Long Short-Term Memory (LSTM) is a type of recurrent neural network (RNN) designed to process sequences of data. Unlike basic models that treat events independently, LSTM networks maintain memory over long input sequences making them ideal for modeling user journeys where the order and timing of touchpoints matter.
LSTM models can learn how:
Early and late touchpoints interact
Time gaps between events affect conversion likelihood
Repeated exposures to certain channels influence outcomes
This makes them especially well-suited for multi-touch attribution, where predicting conversions from complex and varied sequences is essential.
Why Funnel uses LSTM for attribution
Funnel applies LSTM modeling to evaluate how marketing activities contribute to conversions over time. The model uses user-level data, capturing both converting and non-converting journeys, to understand what truly drives performance.
Input data and features
Touchpoint sequences: Clicks, impressions, page visits across sources and channels
Timing: Time gaps between events
Conversion signals: Whether the journey led to a conversion
User-level features: When available, such as device or campaign metadata
Model outputs
Conversion probability for each journey
Attribution weights for each touchpoint, based on its predicted contribution
This allows the model to assign credit based on observed impact, rather than fixed rules.
Modeling interactions and time sensitivity
Unlike rule-based models that assume static weights, LSTM models learn from real behavior:
A branded search click might get more credit only when it follows awareness-building activity.
A display ad seen three days before conversion may be less influential than one seen 30 minutes before.
LSTMs recognize these nuances by learning how sequence, timing, and interaction patterns shape outcomes.
Attributing based on customer lifetime value (CLV)
Funnel extends its LSTM modeling to account not just for conversions, but for long-term customer value:
Touchpoints that lead to repeat purchases receive higher credit.
Attribution is weighted based on predicted CLV, not just immediate conversion.
This CLV-based attribution helps you prioritize investments that lead to sustained growth, not just quick wins.