Data Foundations & Time-Series Predictions
Innkeepr collects and processes first-party event data via the innkeepr.js client library. This data includes both known and anonymous users, and captures rich behavioral signals across websites, shops, mobile apps, and backend systems. These events are transformed into time-series user sessions, enabling Innkeepr’s predictive models to learn from patterns over time. // Image mit user JourneyObjective Functions
Each signal type is generated based on a configurable objective function, which defines the outcome you want to optimize for. Common objectives include:- Conversion likelihood (e.g. purchase intent, signup probability)
- Customer lifetime value (CLV)
- Lead qualification score (MQL probability)
Treatments
Treatments are media-driven user interactions - such as ad clicks, email opens, or landing page visits and a core concept of Innkeepr. Innkeepr uses a 7-day first-touch attribution window across all platforms to assign treatment exposure to resulting conversion events. This design supports a clean alignment with acquisition objectives and makes treatment assignment consistent across Meta, Google Ads, TikTok, and additional sources. Treatments are stored and mapped using deterministic logic using common UTM parameters, ensuring high attribution quality even for anonymous users. // Image mit erweiterung der user Journey um forecast auf objective function und treatments entlang der journeyTreatment Effects & the Causal Engine
At the core of Innkeepr is a causal inference engine that estimates individual treatment effects: the predicted impact of a marketing treatment on a given user’s likelihood to reach an objective. Key features:- Forecasts the incremental lift from a specific ad or channel on conversion probability or CLV
- Scores each user (anonymous or known) for predicted treatment response
- Continuously updated in bi-weekly training cycles to account for seasonality, channel shifts, and behavior change
Signals & Incrementality
Audience Signals
Audience Signals are predictive user segments built from treatment effect scores. They identify users most (or least) likely to respond to marketing, enabling efficient targeting, suppression, and seed audience generation.- Built by segmenting users based on treatment effect scores
- Used for lookalike seeding, spend suppression, or churn prevention
- Uploaded as audiences into Meta, Google, TikTok, and CRM tools
- Refreshed every 1–3 days depending on audience type
Conversion Signals
Conversion Signals are adjusted conversion values that reflect the causal impact of an ad. They help platforms prioritize users who converted because of an ad, not just alongside it, enabling smarter, lift-optimized bidding.- Adjusted conversion events weighted by the user’s treatment effect score
- Used to inform bidding algorithms about the incremental value of a conversion
- Compatible with Meta Conversions API, Google Enhanced Conversions, and TikTok Events API
- Delivered via server-side batch postbacks