All the algorithms you need for real-time behavioral engagement
Our ecosystem of dynamic and behavioral algorithms covers all the bases, helping you understand, predict, explore, and continuously learn from customer behavior in real time — so every engagement is adaptive, contextual, and optimized for impact.
What is an algorithm?
An algorithm is a well-defined step-by-step procedure that takes inputs and produces outputs by specifying what is being computed, how success is measured, and how the result is achieved. The concept traces back to the work of the 9th-century mathematician al-Khwārizmī, whose writings laid the foundations for modern arithmetic and computation. In customer engagement, traditional algorithms often optimize for short-term outcomes like clicks or conversions, but they miss the complexity of real human decision-making.
Behavioral algorithms
Behavioral algorithms are dynamic decisioning systems that optimize customer engagement by modeling real human behavior rather than relying solely on historical performance.
They combine behavioral science and machine learning to decide what to show, to whom, and when — balancing exploration and exploitation to continuously learn from interaction data. Behavioral algorithms help you build customer engagement systems that understand how people make decisions— and adapt in real time to improve relevance, trust, and long-term value.
Algorithm catalog
Spend Personality
Analyzes a customer’s transaction behavior to identify spending patterns (frequency, categories, and value distribution). It uses these signals to classify customers into behavioral spending archetypes, reflecting their spend personality.
Learn MoreMoney Personality
The Money Personality algorithm models a customer’s broader financial behavior, including saving, credit risk, and debt. It creates a profile that describes how a customer manages and relates to money over time.
Learn MoreDigital Personality
The Digital Personality algorithm interprets a customer’s historical digital interactions across touchpoints. It maps these digital interaction patterns into archetypes that reflect how the customer typically engages with digital experiences.
Network Analysis
Looks at how offers are connected to each other based on how people interact with them. Over time, it builds a network of these relationships. This means that offers that are strongly connected to other successful offers naturally rise to the top.
Bayesian Probabilistic
A Naive Bayes–based model that predicts outcomes by calculating the probability of a user–offer match based on historical patterns across features. It learns from past data distributions and applies Bayesian inference to make the most likely prediction in new situations.
Generative Model
The generative model algorithm uses a large language model to score and rank offers by sending customer history and context to an external API and receiving back scores for each option. These scores are then used like any other algorithm to decide which offers to show.
Risk Aversion
The risk aversion algorithm scores offers based on how reliable their past performance is, not just how well they perform on average. It lowers the score of offers with uncertain or highly variable outcomes, favoring more consistent ones even if they have slightly lower success rates.
Sentimental Equilibrium
The sentimental equilibrium algorithm models customer engagement as a balance between the value gained from interacting and the effort or cost of that engagement. It aims to find a steady state where the benefit of engagement equals the cost, representing an optimal long-term engagement level.
Epsilon Greedy
The simplest multi-armed bandit strategy for dynamic interactions. It mostly picks the offer that seems best based on past data, but occasionally tries something random to explore new possibilities. This way, it balances exploiting what works with testing new options.
Coverage-Aware Thompson
A version of Thompson Sampling that ensures less-seen (long-tail) offers still get exposure. It uses past performance to guide selection but boosts under-exposed items so popular and niche offers are fairly balanced.
Prospect Theory
Prospect Theory models how people evaluate risk, where losses feel stronger than gains and probabilities are often distorted, ensuring fair and effective traffic allocation from day one.
Long-Tail Boost MF
The long-tail is a business and statistical strategy of selling a large number of unique, niche items in small quantities, rather than relying solely on a few bestsellers. Long-Tail Boost Matrix Factorization uses collaborative filtering to uncover patterns across users and items, while boosting less popular options.
Ecosystem Rewards (Thompson Sampling)
Ecosystem Rewards is the system’s core learning algorithm, based on Thompson Sampling. It balances exploration and exploitation automatically — favouring high-performing offers while still testing alternatives.
Q-Learning
A reinforcement learning method that learns which offers to show each customer by assigning scores based on past interactions and reward functions. It improves over time by trying different offers, observing customer responses, and updating its decisions to maximize expected reward.
Loss Aversion
People feel losses more strongly than gains. This algorithm prioritizes safer options and avoids showing offers likely to be rejected.
Developer resources
Documentation and next steps
Technical detail lives in the developer docs; strategy connects on the Prediction Platform.
Algorithms overview (developer docs) (opens in new tab)Dynamic recommender user guide (opens in new tab)Personality algorithms (opens in new tab)
Frequently asked questions
Quick answers derived from the same product story as the rest of this page.
They are dynamic decisioning systems that optimize engagement by modeling real human behavior—not only historical click or conversion performance. They combine behavioral science and machine learning to decide what to show, to whom, and when, balancing exploration and exploitation so the system keeps learning from live interaction data.
Each tag describes the kind of job the algorithm does: Analytics builds understanding from networks and behavioral profiles; Predict scores likely outcomes; Decision uses models (including LLM-backed scoring) to rank options; Behavioral encodes how people perceive risk, effort, and loss; Exploration balances trying new offers with exploiting what works; Learning updates policies from rewards and feedback over time.
This page lists 15 named algorithms. You can compose them on the Prediction Platform with Dynamic Experimentation and real-time recommenders so each journey uses the right mix.
Start with the algorithms overview and dynamic recommender user guide on developer.ecosystem.ai, linked in the Documentation section below. Personality-focused modules are also described on ecosystem.ai.
Yes. Spend Personality, Money Personality, and Digital Personality sit beside probabilistic scorers, exploration strategies, and reinforcement-style learners on the same platform APIs and experimentation paths—so you can combine understanding, prediction, exploration, and learning in one engagement system.