> ## Documentation Index
> Fetch the complete documentation index at: https://docs.innkeepr.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Signal Overview

## What Are Signals?

Signals are predictions derived from Innkeepr's **causal models** and come into two different forms:

* [Audience Signals](/guides/signals-and-activation/audience-signals) – Predictive user segments for targeting, lookalikes, and exclusions
* [Conversion Signals](/guides/signals-and-activation/conversion-signals) – Incrementality-weighted conversion values to enhance bidding systems

Both signal types are powered by Innkeepr's **causal engine**, which estimates the true incremental impact of your marketing touchpoints.

## How Signals Work

**Data** <br />
Innkeepr collects behavioral data through client-side tracking ([innkeepr.js](/getting-started/getting-started-guide#install-innkeepr-js), server-side APIs, and warehouse connections. This data is processed into time-series user sessions containing interaction sequences, temporal context, marketing exposure, and user attributes.

**Objectives** <br />
Signals are optimized for specific business outcomes defined through objective functions - the measurable results you want to drive, they are called *Objectives*

Common objectives include:

* **Conversion likelihood** – Binary outcomes like purchases, sign-ups, or installs
* **Customer lifetime value (CLV)** – Predicted value over 90 or 180 days
* **Lead qualification** – MQL probability based on engagement and firmographic data
* **Custom objectives** – Any measurable outcome from tracked events or calculated traits

Objectives can be filtered by additional traits, such as "CLV for new customers in EU markets" or "7-day retention from paid social traffic."

<Info>
  Learn More

  * Objectives
</Info>

**Treatments & Attribution** <br />
Treatments represent marketing touchpoints - ad clicks, email opens, campaign landing page visits. Innkeepr uses **7-day first-touch** attribution to assign treatment exposure to user sessions based on UTM parameters, referrer data, and platform click IDs.

Treatments map to platform-native campaign structures: ad sets (Meta), ad groups (Google/TikTok), and asset groups (Performance Max). This granularity reveals which specific campaign elements drive the most **incremental lift**.

Treatment data is the foundation for signal learning. The causal engine observes which users were exposed to which treatments and what outcomes they achieved, then estimates individual treatment effects - the predicted incremental impact of a marketing treatment on a user's likelihood to reach an objective.

Treatment effects answer: *"If this user sees this ad, how much will their conversion probability increase compared to baseline?"*

## Signal Types

**Audience Signals** <br />
Audience signals are predictive user segments built from **treatment effect** scores. They identify users most (or least) likely to deliver incremental value from marketing exposure.

**Use cases:**

* **Seed audiences** – High-incrementality users (top 10-20%) for lookalike expansion
* **Exclusion audiences** – Low-incrementality users to suppress wasted spend
* **Retargeting audiences** – Users predicted to respond incrementally to re-engagement

<Info>
  Learn more

  * [Audience signals](/guides/signals-and-activation/audience-signals)
</Info>

**Conversion Signals** <br />
Conversion signals adjust conversion values by incremental contribution. Instead of passing raw transaction amounts to platform algorithms, Innkeepr weights each conversion based on its predicted lift over baseline.

**How it works:** When a €100 conversion occurs and the predicted treatment effect indicates 60% **incremental lift**, the signal passes €60 as the incrementality-weighted value. This trains bidding algorithms to prioritize users and touchpoints that deliver true causal impact.

<Info>
  Learn more

  * [Conversion signals](/guides/signals-and-activation/conversion-signals)
</Info>

**Signal Quality Over Time** <br />
As more **behavioral data** flows through Innkeepr, treatment effect models become more precise. Early implementations may have wider confidence intervals; after 2-3 model training cycles (4-6 weeks), predictions stabilize and signal quality improves measurably.

**Monitoring** <br />
Innkeepr tracks signal health through **prediction accuracy**, audience overlap rates, conversion signal delivery rates, and model confidence scores.