Last Monday, on November 13, at the DevDay event, Sam Altman unveiled GPT-4, providing developers

with customization access to ChatGPT. Developers can now access web browsing, OpenAI's DALL-E,
Code Interpreter, and Actions feature for external access like email and databases.

Did you know this release opens doors for your company to build its Internal AI?
If you're in the financial industry, you can create AI for various business cases, such as:

1. Automated Document Processing

Automate the extraction and analysis of information from unstructured documents. This can
streamline tasks such as document verification, loan application processing, and compliance
checks.

2. Predictive Maintenance for ATMs

Forecast potential issues with ATMs. This helps in scheduling proactive maintenance, reducing
downtime, and improving overall ATM availability.

3. Anomaly Detection for Transaction Monitoring

Detecting anomalies in transaction patterns. This can enhance fraud detection by identifying
unusual activities and triggering alerts for further investigation.

4. Dynamic Credit Scoring

Implement a dynamic credit scoring system that continuously updates a customer's credit score
based on their real-time financial behavior. This provides a more accurate representation of
creditworthiness.

5. AI-Enhanced Financial Advisory Services

Analyzes a customer's financial data, spending patterns, and goals to provide personalized
investment advice, budgeting tips, and financial planning recommendations.

6. Automated Invoice Processing for Business Clients

Automates the processing of invoices for business clients. This can include extracting relevant
information, matching invoices with purchase orders, and facilitating seamless payment
processes.

7. Credit Limit Optimization

Develop a system that dynamically adjusts credit card limits based on the customer's financial
behavior, credit score changes, and spending patterns. This ensures that customers have
appropriate credit limits aligned with their financial situations.
STEP 1
Data Collection
Collecting data from various data points is the initial step. The gathered data
should align with the business case you intend to develop. For creating an AI
Credit Limit Optimization, essential data points include:

Credit Score Spending Patterns

Income Information Credit Utilization
Current Debt Obligations Length of Credit History
Payment History Recent Credit Inquiries

Additional data points enhance AI accuracy. In this phase, you require IT talents
such as Data Engineers for the ETL process. Some data may need extraction from
external sources, necessitating developers to facilitate data collection through
APIs, scraping, FTP, etc.
STEP 2
Choose Frameworks
Select suitable AI frameworks tailored to your needs. Assuming you aim to develop AI for
Credit Limit, here are 5 recommended framework options:

1. TensorFlow
TensorFlow, developed by Google, is a versatile and widely-used open-source
machine learning framework. It's suitable for building complex neural network
models, making it applicable to tasks like credit scoring and predicting credit limits.

2. PyTorch
PyTorch, developed by Facebook, is known for its dynamic computational graph
and ease of use. It's well-suited for tasks where flexibility in model architecture and
dynamic adjustments are crucial, making it suitable for credit limit optimization
models.

3. scikit-learn
scikit-learn is a popular and user-friendly machine learning library in Python. While
it might not be as deep or specialized as some deep learning frameworks, it is
excellent for implementing traditional machine learning algorithms, such as
decision trees or support vector machines, which can be effective in credit scoring.

4. XGBoost
XGBoost is an efficient and scalable open-source library for gradient boosting. It's
particularly powerful for building ensemble models and has been successful in
various machine learning competitions. XGBoost can be applied to credit scoring
and risk assessment tasks.

5. LightGBM
LightGBM is another gradient boosting framework developed by Microsoft. It's
known for its speed and efficiency, making it suitable for large datasets and tasks
such as credit risk modeling where performance is critical.

The choice of framework also needs to consider the availability of talent. The abundance
of Python Developers makes it much easier to recruit and train with additional tools.
STEP 3
Develop Model
After collecting and processing the data, the next step is to create the model.
Assuming you want to create AI for Credit Limit, here are 3 recommended choices for building
the Machine Learning algorithm :

1. Gradient Boosting Machines (GBM)

GBM is an ensemble learning method that builds typical decision trees and combines
their predictions to create a strong predictive model. GMB handles complex relationships
in data, provides high predictive accuracy, and is suitable for handling both numerical
and categorical features.

2. Neural Networks (Deep Learning)

Neural networks, especially deep learning architectures, can capture intricate patterns
and relationships in data. Neural Networks excel in capturing non-linear relationships and
might adapt well to changing customer behaviors.

3. Random Forest
Random Forest is an ensemble learning method that builds multiple decision trees and
merges their outputs. Random Forests handle both categorical and numerical features
effectively.

At this stage, you need three essential teams:

Domain Expert to provide valuable insights of various data points, input on machine
learning model, dan validation and evaluation of machine learning models

Data Scientist to Develop and fine-tune machine learning models, analyze data
patterns, and optimize algorithms

ML Engineer to Focus on the deployment and integration of machine learning

models into the banking systems.
STEP 4
Integration
After training the created model, the next step is integrating it into your company's system. In the
context of the AI Credit Limit mentioned above, here is the integration process you should undertake:

A. Model Export E. Security Measures

Export the trained machine learning Use encryption protocols to secure data
model into a format that can be easily in transmission between systems and
tegrated into the banking systems. ensure that access to the model is
restricted to authorized users.
B. API Development
Create an Application Programming F. Monitoring and Logging
Interface (API) that exposes the Set up monitoring and logging
functionality of the machine learning mechanisms to track the performance of
model. the deployed model.

C. Model Deployment G. Scalability Considerations

Deploy the model on cloud platforms (e.g., Implement load balancing and scaling
AWS, Azure, GCP) or on-premises servers, strategies to handle increased demand.
depending on the bank's infrastructure
and security requirements. H. Integration Testing
Conduct thorough integration testing to
D. Data Input Handling ensure that the deployed model interacts
Develop mechanisms to handle input correctly with the banking systems. Test
data from the banking systems. This different scenarios, edge cases, and
includes preprocessing the input data to inputs.
match the format expected by the
machine learning model.

At this stage, you need at least a few teams:

1. Developers, to deploy machine learning models into production. And developing APIs
for model integration into banking systems.

2. DevOps Engineer, to :
- Manage the deployment infrastructure and environment.
- Implement security measures for data and model protection.
- Monitor and optimize system performance.
- Setup continuous integration and deployment (CI/CD) pipelines.

3. QA Engineer, to conduct through integration testing of the deployed model.

How long does it take? With the context of AI Credit Limit Optimization above, the estimated time
you need is as follows:

1. Machine Learning Model Development

- Developing, training, and fine-tuning machine learning models can take several
weeks to a few months, depending on the complexity of the models and the size
of the dataset.
- Estimated Time: 3 - 6 months

2. Data Engineering and Infrastructure

- Setting up data infrastructure, including databases and data pipelines, may take
a few weeks to a couple of months.
- Estimated Time: 1 - 2 months

3. Integration and Deployment

- Integrating the model into banking systems, developing APIs, and deploying the
solution could take a few weeks to a few months.
- Estimated Time: 2 - 4 months

4. Security Measures
- Implementing security measures may require a few weeks to a couple of months,
depending on the complexity of the security protocols.
- Estimated Time: 1 - 2 months

5. Testing and Quality Assurance

- Thorough testing, including integration testing and quality assurance efforts, may
take several weeks to a few months.
- Estimated Time: 2 - 4 months

6. DevOps and Infrastructure Scaling

- Setting up DevOps practices, infrastructure scaling, and implementing continuous
integration and deployment pipelines might take a few weeks to a couple of
months.
- Estimated Time: 1 - 2 months

7. Project Management and Documentation

- Project management and documentation efforts will be ongoing throughout the
project but may require dedicated time during specific phases.
- Estimated Time: Throughout the project

Total Estimated Development Time: 8 Months - 2 years. This time is influenced by several factors
such as the complexity of the built AI, the availability of the team, and the readiness of the internal
company itself.
IT Outsourcing - Cybersecurity - Analytics - IT Supports

ronald@akasia.id www.akasia.id

Sovereign Plaza 21th Floor, TB Simpatupang - Jakarta Selatan