Highlight:

  • 22 products across 5 categories
  • Clickable image_url links to view products
  • Each product ready for video generation

Act 2: Creative Generation with AI (5-6 min)

Scene 2.1: Explore Variation Options

Query:

What variation presets do I have for video generation?

What to Highlight:

  • Diversity presets (model ethnicities)
  • Setting presets (studio, beach, urban, cafe, etc.)
  • Mood presets (elegant, romantic, bold, playful)
  • These enable A/B testing

Scene 2.2: Generate a Video

Query (Use existing campaign):

Generate a video for campaign 2 with a European model on a rooftop at sunset, with a sophisticated elegant mood

Query (Create new campaign):

Create a campaign for the sage-satin-camisole at our Miami Beach store, then generate a video with a Latina model on a beach at golden-hour, romantic serene mood

What to Highlight:

  • Two-stage pipeline:
    1. Stage 1: Gemini generates scene image
    2. Stage 2: Veo 3.1 animates into 8-second video
  • Video starts in "generated" status (pending review)

Demo Tip: While generating (~2-3 min), explain:

"The AI first creates a scene image showing a model wearing our product, then animates it into a cinematic video."


Scene 2.3: Batch Generate Variations

Query:

Generate three video variations for campaign 2:
1. African model in studio with dramatic lighting, bold energy
2. Asian model in cafe setting, warm sophisticated mood
3. European model walking in urban street at day, dynamic energy

What to Highlight:

  • Batch generation for efficiency
  • Each variation named descriptively
  • All pending review before going live

Variation Parameters Reference

Category Options
Model asian, european, african, latina, south-asian, middle-eastern, diverse
Setting studio, beach, urban, cafe, rooftop, garden, nature, office, street
Mood elegant, romantic, bold, playful, sophisticated, mysterious, serene
Lighting natural, studio, dramatic, soft, golden, neon, moody
Time golden-hour, sunrise, day, sunset, dusk, night
Activity walking, standing, sitting, dancing, spinning, posing, running
Camera orbit, pan, dolly, static, tracking, crane, handheld

Act 3: Human-in-the-Loop Review (4-5 min)

Scene 3.1: The Review Table

Query:

Show me the video review table with public links

What to Highlight:

  • Card-based format with full details
  • Clickable WATCH VIDEO links for preview
  • Emphasize: "Nothing goes live without human approval"

Scene 3.2: Batch Activation

Query:

Activate videos 5 and 7

What to Highlight:

  • Batch activation with activate_batch([5, 7])
  • Status changes to "activated"
  • 30 days of metrics generated on activation
  • Metrics only start after human approval

Scene 3.3: Pause and Archive (Optional)

Query:

Pause video 6 temporarily
Archive video 8

What to Highlight:

  • Full lifecycle control: generated -> activated -> paused -> archived
  • Paused videos preserve metrics history

Act 4: Analytics & Optimization (4-5 min)

Scene 4.1: Campaign Metrics

Query:

Get metrics for campaign 2 over the last 30 days

What to Highlight:

  • In-store retail media metrics (not digital):
    • Impressions: Ad displays on in-store screens
    • Dwell Time: Seconds shoppers viewed the ad
    • Circulation: Foot traffic past display
    • RPI (Revenue Per Impression): Primary KPI
  • Weekend patterns visible (40% higher)

Scene 4.2: Generate Charts

Query:

Generate a trendline chart showing revenue per impression for campaign 2 over 30 days

Alternative Queries:

Create a bar chart of weekly impressions for campaign 3
Generate a comparison KPI card for campaign 1
Create an infographic visualization of campaign 2 performance

What to Highlight:

  • AI-generated charts using Gemini
  • Anti-hallucination: Uses ONLY real data
  • Saved as artifact for download

Scene 4.3: Compare Campaigns

Query:

Compare all four campaigns side by side

What to Highlight:

  • Side-by-side metrics comparison
  • Rankings by RPI and total revenue
  • Identifies winner with explanation

Act 5: Geographic Intelligence (3-4 min)

Scene 5.1: Campaign Map

Query:

Show me all campaign locations with Google Maps links

What to Highlight:

  • Clickable Google Maps URLs for each store
  • Performance metrics per location

Scene 5.2: Generate Map Visualization

Query:

Generate a performance map showing all campaigns on a US map in infographic style

What to Highlight:

  • AI-generated infographic using Gemini
  • Revenue bubbles sized by performance
  • Regional summary panel

Scene 5.3: Regional Comparison

Query:

Generate a regional comparison map showing West Coast vs East Coast RPI

Act 6: Apply Winning Formula (2 min)

Scene 6.1: Scale Success

Query:

Apply the winning formula from video 5 to campaign 3

What to Highlight:

  • Extracts winning characteristics: mood, setting, lighting, camera
  • Applies to different product at different location
  • New video generated with proven approach

Closing: Value Summary

Key Points:

  1. Multi-Agent Collaboration - 4 specialized agents working together
  2. Product-Centric Campaigns - Clear attribution per product per store
  3. Creative at Scale - Multiple video variations with AI
  4. Human Control - HITL review and approval
  5. In-Store Analytics - Retail-appropriate metrics
  6. Geographic Intelligence - Google Maps integration
  7. Data-Driven Optimization - Apply winning formulas

Quick Reference: Demo Queries

Campaign Management

List all campaigns
Show me campaign 2 details
Create a campaign for the sage-satin-camisole at our Miami Beach store
What campaigns are in draft status?

Product Browsing

Show me all available products with their image links
List products in the dress category
Show me outerwear products with URLs

Video Generation

What variation presets are available?
Generate a video for campaign 2 with a European model on a rooftop at sunset
Generate 3 variations for campaign 3 with different settings

HITL Review

Show me the video review table with public links
Activate videos 5 and 7
Pause video 6
What's the status of video 5?

Analytics

Get metrics for campaign 2 over the last 30 days
Top 5 videos by RPI
Compare campaigns 1, 2, 3, 4
Generate a trendline chart for campaign 1

Maps

Show me all campaign locations with Google Maps links
Generate a performance map showing all campaigns
Generate a regional comparison map

Optimization

Apply winning formula from video 5 to campaign 3
Get video properties for video 5

Campaign Reference (Pre-loaded)

ID Product Store City
1 Blue Floral Maxi Dress Westfield Century City Los Angeles, CA
2 Elegant Black Cocktail Dress Bloomingdale's 59th Street New York, NY
3 Black High Waist Trousers Water Tower Place Chicago, IL
4 Emerald Satin Slip Dress The Grove Los Angeles, CA

Troubleshooting

Issue Solution
"No metrics" for campaign Videos must be activated first
Video generation slow Normal - Veo 3.1 takes 2-3 minutes
Chart not showing data Check that campaign has activated videos
Static map API error Set GOOGLE_MAPS_API_KEY or use AI-generated map
Video links show gs:// URLs Add "with public links" to query

Models Used

Purpose Model
All Agents gemini-2.5-pro
Scene Images gemini-2.5-flash-image
Video Animation veo-3.1-generate-preview
Charts/Maps gemini-2.5-flash-image

Environment Variables

Variable Purpose Required
GOOGLE_CLOUD_PROJECT GCP project Yes
GCS_BUCKET Cloud Storage bucket Yes
GOOGLE_MAPS_API_KEY Static Maps API Optional