Customer Scoring

Summary¶

In the realm of financial management, accurately predicting customer scores is essential to providing a reliable customer segmentation, risk analysis and targeting tool. Dedomena AI's Customer Scoring Neuron leverages advanced artificial intelligence techniques to predict customer scores based on transactional historical data, offering an innovative and efficient way to assess risk, obtain credit rating, understand solvency, optimize marketing strategies and improve customer relationships with relevant insights.

This artificial intelligence system is a cutting-edge solution for credit scoring, payment default risk assessment, and user evaluation based only on transactional banking data. It simplifies the input requirements by only needing at least three months of transactional history, with optimal reliability achieved at 12 months. By analyzing customer behavior, preferences, spending and income patterns and interactions, it assigns scores that help identify customer potential, with results ranging from 5 (Top) to 2 (Low) for regular users and 1 for potentially fraudulent users. This ensures a robust and reliable classification system while also providing insights into potential credit approval and associated terms.

The innovation lies not only in the system's technical capabilities and superior performance but also in its focus on data privacy and usability. With proprietary anonymization techniques, explainable AI, and the ability to predict future balances and default risks, the system redefines the standards for customer scoring in the financial sector. Moreover, it integrates a credit approval analysis to provide detailed insights into loan eligibility and potential terms, making it a comprehensive tool for financial institutions.

Warning

This classification is based on the probability of a user being in a certain level. It is essential to consider other factors and perform additional checks to confirm the user's score. Also, this algorithm runs on the fly, Dedomena AI does not store any data. You must copy the results to avoid losing the information.

Say goodbye to uncertainty in customer scoring. Embrace the power of Dedomena's Customer Scoring Neuron and make informed decisions with confidence.

Key Properties

The system's functionality is built on several key innovations:

Simplified Input Requirements: Unlike other models, this system only requires at least three months of transactional banking data from all active accounts of the individual being assessed. Optimal performance is achieved with 12 months of data. The system cab ne connected securely to company or entity data sources, being implemented on-premise or using Dedomena's PSD2-compliant connectors.
Proprietary Anonymization: Sensitive data is protected through advanced, proprietary anonymization layers that ensure data security.
Ensemble Models: The data is processed through an ensemble of intelligent models that extract insights, categorize transactions into income and expense categories, segment users, and estimate future balances. These architectures are the result of years of research and provide superior performance.
Credit Scoring and Default Risk Models: The final scoring is determined by two proprietary models—one for credit scoring and another for payment default risk, which requires the requested loan amount. These models deliver explainable results, giving financial institutions confidence in their decision-making processes.
Scalability and Updates: The system ensures scalability and seamless integration through an API. It is designed for biannual updates with minimal steps to keep the models and algorithms current.
Automation and Explainability: Best practices in machine learning lifecycle management have been applied, including bias reduction through synthetic training data. Explainability tools provide transparency into the scoring and risk assessment processes.

Use Cases

Here are the key use cases for the Customer Scoring Neuron:

Credit Scoring and Approval: Assessing an individual’s creditworthiness and determining the terms of a potential loan based on their transactional history and scoring insights.
Risk Management: Evaluating the likelihood of payment default and taking proactive measures to mitigate potential losses.
Marketing Strategies: Optimizing marketing campaigns by targeting customers based on their scores, preferences, and behavior.
Customer Segmentation: Identifying customer segments and tailoring products and services to meet their specific needs.
Customer Relationship Management: Enhancing customer relationships by providing personalized experiences and offers based on their scores and resulting insights.
Predictive Insights: Generating insights into future balances and spending behavior to inform financial planning and decision-making.
Fraud Detection: Identifying potentially fraudulent users with a dedicated scoring level (1) and recommending actions to safeguard business operations.
Business Growth: Driving strategic growth by leveraging customer scores to improve marketing effectiveness, increase customer satisfaction, and expand financial services.

Dedomena AI’s Customer Scoring Neuron is the ultimate solution for financial institutions, enterprises, retailers, fintechs, among other companies in diferent sectors, seeking a comprehensive, secure, and innovative tool for customer evaluation and risk assessment.

API & Endpoints¶

/v3/score¶

POST v3/score

This endpoint classifies users into 4 categories: 5 (top), 4 (high), 3 (medium) and 2 (low).

Headers:

token: available via free trial or subscription.

Parameters:

user_type (str): Type of user. Right now the only possible value is PERSON.
user_id (str): Unique anonymous id that identifies the user.
user_created (str): Date and time when the user was created.
days_threshold (int): Number of days to consider for the scoring. Default is 30. This parameter is optional.
txns_threshold (int): Number of transactions to consider for the scoring. Default is 2. This parameter is optional.
credit_amount (float): Amount of the credit requested. This parameter is optional and by default is 10000.
credit_term (int): Term of the credit or amount of months to pay the credit. This parameter is optional and by default is 24.
tin (float): Total Interest on the credit. This parameter is optional and by default is 3.5.

Request Body:

It should be a JSON array with the following fields:

account_id (str): Unique anonymous id that identifies the account.
transaction_currency (str): Currency of the transaction.
user_balance (float): User balance at the time of the transaction.
transaction_amount (float): Amount of the transaction.
concept (str): string with a description/concept of the transaction.
transaction_date (str): Date and time of the transaction.

Request Body - JSON Array¶

[
    {
        "account_id": "account_123",
        "transaction_currency": "USD",
        "user_balance": 1000.0,
        "transaction_amount": 50.0,
        "concept": "Deposit",
        "transaction_date": "2024-06-30T12:00:00"
    },
    {
        "account_id": "account_123",
        "transaction_currency": "USD",
        "user_balance": 900.0,
        "transaction_amount": -100.0,
        "concept": "Withdrawal",
        "transaction_date": "2024-06-30T12:30:00"
    }
]

Response¶

200

[
    {
        "user_id": "user_123",
        "pred_category": 5,
        "pred_category_name": "Top",
        "probability_low": 8.9357,
        "probability_medium": 0.0005,
        "probability_high": 0.0111,
        "probability_top": 0.9882,
        "insights": {},
        "credit_info_user": "Elegible client. Regular activity",
        "loan_approved": True,
        "loan_term": 24,
        "monthly_payment": 250.0,
        "monthly_taxes": 20.5,
        "payment_probability": 0.92
    }
]

Description:

The response will include the predicted category for each user, along with the probabilities of being in each category:

user_id (str): Unique anonymous id that identifies the user.
pred_category (int): Predicted category of the user (2: Low, 3: Medium, 4: High, 5: Top).
pred_category_name (str): Predicted category name of the user (Low, Medium, High, Top).
probability_low (float): Probability of the user being in the low category.
probability_medium (float): Probability of the user being in the medium category.
probability_high (float): Probability of the user being in the high category.
probability_top (float): Probability of the user being in the top category.
insights (dict): Additional insights about the user.
credit_info_user (str): Credit information for the user.
loan_approved (bool): Loan approval status.
loan_term (int): Loan term or amount of months to pay the approved credit.
monthly_payment (float): Monthly total payment for the approved credit.
monthly_taxes (float): Monthly taxes for the approved credit.
payment_probability (float): Payment probability for the approved credit.

400

[
    {
        "detail": "Data not valid"
    }
]

401

[
    {
        "detail": "Could not validate credentials"
    }
]

405

[
    {
        "detail": "User type not allowed"
    }
]

406

[
    {
        "detail": "the number of observations in the payload exceed the available quote"
    }
]

422

[
    {
        "detail": "Unprocessable Entity"
    }
]

429

[
    {
        "detail": "The maximum monthly number of calls or predictions for your account has been exceeded"
    }
]

Metrics¶

Dedomena AI model's performance metrics ensure high accuracy and reliability:

Summary:

Accuracy	Precision	Recall	F1 Score
0.9630	0.9614	0.9580	0.9596