Set Up Seed Audiences on Meta

​

Overview

Innkeepr Seed Audiences are custom audience lists uploaded to Meta Ads Manager. They contain visitors ranked by predicted incremental lift based on your selected objective. Meta uses these lists as a basis for Lookalike Audiences (LALs) in ad sets. This guide covers the setup of Innkeepr Seed Audiences on Meta, including audience structure, testing, and scaling.

​

Prerequisites

Technical Requirements:

• Active Meta Ads Manager account
• Innkeepr integration connected to your Meta account
• Sufficient conversion data for signal generation (minimum 30 days recommended)

Timeline at a Glance:

​

Audience Types

Innkeepr provides two seed audience types for Meta: Custom Audiences User lists uploaded directly to Meta. These lists contain visitors ranked by predicted incremental lift based on your selected objective. The objective determines what the ranking optimizes for — for example conversion probability, customer lifetime value, or average order value. They are used as source audiences for Lookalike Audiences. Value-Based Audiences Lists of users identified by email address, where each entry has an assigned value based on predicted incremental impact. Unlike custom audiences, value-based audiences require an email address for each user — visitor-level data without email is not sufficient. Meta uses these values to build Lookalike Audiences that prioritize finding users similar to the highest-value entries, not just the full list. This gives Meta an additional signal for LAL expansion. Email handoff to Innkeepr works via shop system integration (e.g. Shopify) or Innkeepr Identify. Learn more:

• Shop System Integrations
• Innkeepr Identify

​

Setup Instructions

Step 1: Create Audience Buckets There are different ways to use seed audiences on Meta, and not every setup works equally well across all accounts. The optimal configuration depends on factors like audience size, conversion volume, and product category. As of March 2026, we recommend the following setup as a starting point: Recommended bucket structure: The logic: more selective source audiences (smaller percentile) are paired with smaller LALs to maintain signal quality. Broader source audiences can support larger LALs because the expansion is less sensitive to individual user variation. Step 2: Upload Audiences to Meta

1. Create audiences in Innkeepr for each percentile bucket
2. Select “Meta” as your destination
3. Upload — Innkeepr sends innkeepr_targeting events to your connected Meta pixel
4. Verify events in Meta Events Manager — after the first upload, the innkeepr_targeting event appears in your Events Manager. You need to approve the event before Meta processes the audience data. Without this approval, audiences will not populate.
5. Create Lookalike Audiences in Meta Ads Manager based on each uploaded bucket, using the LAL sizes from the table above

Step 3: Set Up the Campaign Structure Test Innkeepr audiences against broad targeting within a single campaign using separate ad sets. Campaign structure: Important: Keep all other variables identical across ad sets:

• Same creatives (images, videos, copy) in every ad set
• Same optimization event (e.g. Purchase, Add to Cart)
• Same bid strategy
• Same placements (or all set to Advantage+)

The only difference between ad sets should be the audience targeting. If creatives or other settings differ, performance differences cannot be attributed to the audience signal. Step 4: Set Budget Allocation Start all ad sets with the same daily budget using ad set budget optimization (ABO). Do not use campaign budget optimization (CBO / Advantage Campaign Budget) during the test phase — CBO shifts spend unpredictably between ad sets, making it difficult to compare performance. After 1–2 weeks, you can begin scaling individual ad sets based on observed performance (see Monitor & Optimize).

​

Monitor & Optimize

Weeks 1–2: Observation Phase

• Do not adjust budgets, bids, or creatives during the initial learning phase
• Monitor that all ad sets are spending their allocated budget and exiting the learning phase
• Verify that conversions are being tracked correctly across all ad sets

Weeks 2–4: Evaluation

• Compare ROAS, new customer share, and AOV across ad sets
• Identify which buckets outperform broad targeting and by how much
• If a bucket underperforms broad consistently, it can be paused

After Week 4: Dynamic Scaling Once you have sufficient data (2–4 weeks minimum), begin scaling:

• Increase budget on the best-performing Innkeepr ad sets incrementally (20–30% per increase, not more)
• Measure how much additional spend the Innkeepr audiences can absorb while maintaining comparable performance to broad
• The goal is to determine what share of total campaign spend Innkeepr audiences can cover at equal or better performance than broad targeting
• Continue monitoring after each budget increase — allow 3–5 days for Meta’s algorithm to stabilize before evaluating

Scaling question to answer: How much additional spend can Innkeepr audiences absorb compared to broad targeting, at the same or better CPA/ROAS? This determines the incremental budget capacity that Innkeepr unlocks for your Meta campaigns.

​

Best Practices

Creative Isolation The test is only valid if creatives are identical across all ad sets. If you want to test new creatives or add additional creatives during the test, add them to all ad sets at the same time. Never add creatives to only some ad sets — this makes performance differences between ad sets unattributable. Lookalike Size Selection The recommended LAL sizes (1%, 5–10%, 10%) are starting points. Depending on your market size and target geography, you may need to adjust:

• Smaller markets (e.g. single country, DACH): start with 1%, 3%, 5%
• Larger markets (e.g. EU-wide, US): 1%, 5–10%, 10% works well

​

Troubleshooting

Audiences Not Populating All Buckets Underperform Broad Performance Drops After Sales Phases Performance drops can coincide with sales events (e.g. Black Friday, seasonal promotions). During and after these phases, audience composition and conversion patterns shift temporarily. Innkeepr typically detects these phases automatically and adjusts the underlying models accordingly. If a performance drop persists for more than 1–2 weeks after a sales event has ended, reach out to the Innkeepr team to verify that the model has recalibrated correctly.