​
The Core Equation

Tactics = Targeting + Inventory Every tactic is a bet on a specific combination performing well. The platform continuously discovers, tests, and optimizes these bets using a multi-armed bandit approach.

​
Campaign Modes

Two fundamentally different approaches:

​
BYOT (Bring Your Own Targeting)

You define the audience, we find the inventory
  • Start with your brand story or custom signals
  • Platform discovers run-of-site inventory that reaches them
  • Optimization finds the most cost-effective paths to your audience
  • Example: “Find my enterprise IT buyers wherever they browse”

​
Collab (Collaborative)

Publishers package inventory + their data
  • Publishers combine their inventory with first-party data
  • Pre-packaged products with built-in audience insights
  • You leverage publisher expertise and relationships
  • Example: “ESPN’s fantasy football enthusiasts package”

​
The Optimization Engine

​
Multi-Armed Bandit in Action

Think of each tactic as a slot machine with unknown payouts. The platform must decide which machines to play (and how much to bet) while learning their returns. Phase 1: Discovery
Campaign starts. Platform creates 20-50 initial tactics, each getting small test budgets. Like a gambler trying every machine once. Phase 2: Exploitation vs Exploration
The eternal tension:
  • Exploit: Pour money into the tactics showing 8+ effectiveness scores
  • Explore: Keep testing new combinations that might beat current winners
The algorithm dynamically adjusts this balance based on:
  • Campaign maturity (explore more early, exploit more late)
  • Performance variance (stable = exploit, volatile = explore)
  • Market changes (new inventory appearing, seasonality)
Phase 3: Pacing Control
Even winning tactics need throttling:
  • Prevent budget exhaustion in first hours
  • Maintain presence throughout flight dates
  • Adjust for audience behavior patterns
  • React to real-time performance shifts

​
Tactic Seed Data Cooperative

When your brand agent opts into the tactic seed data cooperative (tacticSeedDataCoop: true), the platform leverages aggregated data to more effectively seed new tactics:

​
1. Historical Pricing & Delivery Data

Product ABC123 delivered 10,000 impressions/day at $1.23 CPM from 9/1/25 to 9/9/25
Product DEF456 delivered 25,000 impressions/day at $2.45 CPM from 9/1/25 to 9/9/25
This helps seed tactics with:
  • Realistic CPM starting points based on actual delivery costs
  • Expected inventory availability for initial budget allocation
  • Delivery rate estimates for new tactic pacing

​
2. Performance Quintiles by Campaign Category

Product ABC123: Top 20% of tactics for food & beverage campaigns from 9/1/25 to 9/9/25
Product DEF456: Bottom 20% of tactics for automotive campaigns from 9/1/25 to 9/9/25
This provides:
  • Category-specific quality rankings to prioritize high-performing inventory
  • Avoid poor performers by learning from aggregated campaign data
  • Better initial tactic selection based on proven category performance
How this improves new tactic seeding:
  • Skip inventory with consistently poor category performance
  • Start with higher confidence in proven inventory sources
  • Allocate initial test budgets more effectively
Without opting in, new tactics start with:
  • Publisher-provided base rates only
  • No category performance guidance
  • Equal budget allocation across all discovered inventory
Enable via brand agent settings: 
await updateBrandAgent({
  brandAgentId: "ba_123", 
  tacticSeedDataCoop: true  // Opt into tactic seeding data cooperative
});

​
How It Works

​
Inventory Sourcing

Tactics pull inventory from the Ad Context Protocol (AdCP):
  • Publishers define media products with targeting capabilities
  • Platform matches these to campaign requirements
  • Creative formats are matched to available placements
  • Quality signals ensure brand safety

​
Targeting Power-Ups

Three ways to sharpen your aim:
  1. Brand Stories: “Find people like my best customers”
  2. Custom Signals: Your CRM segments, pixel data, lookalikes
  3. Third-Party Signals: LiveRamp, Oracle, TransUnion (coming soon)
The platform tests how each targeting method performs across different inventory sources.

​
Tactic Scoring

Three-component scoring drives budget allocation: Each tactic gets a weighted score combining:
  • Quality Score (Scope3): Media quality, viewability, IVT detection
  • Outcome Score (You): Your normalized measurement data (ROAS, conversions, etc.)
  • Story Affinity (Scope3): Alignment with your selected brand stories
Budget flows to higher-scoring tactics through multi-armed bandit optimization. See Scoring Guide for detailed explanation of weights and configuration.

​
Control Modes

​
Let the Platform Drive

// BYOT example - your targeting, our inventory discovery
const campaign = await createCampaign({
  brandAgentId: "ba_123",
  name: "Q4 Prospecting",
  prompt: "Find SMB decision makers interested in cloud migration",
  
  // Configure scoring for performance focus
  scoringWeights: {
    quality: 0.3,    // 30% media quality
    outcome: 0.7,    // 70% your conversion data
    affinity: 0.0    // 0% story alignment (B2B focus)
  },
  outcomeScoreWindowDays: 14,  // B2B longer sales cycle
  
  budget: { total: 50000, currency: "USD" }
});
// Platform creates tactics, runs multi-armed bandit optimization

​
Take the Wheel

// Direct control - you know exactly what you want
const campaign = await createCampaign({
  brandAgentId: "ba_123",
  name: "Premium Sports Deal",
  manualTactics: [
    {
      inventoryId: "espn_nfl_package",
      budget: 25000,
      targeting: { segments: ["existing_customers"] }
    },
    {
      inventoryId: "foxsports_streaming", 
      budget: 15000
      // No targeting = run of network
    }
  ]
});

​
Learn More

For deeper understanding of how tactics work within campaigns:

Campaign Optimization

Performance TuningLearn how the multi-armed bandit algorithm optimizes tactic performance.

Signals Guide

Targeting DataUnderstand how signals enhance tactic targeting capabilities.

​
The Bottom Line

  • Every tactic is a bet on a targeting + inventory combination
  • BYOT vs Collab - Bring your data or use publisher packages
  • Multi-armed bandit continuously reallocates budget to winners
  • Exploration never stops - Always testing for better combinations
  • You choose control level - Full automation or manual override