Loss Aversion

Based on the behavioral economics principle that people feel losses roughly twice as strongly as equivalent gains, this algorithm amplifies the perceived failure rate for under-performing offers, making the system more aggressive about avoiding offers that customers tend to reject. It combines a loss-adjusted learned probability with an Upper Confidence Bound (UCB) exploration term.

Algorithm

Config value: "approach": "behaviorAlgos", "sub_approach": "lossAversion"

Training phase:

For each offer, the raw acceptance rate is adjusted using the loss aversion factor:

\(\text{adjustedRate} = \begin{cases} \text{rawRate} & \text{if rawRate} \geq 0.5 \\ \text{rawRate} \times \text{lossAversionFactor} & \text{if rawRate} < 0.5 \end{cases}\)

An Upper Confidence Bound exploration term provides additional exploration for under-sampled offers:

\(\text{UCB} = \text{adjustedRate} + \text{explorationDecay} \times \sqrt{\frac{2 \ln(\text{totalCount})}{\text{count}}}\)

Where explorationDecay is 1.0 for under-performing offers (rawRate < 0.5) and \((1 - \text{rawRate})\) otherwise.

Scoring phase:

\(\text{finalScore} = \text{ucbWeight} \times \text{learnedProb} + (1 - \text{ucbWeight}) \times \text{UCB}\)

Scores are then normalized across all offers to sum to 1.0.

Parameters

• lossAversionFactor: Multiplier for under-performing offers (rawRate < 0.5). Default: 2.0. Higher values penalize poor performers more aggressively.
• ucbWeight: Balance between learned probability and UCB exploration. Default: 0.5. Range: 0.0 (pure UCB) to 1.0 (pure learned probability).
• explorationFactor: Controls UCB exploration term magnitude. Default: 2.0.
• Processing Window: Time window in milliseconds for historical data.
• Historical Count: Max records to process per update cycle.

Cold Start

Recommendations are always returned. The real-time training path in RollingBehavior always produces a scored options array, regardless of whether the algorithm has interaction history:

• No history: Every offer in the options store receives a uniform random score. All offers are ranked and passed to the post-score class.
• Early history: New offers receive a smoothing alpha of 1.5 added to their first observation, preventing extreme initial scores. The UCB exploration term gives under-sampled offers a natural boost.
• Partial history: Offers scored by Loss Aversion use the loss-adjusted probability blended with UCB; unscored offers receive a random score. Both are included in the result.

The scored options are then sorted by arm_reward and handed to the configured dynamic post-score class, which controls the final offer selection and response formatting.

When To Use

• When the cost of showing a rejected offer is high (customer churn risk)
• When you want the system to quickly stop recommending poor performers
• When you prefer conservative recommendations
• Financial services, insurance, or high-stakes offer environments

When NOT To Use

• When you want to give new or unpopular offers a fair chance (use Coverage-Aware Thompson)
• When acceptance rates are naturally low across all offers
• When acceptance rates are naturally low across all offers (the loss aversion factor may over-penalize all arms equally)

Example

from prediction.apis import deployment_management as dm
from prediction.apis import online_learning_management as ol
from prediction import jwt_access
 
auth = jwt_access.Authenticate("http://localhost:3001/api", ecosystem_username, ecosystem_password)
 
deployment_id = "demo-loss-aversion"
 
online_learning_uuid = ol.create_online_learning(
        auth,
        algorithm="ecosystem_rewards",
        name=deployment_id,
        description="Loss Aversion configuration",
        feature_store_collection="set_up_features",
        feature_store_database="my_mongo_database",
        options_store_database="my_mongo_database",
        options_store_collection="demo-deployment_options",
        randomisation_processing_count=5000,
        randomisation_processing_window=604800000,
        contextual_variables_offer_key="offer",
        create_options_index=True,
        create_covering_index=True
)
 
online_learning = dm.define_deployment_multi_armed_bandit(epsilon=0, dynamic_interaction_uuid=online_learning_uuid)
 
parameter_access = dm.define_deployment_parameter_access(
    auth,
    lookup_key="customer_id",
    lookup_type="string",
    database="my_mongo_database",
    table_collection="customer_feature_store",
    datasource="mongodb"
)
 
deployment_step = dm.create_deployment(
    auth,
    project_id="demo-project",
    deployment_id=deployment_id,
    description="Loss Aversion demo deployment",
    version="001",
    plugin_post_score_class="PlatformDynamicEngagement.java",
    plugin_pre_score_class="PreScoreDynamic.java",
    scoring_engine_path_dev="http://localhost:8091",
    mongo_connect=f"mongodb://{mongo_user}:{mongo_password}@localhost:54445/?authSource=admin",
    parameter_access=parameter_access,
    multi_armed_bandit=online_learning
)