Audience Signals

​

Overview

Audience Signals are predictive user segments built from treatment effect scores. They identify users by predicted incremental response—who will convert because of your ads, not just who looks similar to past converters. Audience Signals sync directly to advertising platforms through native APIs, enabling lookalike expansion, direct targeting, and spend suppression without manual exports.

​

Prerequisites

Data Requirements Audience Signals require three data inputs for treatment effect modeling: Event data – Behavioral interactions (page views, product views, add-to-cart, purchases, signups) from:

• Innkeepr.js (client-side tracking)
• Server-side APIs (backend events, transaction data)
• CDP integrations (Segment, RudderStack, mParticle, Braze)
• Data warehouses (BigQuery, Snowflake, Redshift)

Treatment exposures – Marketing touchpoint data showing which users were exposed to which treatments, captured through:

• UTM parameters (automatically parsed)
• Platform click IDs (gclid, fbclid, ttclid)
• Referrer data and campaign identifiers

User identifiers – Matching keys for signal activation:

• First-party cookies (anonymous tracking)
• User IDs (logged-in users)
• Email addresses (hashed for platform matching)

Timeline

​

Use Cases

Innkeepr supports three primary use cases for Audience Signals: Seed Audiences – Acquisition/Lookalikes
High-incrementality user lists used to seed lookalike audiences. Platforms use these seeds to find similar prospects, improving acquisition efficiency. Retargeting Audiences – Re-engagement
Users with high predicted recovery or repeat purchase probability. These segments identify users who interacted with your brand but haven’t converted, and who are predicted to respond incrementally to retargeting. Exclusion Audiences – Spend Protection
Low or negative-incrementality users (bottom 20-40% by predicted lift). These audiences should be excluded from campaigns to suppress wasted spend on users unlikely to convert incrementally.

​

Setup Instructions

Step 1: Define Your Objective
The objective function defines what business outcome your signals optimize for. All treatment effect predictions - and therefore all audience segments - maximize incremental impact on this objective. Examples:

• Purchase probability within 7 days of ad exposure
• 90-day CLV for new customers
• Signup likelihood for free trial visitors
• Repeat purchase probability for existing customers

Multiple objectives can run in parallel for different signal strategies. Step 2: Configure Audience Segments
Configure segments based on treatment effect score distributions:

1. Select audience type – Choose your signal strategy:
   • Acquisition seeds (top 10-20% by predicted lift)
   • Exclusion audiences (bottom 20-40% by lift or negative effects)
   • Retargeting segments (high recovery probability)
2. Link objective – Select the objective this audience optimizes for
3. Set treatment context (optional) – Specify which historical campaigns this audience learns from. Linking past campaigns enables faster pattern recognition.

Step 3: Activate Across Destinations
Audience Signals sync directly to connected platforms through native APIs. Supported destinations:

​

Best Practices & Optimization

Segment Configuration

• Start with broader segments (top 20-30%) and narrow based on performance data
• Use separate audiences per objective - don’t mix CLV and conversion likelihood segments
• Refresh exclusion audiences more frequently (daily) than acquisition seeds (every 2-3 days)

Testing Recommendations

• Test audience signals incrementally against existing targeting using control/test campaign structures
• Allow 2-3 model training cycles (4-6 weeks) before evaluating signal quality
• Compare performance against traditional broad, behavioral or demographic audiences

Common Pitfalls

• Using exclusion audiences for targeting - these are optimized to identify non-incremental users
• Evaluating too early - signal accuracy improves after initial training cycles
• Overlapping audiences - ensure seed and exclusion segments don’t compete in the same campaigns

​

Troubleshooting

Audience Not Populating
Issue: No users in audience after 48 hours Solutions:

• Verify sufficient event data is flowing (minimum 5,000 monthly sessions)
• Check that user identifiers are being captured correctly
• Confirm platform connection is active and permissions are granted

Low Match Rates
Issue: Platform audience size is significantly smaller than Innkeepr audience Solutions:

• For email-matched audiences: Verify email formatting and hashing
• For pixel-based audiences: Ensure users have visited your website
• Allow 24-48 hours for full platform matching to complete

Signal Quality Issues
Issue: Audience performance doesn’t improve over baseline Solutions:

• Verify objective function aligns with actual business goals
• Check that sufficient conversion volume exists (300+ monthly conversions recommended)
• Allow more training cycles - early models have wider confidence intervals