AI-900 Fundamentals

🎓 Don't Forget Your Learning Badge!

Congratulations on completing your study! You can redeem your learning badge here to showcase your achievement.

Describe Artificial Intelligence workloads and considerations (15–20%)

Identify features of common AI workloads

Identify computer vision workloads

Computer vision workloads use AI to process and extract useful information from images and videos. In finance, this could mean analyzing scanned documents, receipts, or visual records.
A key computer vision workload is Optical Character Recognition (OCR), which automatically reads text from images, such as printed or handwritten amounts on invoices or checks. This helps turn paper-based information into digital, searchable data.
Image analysis workloads can detect objects, faces, and even inappropriate content in financial media. For example, software can classify and describe items in a scanned contract or identify the presence of certain logos within images for compliance checking.
Face recognition can be used to enhance security and identity verification, ensuring only authorized employees access sensitive financial data or accounts.
Tagging and describing image content makes storing and searching financial documents faster and easier, as images become indexed with relevant keywords and summaries automatically.

Example: A finance company uses Azure AI Vision OCR to automatically read and digitize handwritten dollar values from scanned checks, saving hours of manual data entry and reducing errors.

Use Case: A bank uses computer vision to scan and analyze thousands of customer-submitted documents. AI automatically extracts key information (like names, account numbers, amounts) from these documents and checks for security features (like signatures and logos), helping speed up loan application processing and compliance reviews.

For more information see these links:

Identify natural language processing workloads

Natural language processing (NLP) workloads involve using AI to understand and manage human language, such as emails, documents, messages, or chat transcripts. In finance, this helps automate tasks that require reading and interpreting large volumes of text data.
Common NLP tasks in finance include sentiment analysis (identifying positive or negative tone in financial news or earnings reports), entity recognition (extracting names of companies, financial products, or regulatory bodies from documents), and document classification (categorizing emails, contracts, or compliance reports).
NLP workloads can use prebuilt models provided by platforms like Azure AI Language for common tasks, or customized frameworks like Spark NLP on Azure Databricks for more advanced or large-scale text processing needs. These tools make it easier to automate complex tasks, extract valuable insights, and support better decision-making.

Example: A financial analyst uses Azure AI to scan daily news articles and automatically flag any news mentioning their company’s name or major competitors. The system uses entity recognition to extract company names and sentiment analysis to highlight negative news that may require further attention.

Use Case: A compliance team processes thousands of customer emails daily looking for signs of fraudulent activity or regulatory risk. By applying NLP on Azure Databricks with Spark NLP, they automate the detection of risky terms or suspicious requests, helping prioritize which emails need manual review.

For more information see these links:

Identify document processing workloads

Document processing workloads involve using AI tools to automatically read, extract, and organize data from financial documents. This can include bank statements, invoices, tax forms, and identification documents.
Common document processing tasks include digitizing paper-based records using OCR (Optical Character Recognition), extracting key information like names, dates, and amounts, and classifying documents by type and content.
In finance, document processing helps save time, reduce manual errors, and ensure data consistency by automating repetitive tasks such as data entry and form review. AI solutions like Azure Document Intelligence and AI Builder offer tools with prebuilt models tailored for finance.
Choosing the right tool depends on your document types and business needs. For well-structured financial forms with consistent templates, prebuilt models work best. More complex or diverse documents may require custom AI workflows using tools like Azure AI Content Understanding or Azure OpenAI.
Document processing workloads often include compliance, security, and review steps to protect sensitive financial information. AI systems integrate role-based access, encryption, and human review features to help meet industry standards and regulations.

Example: A bank receives loan application forms from customers. Using Azure Document Intelligence, the bank automatically scans and extracts key information such as applicant name, address, income, and identification number, eliminating manual data entry and speeding up application review.

Use Case: A new employee at a financial firm uses an AI-powered document processing solution to automate customer identity verification. When a customer uploads their driver’s license, the system extracts details like name, date of birth, and license number, securely stores the data, and checks it against compliance requirements—all without manual intervention.

For more information see these links:

Identify features of generative AI workloads

Generative AI creates new and original content, such as text, images, or code, rather than simply analyzing or classifying existing data. In finance, this could mean generating financial reports, summaries, or customer messages automatically.
These models use large amounts of training data to learn patterns, which allows them to produce outputs that are unique and contextually relevant. For example, a generative AI trained on financial documents can answer customer questions about investments in natural language.
Generative AI workloads are often delivered through managed cloud services like Azure OpenAI Service, which provide ready-to-use models and interfaces, simplifying integration into finance applications.
Security, privacy, and compliance are crucial for generative AI in finance. Workloads are designed to protect sensitive customer data and comply with financial regulations, often incorporating features like access controls and activity monitoring.
Generative AI workloads require a strong data platform to store, process, and feed large datasets for model training and inference, supporting tasks like prompt engineering or fine-tuning for finance-specific scenarios.

Example: A bank uses a generative AI-powered chatbot to automatically answer customer questions about credit card features, recent transactions, or loan options using natural, conversational language.

Use Case: An investment firm deploys a generative AI solution to automatically generate personalized investment portfolio summaries for clients every month. These summaries include plain-language explanations of market trends, portfolio performance, and suggested next steps, saving advisors time and giving clients clearer insights.

For more information see these links:

Identify guiding principles for responsible AI

Describe considerations for fairness in an AI solution

Use diverse and representative data: AI systems in finance should be trained with data that reflect a wide range of backgrounds, financial situations, and demographics to minimize bias and treat all users fairly.
Regular auditing for bias: Periodically check the AI models and data for any signs of bias or unfairness. This helps identify and address issues that might affect certain groups differently.
Transparency to users: Clearly communicate how AI decisions are made and why the AI solution was chosen. Let customers know that AI is used and explain the factors considered in financial decisions.
Human oversight and feedback: Include human review in decision processes, and encourage users to provide feedback if they notice unfair outcomes. This helps continually improve the AI system.
Continuous monitoring and improvement: Set up processes to regularly review and update the AI solution based on new data, user feedback, and changing ethical standards to ensure ongoing fairness.

Example: A bank uses an AI model to evaluate loan applications. To make the process fair, the bank trains the AI with data from various income levels, geographic locations, and age groups. This reduces the chance that the AI will favor one group of applicants over another.

Use Case: A financial advisor app powered by AI analyzes credit scores and spending habits to recommend savings plans. To ensure fairness, the app includes periodic audits, allows users to report concerns, and uses updated, diverse data for retraining. This helps the app provide unbiased advice to customers regardless of background.

For more information see these links:

Describe considerations for reliability and safety in an AI solution

Regularly test and monitor AI systems to ensure they produce consistent and accurate results, especially when handling sensitive financial data. Reliability means the system works as expected, even as conditions or input data change.
Implement strong security measures such as encryption, access controls, and secure data transmission to protect financial information from unauthorized access, breaches, or manipulation.
Follow industry and regulatory guidelines for financial services to make sure the AI solution operates safely and meets compliance standards. This helps prevent legal issues and builds trust with customers.
Establish clear troubleshooting and incident response processes so the organization can quickly address any issues or failures in the AI system, minimizing risk and disruption.
Engage diverse stakeholders in reviewing and assessing the AI solution to identify potential risks or limitations and implement improvements for safe use.

Example: A bank uses an AI-powered credit scoring system to determine loan approval. To ensure reliability and safety, the bank regularly checks the system with test cases, encrypts customer financial data, and monitors performance. If the system makes an unexpected error, a team reviews logs and takes corrective action quickly.

Use Case: A financial services firm deploying an AI chatbot for customer service uses secure channels, applies strict user authentication, and monitors chatbot interactions for errors or unusual behavior. The team conducts regular reliability tests and immediately responds to incorrect or unsafe suggestions, ensuring customer safety and trust.

For more information see these links:

Describe considerations for privacy and security in an AI solution

Protect financial data by restricting access to only authorized users or groups. Use role-based access controls (RBAC) and apply the principle of least privilege so sensitive information is not exposed unnecessarily.
Encrypt data both when it is stored (at rest) and when it is transferred (in transit) using strong encryption technologies. This helps prevent data breaches or unauthorized access, especially important for storing customer financial records.
Monitor and audit AI systems regularly by keeping detailed logs of who accesses data and AI models. This allows you to detect and respond to suspicious activity, and maintain compliance with financial data protection regulations.

Example: A bank uses an AI-driven chatbot to answer customer questions about their accounts. The bank ensures that all communication between the chatbot, customers, and internal systems is encrypted. Customers’ personal data is only accessible to employees with the right clearance level, and all data access is logged for auditing purposes.

Use Case: A small financial advisory firm deploys an AI solution to analyze clients’ investment portfolios. They configure Azure role-based access control so only licensed financial advisors can access sensitive customer data. All AI model files are stored in Azure Blob Storage with encryption and versioning enabled. The system maintains automated audit logs to monitor access and modifications, helping the firm comply with financial regulations.

For more information see these links:

Describe considerations for inclusiveness in an AI solution

Use diverse and representative data: Make sure the data used to train AI systems reflects the real range of customers, backgrounds, and situations found in finance – including different ages, genders, ethnic groups, and locations.
Test for potential biases: Regularly evaluate AI models for patterns where certain groups may be disadvantaged or excluded from financial services, and take steps to correct these issues if found.
Include voices from different backgrounds: Involve a range of stakeholders, including those from underrepresented communities, in designing and checking AI solutions to ensure that they meet different needs.
Communicate transparently: Clearly explain to users how and why AI solutions make decisions, and offer channels for users to provide feedback if something seems unfair or confusing.

Example: A bank is developing an AI-based loan approval system. To ensure inclusiveness, the team includes data from customers of different regions, languages, and income levels in its training data. They regularly check if the AI is unintentionally favoring one group over another and adjust the model if problems are found.

Use Case: A financial services company wants to launch a new AI tool to help customers build credit scores. By involving community representatives in the design process and using data from a wide range of populations, the company ensures that the tool provides fair recommendations for customers from various backgrounds, including immigrants and people with non-traditional employment histories, reducing the risk of exclusion.

For more information see these links:

Describe considerations for transparency in an AI solution

Clearly communicate to users when AI is involved in making decisions that affect them, such as when an automated system helps approve or deny a loan application.
Document how the AI solution works, explaining in simple language what data it uses, how it makes decisions, and any limitations or risks—this documentation should be accessible to both staff and customers.
Regularly review and update AI models while sharing updates or changes with stakeholders, ensuring they are aware of how the system is maintained and monitored for accuracy, fairness, and security.
Provide ways for users to ask questions or challenge decisions made by the AI, ensuring openness about how decisions can be reviewed or appealed.
Be transparent about privacy and security practices, clearly stating how customer data is used or protected in relation to the AI solution.

Example: A bank introduces an AI-powered credit scoring tool for loan approvals. The bank informs applicants when AI is used in the decision-making process and provides an explanation summary of which main factors influenced the outcome. If an applicant is denied a loan, the bank offers a simple explanation and contact details to request more information.

Use Case: A finance professional uses a dashboard powered by AI to detect unusual transactions. To be transparent, the dashboard displays how alerts are generated (e.g., pattern analysis, past behavior) and offers links to plain-language guides on how the AI flags transactions. This helps the user confidently spot errors or ask the compliance team for clarification.

For more information see these links:

Describe considerations for accountability in an AI solution

Clearly assign roles and responsibilities for AI oversight. Designate specific individuals or teams who are accountable for monitoring the AI system’s performance, making decisions, and ensuring alignment with policies and regulations. This helps prevent ambiguity and ensures someone is always responsible for AI-related issues.
Establish transparent processes for AI decision-making. Document how AI models make decisions and communicate this information to relevant stakeholders and users. Transparency builds trust and allows for quicker identification and correction of errors or unfair outcomes.
Regularly review and update the AI system for compliance with financial regulations. The finance industry faces strict rules around data usage, privacy, and fairness. Consistently auditing the AI solution helps ensure it remains compliant with changing laws and standards.
Implement feedback and escalation procedures. Set up channels for users and employees to report issues or concerns, and ensure there are procedures for quickly escalating and addressing problems or unintended consequences caused by AI.

Example: A bank launches an AI system to automatically approve or deny loan applications. An internal team is assigned to review loan approval decisions, monitor regulatory requirements (like anti-discrimination laws), and update the system if any bias or errors are found. They also document decisions so customer service staff can explain them to applicants, maintaining transparency.

Use Case: A junior data analyst in a bank uses a dashboard to track the performance of an AI-powered credit scoring system. The dashboard showcases which team is accountable if an issue arises, provides an explanation for each credit decision, and allows users to submit feedback or complaints. Periodic audits are triggered automatically to ensure compliance with regulations. The analyst knows whom to contact when a problem is detected and can easily report concerns using built-in features.

For more information see these links:

Describe fundamental principles of machine learning on Azure (15-20%)

Identify common machine learning techniques

Identify regression machine learning scenarios

Regression is a type of machine learning task used to predict a continuous number, such as price, profit, or sales figures, rather than a category or label.
In financial scenarios, regression models analyze relationships between different factors (features) and a numerical outcome, like how interest rates or economic indicators might influence stock prices.
To build a regression model, you use historical data where the actual values are known, train the model to learn patterns, and then use it to estimate values for new or unseen situations.
Regression is suitable for scenarios such as forecasting future stock prices, predicting the value of transactions, or estimating expenses based on market trends — whenever the output is a number that can vary widely.
Actionable insight: Start by clearly identifying your numerical prediction goal (e.g., next quarter’s sales) and gather related data points, as the choice of features strongly affects the accuracy of regression models.

Example: Imagine you want to predict the future price of a company’s stock. You can use regression to model how factors such as previous stock prices, overall market performance, and economic reports affect the price. For instance, if historical data shows stock prices increasing with positive earnings announcements, the regression model can learn this relationship and help estimate future prices.

Use Case: A financial analyst uses a regression model to estimate the expected returns for a portfolio. By inputting past performance data, current interest rates, and economic trends into the model, they can predict how the portfolio’s total value might change over the next year. This helps inform investment decisions and strategy.

For more information see these links:

Identify classification machine learning scenarios

Classification is a type of machine learning task where data is categorized into groups, such as ‘fraudulent’ or ‘legitimate’ transactions. It’s widely used in finance to make quick, automated decisions based on existing labeled data.
There are two main types of classification: binary classification (choosing between two categories) and multiclass classification (choosing among three or more categories). For example, binary classification helps decide if a transaction is fraud or not, while multiclass classification could label customers as ‘low,’ ‘medium,’ or ‘high risk.’
To build a classification model, you need labeled training data (examples where you already know the category), such as historical transactions marked as fraudulent or legitimate. The trained model then predicts the category for new, unseen data, helping with real-time decision-making.
Common scenarios in finance for classification include fraud detection, loan approval, customer segmentation, and identifying suspicious activities. These uses help businesses reduce losses and improve customer service.
Actionably, beginners can start by gathering clean, labeled data, choosing a simple classification algorithm (like logistic regression or decision trees), and training a model to predict outcomes relevant to their business, such as whether a customer will default on a loan.

Example: A bank uses classification to automatically flag credit card transactions as ‘fraudulent’ or ‘legitimate.’ The system looks at past examples and learns patterns, so it can alert the bank to risky transactions in real time.

Use Case: A beginner working in finance collects past data on loan applications, including whether each was repaid or defaulted. By using a binary classification model, they can predict if a new applicant is likely to default, helping the bank make safer lending decisions and manage risk more effectively.

For more information see these links:

Identify clustering machine learning scenarios

Clustering is a machine learning technique that groups data points based on similar characteristics, without using labeled data. This is known as unsupervised learning.
In finance, clustering helps discover natural segments in customer data, such as grouping clients by spending habits, risk profile, or investment behavior.
Clustering algorithms (like K-Means) work by analyzing data and assigning each data point to a cluster, helping identify patterns and relationships that are not obvious through simple observation.
By understanding clusters, financial institutions can tailor products, personalize communication, and improve customer service for each segment.
Clustering can also be used to detect unusual patterns, which may help in identifying potential fraud or outlier behaviors.

Example: A bank analyzes transaction data from thousands of customers and uses clustering to group clients into segments—such as frequent travelers, high-volume business owners, and everyday users. Each group has distinct financial needs.

Use Case: A financial advisor uses clustering to segment retail clients based on investment behavior. This allows the advisor to offer customized financial products and targeted educational content to each group, increasing client engagement and satisfaction.

For more information see these links:

Identify features of deep learning techniques

Deep learning uses artificial neural networks with multiple layers (‘deep’), which allows models to automatically discover complex patterns in large amounts of data.
Deep learning models can learn to create and extract features from raw, unstructured data (like text, images, or audio) without needing humans to manually design these features.
Training deep learning models requires powerful hardware (often GPUs) and substantial data, but they excel at handling tasks where the data is complicated or voluminous, such as image recognition or language processing.

Example: A bank wanting to automatically detect fraudulent transactions from millions of records can use a deep learning model. Unlike older systems that rely on fixed rules, the model learns distinct patterns of fraud by itself from the data.

Use Case: Credit card companies use deep learning to monitor transaction data in real time, flagging unusual patterns that might indicate fraud. This helps protect customers and reduces financial losses, with minimal manual intervention required.

For more information see these links:

Identify features of the Transformer architecture

Transformers use an attention mechanism, which allows the model to focus on the most relevant parts of a sequence when making predictions. For example, when analyzing a sentence, attention helps the model determine which words are most important for understanding the meaning.
The architecture has two key parts: the encoder and the decoder. The encoder processes input data (like a financial news article) and turns it into a meaningful numeric representation. The decoder then uses this to generate output, such as a summary or prediction.
Transformers handle sequences (like text or time-series financial data) efficiently. Unlike older models (like RNNs), transformers can consider all parts of the sequence at once, making them well-suited for processing long documents or tracking market prices over time.
Popular transformer-based models include BERT and GPT, which are used for tasks such as text classification, summarization, and question answering.
With transformers, organizations can use pre-trained models and adapt them to specific finance tasks, making AI adoption faster and more accessible even with limited data.

Example: A financial analyst wants to quickly summarize daily market news. A transformer can read large news articles and generate concise summaries, helping the analyst stay informed about market trends without reading every article in full.

Use Case: A bank uses a transformer-based model deployed on Azure Machine Learning to automatically scan and summarize earnings reports, flagging any critical changes or risks. This actionable information helps analysts respond faster to important financial developments.

For more information see these links:

Describe core machine learning concepts

Identify features and labels in a dataset for machine learning

Features (also known as input variables) are the individual pieces of information used by a machine learning model to make predictions. In finance, features might include client income, age, account balance, transaction history, or credit score.
Labels (also known as target variables or outcomes) are the values that the model is trying to predict or classify. For supervised learning, each data point has a known label, such as whether a loan was repaid (yes/no) or the dollar amount of a stock price movement.
Properly identifying and separating features from labels is crucial for training accurate machine learning models. Features provide the model with context, while the label represents the answer or result the model should learn to predict.
Features and labels need to be gathered, formatted, and often labeled manually or via tools (such as Azure Machine Learning Data Labeling) before they can be used to train a model. Ensuring the accuracy and consistency of labels is essential for effective model training.
In many financial applications, using historical labeled data (where the outcomes are known) enables organizations to train predictive models to automate and improve business decisions, like loan approval or fraud detection.

Example: Suppose a bank wants to predict which loan applicants are likely to repay their loans. The dataset includes features such as applicant income, employment status, loan amount, and years of credit history. The label is whether the applicant ended up repaying the loan (‘repaid’ or ‘defaulted’). The model uses the features to learn patterns that help it predict the label for new applicants.

Use Case: A financial analyst uses Azure Machine Learning to create a labeled dataset for a credit card fraud detection project. Transaction data (features: transaction amount, location, time, account age) is labeled as ‘fraudulent’ or ‘not fraudulent’ (label). By clearly identifying which columns are features and which is the label, the analyst can train a machine learning model to detect likely fraud cases in new transactions.

For more information see these links:

Describe how training and validation datasets are used in machine learning

In machine learning, training and validation datasets are used to develop and evaluate a model. The training dataset is the portion of your data used to teach the model to understand patterns and relationships. The model learns by adjusting its parameters to reduce prediction errors on this data.
The validation dataset is a separate set of data not seen by the model during training. It is used to tune the model’s settings, such as deciding how complex the model should be, and to monitor how well the model generalizes to new, unseen data. If performance on the validation data is poor, adjustments are made before finalizing the model.
Splitting your data into training and validation sets is crucial for preventing overfitting, where the model learns patterns that are too specific to the training data and fail to work for new data. Common strategies for splitting include random split, chronological split (based on time), or manual split, especially in financial datasets with time or class imbalances.
It’s important to only apply special data treatments (like SMOTE, which balances imbalanced classes by creating synthetic examples) to the training data, not to the validation or test sets. This ensures that the validation performance reflects how the model would work in real-world scenarios.

Example: Suppose a bank wants to predict which customers are likely to leave (churn). The bank collects data on customer transactions and demographics. They randomly split 10,000 customer records into 6,000 for training and 2,000 for validation, and keep the remaining 2,000 as a test set. The model learns using the training data, and its performance is checked on the validation data. If the model does well on the validation set but poorly on the test set, the bank knows it has probably overfitted and should re-examine its approach.

Use Case: A financial institution is building a model to detect fraudulent transactions. They split historical transaction data into training and validation datasets. The model is trained to recognize suspicious behavior using the training data, while its settings are optimized based on the validation data. By keeping the validation data separate, the institution ensures the model is accurate when applied to new transactions, helping to minimize false alarms and missed fraud cases.

For more information see these links:

Describe Azure Machine Learning capabilities

Describe capabilities of automated machine learning

Automated machine learning (AutoML) streamlines the process of building ML models by automatically selecting the best algorithms, tuning settings, and preparing data with minimal manual intervention. It reduces the complexity and time required, making it easier for people without a data science background to use machine learning.
AutoML in platforms like Azure and Microsoft Fabric integrates with existing data sources and tools, allowing users to quickly train and deploy models directly from their financial data. This integration supports large datasets and helps users get fast, accurate results.
AutoML handles repetitive tasks like preprocessing, model training, and evaluation. By automatically experimenting with different algorithms and configurations, it finds the optimal model for a given problem, improving efficiency and helping users focus on interpreting results and making decisions.

Example: A bank wants to predict which loan applicants are most likely to default. With AutoML, a financial analyst uploads historical loan data to Azure Machine Learning and lets AutoML try various algorithms and settings to build the best predictive model. The system evaluates and recommends the most accurate solution, even for users new to AI.

Use Case: A credit analyst uses AutoML in Azure Machine Learning to quickly create a model that classifies customers based on risk of default, enabling faster and more consistent loan approval decisions without needing in-depth machine learning expertise.

For more information see these links:

Describe data and compute services for data science and machine learning

Azure Machine Learning provides integrated data services that let you securely store, access, and manage your financial data in the cloud. With Azure Datastores and Datasets, you can easily connect to data sources like Azure SQL Database, Azure Data Lake, or even simple files in Blob storage, which makes your data ready for machine learning projects.
To perform data science and machine learning tasks, Azure offers compute services such as Azure Machine Learning compute instances, Azure Synapse Spark pools, and Azure Databricks. These services provide scalable and managed environments for handling large data processing, model training, and testing.
With Azure Machine Learning, you can automate many steps, such as data preparation, model training, and experiment tracking. Automation and management features help ensure that your data science workflows are repeatable and secure—essential for meeting governance and compliance needs in the finance industry.
Azure’s integration with other data and analytics services like Power BI enables you to visualize results, while options like autoscaling and pay-as-you-go compute save costs by only using resources when needed.
These data and compute services also support collaboration by allowing teams to share workspaces, access registered datasets, and reuse trained models, making it easier for financial institutions to work together on large-scale AI and analytics projects.

Example: A retail bank wants to analyze customer transaction data to detect unusual or potentially fraudulent activity. They upload their transaction records into Azure Blob Storage, register this data in Azure Machine Learning as a dataset, and use an Azure ML compute instance to train a fraud detection model. The team can easily scale up resources for processing large volumes of data and securely manage access, ensuring sensitive financial data stays protected.

Use Case: A financial analyst at an investment firm uses Azure Machine Learning and Azure Synapse Spark pool to analyze historical stock price data and make predictions about future movements. By using Azure’s data and compute services, the analyst can efficiently process massive datasets, build predictive models faster, collaborate with colleagues, and create dashboards in Power BI to share insights with decision makers—all while meeting industry security and compliance requirements.

For more information see these links:

Describe model management and deployment capabilities in Azure Machine Learning

Model registration and versioning in Azure Machine Learning lets you securely store, organize, and keep track of machine learning models. Each model registered in your workspace receives a unique name and version, making it easy to manage model updates and audit changes over time.
Azure Machine Learning streamlines deployment by allowing you to package models with all necessary dependencies and scripts, and then easily deploy them as endpoints for real-time or batch predictions. You can test deployments locally before moving to the cloud, helping prevent errors and ensuring reliability.
With Azure Machine Learning, you can monitor deployed models for performance, data drift, and operational issues, receiving alerts if something changes. The platform automatically logs metadata about how and when models are trained, deployed, or updated—helping you meet finance industry requirements for transparency and regulatory compliance.

Example: A bank uses Azure Machine Learning to register a credit scoring model. Each new version of the model, trained with updated customer data, is tracked automatically. The bank can safely deploy the latest model to production as an endpoint, making instant credit decisions for loan applicants via the bank’s website.

Use Case: A financial analyst with no AI background can use Azure Machine Learning to manage and deploy a fraud detection model. The analyst can register different versions of the model, safely test them using controlled rollout features, and monitor performance. If the model catches more fraudulent transactions, the analyst can promote it to serve more of the business, all while keeping a clear record for compliance audits.

For more information see these links:

Describe features of computer vision workloads on Azure (15–20%)

Identify common types of computer vision solution

Identify features of image classification solutions

Image classification solutions use machine learning models to automatically sort images into different predefined categories based on visual features, such as shapes, colors, and textures.
Pretrained models like Inception or Custom Vision allow you to quickly start classifying images by reusing knowledge gained from millions of prior images, even with limited new training data.
These solutions require you to provide labeled images—each image must be tagged with the correct category—so the model learns to recognize patterns distinguishing each class.
Modern tools, such as ML.NET Model Builder and Azure Custom Vision, make it easy for beginners to train and deploy image classification models without needing deep expertise in AI or programming.
Once trained and tested, image classification models can be integrated into business applications to automate tasks such as visual verification, fraud detection, or quality control.

Example: A bank uses an image classification model to automatically review images of submitted ID documents (like passports or driver’s licenses) and classify them as valid, expired, or potentially fraudulent, reducing the need for manual checking by staff.

Use Case: Automated check processing: A financial institution deploys an image classification solution to automatically sort scanned checks into categories (e.g., personal, business, government, or suspicious) before further processing, saving time and minimizing the risk of processing errors or fraud.

For more information see these links:

Identify features of object detection solutions

Object detection solutions identify specific objects within images and provide their location using bounding boxes, which are rectangles that show where each object is found in the image.
These solutions not only recognize what objects are present (like people, vehicles, documents, or money) but also count how many instances of each object appear in a single image.
Object detection solutions can help automate tasks by sorting, organizing, or flagging items in images, which saves time and reduces errors compared to manual review.
Most object detection tools use artificial intelligence models trained on many examples to accurately distinguish between different object types, even in crowded or complex images.
Some limitations exist, including difficulty detecting small or overlapping objects and not distinguishing between brands or product names unless paired with additional features.

Example: A bank uses object detection on security camera footage to automatically identify and count the number of people present at each teller booth during business hours.

Use Case: In finance, object detection can be used to automate the process of verifying document authenticity by detecting the presence and correct positioning of required items (like signatures, stamps, or IDs) on scanned checks or application forms, helping staff process documents faster and reducing the risk of missing important details.

For more information see these links:

Identify features of optical character recognition solutions

OCR solutions extract text from images or scanned documents, including both printed and handwritten content. In finance, this means you can quickly turn paper-based records (such as receipts and invoices) into readable, searchable digital text.
Modern OCR systems provide accuracy indicators for extracted text, highlighting words and phrases they recognize confidently. This allows users to review or automate processes based on reliability thresholds—for example, sending low-confidence results for human review.
OCR supports multiple languages and document types, making it useful for global finance operations that handle invoices and statements in different languages and formats. Deployment options include cloud, API, and on-premises containers, enabling flexible integration with existing finance systems.
OCR enables automation of manual data entry tasks. By extracting text lines, words, and even the location in a document, financial teams can streamline workflows such as expense management or invoice processing.
Feedback and training features help OCR systems improve over time. When errors occur, corrections made by users can be sent back to the OCR service, which then learns and adapts for future document processing—reducing repeated mistakes.

Example: A financial analyst receives a batch of scanned vendor invoices in PDF format. Using an OCR solution, these PDFs are quickly converted into electronic records. The analyst can now search, sort, and process these invoices digitally, saving hours compared to manual retyping.

Use Case: An accounts payable department in a finance company uses OCR to automatically extract relevant fields (vendor name, amounts, invoice dates) from incoming invoices emailed as PDFs or images. The extracted data populates their finance software, and only invoices below a confidence threshold are flagged for manual review, reducing errors and speeding up processing.

For more information see these links:

Identify features of facial detection and facial analysis solutions

Face Detection: The ability of AI systems to locate human faces in images or video frames, returning coordinates for each detected face. This is the foundational step for any further facial analysis or recognition tasks.
Face Analysis: Once a face is detected, facial analysis can estimate attributes like age range, head pose, and image quality. This step helps evaluate whether a face is suitable for identity verification. Some solutions also support quality analysis to ensure the detected face is clear enough for accurate processing.
Face Recognition and Search: Facial solutions can match a detected face to individuals already enrolled in a directory, enabling person identification or verification. Features like one-to-one (verifying if a live face matches an ID photo) and one-to-many (searching for a match among many enrolled faces) are key capabilities. Systems may also provide a person directory to organize, enroll, and easily manage multiple face records securely.

Example: A bank uses an AI-based face detection and recognition system to simplify customer sign-ins on its mobile app. When a customer wants to log in, they take a selfie. The app detects the face, analyzes the image quality, and then compares the selfie to the customer’s photo on file to verify their identity, allowing quick and secure access.

Use Case: In finance, a common use case is fraud prevention during remote onboarding: When a new customer opens an account online, the system detects and analyzes their face from a selfie, checks image quality, and verifies the face against a government-issued ID photo. This ensures the customer is who they claim to be, reducing identity fraud risks while speeding up the account opening process.

For more information see these links:

Identify Azure tools and services for computer vision tasks

Describe capabilities of the Azure AI Vision service

Azure AI Vision can extract text from images using Optical Character Recognition (OCR). This means it can read and digitize information from documents, receipts, invoices, and other printed or handwritten materials, making it easier to process financial data automatically.
The service analyzes images to detect and describe objects, faces, people, and other visual elements. For example, it can identify logos, count people in a photo, or generate automatic captions and tags, which helps with organizing and searching through large collections of images.
Azure AI Vision can moderate images by detecting potentially sensitive or inappropriate content, ensuring that all visual data stored and shared by a financial institution meets compliance and content guidelines.
The platform supports secure and responsible handling of customer data, which is crucial for the finance industry. Microsoft provides detailed privacy and security policies to protect sensitive information analyzed via Azure AI Vision.
Easy integration through prebuilt APIs and tools like Vision Studio allows finance professionals who are new to AI to quickly get started analyzing images or processing documents without deep programming knowledge.

Example: A bank receives thousands of scanned checks and invoices daily. By using Azure AI Vision’s OCR capability, the bank can quickly extract important details such as amounts, dates, and payee names directly from image files, streamlining data entry and reducing manual errors.

Use Case: A financial services firm needs to organize and search through a large digital archive of scanned documents and images. By implementing Azure AI Vision, they automatically extract text from these images and tag key information (such as client names or account numbers), making it easy for staff to locate documents and ensure compliance with record-keeping regulations.

For more information see these links:

Describe capabilities of the Azure AI Face detection service

The Azure AI Face detection service uses advanced AI algorithms to locate human faces in images. It can return details like the position of each face, which is helpful for many computer vision applications.
In addition to detecting faces, the service can analyze facial features and predict attributes such as head pose, glasses, or facial hair. These predictions help in ensuring image quality before using photos for identification purposes.
Face detection is the foundation for more advanced tasks like facial recognition (identifying or verifying individuals) and access control. It must find and analyze faces before any further action, making it a crucial first step in building secure, user-friendly applications.

Example: A financial company uses Azure AI Face detection to verify the identity of users during online account opening. When a user uploads their photo and a picture of their ID, the system detects and compares the faces to help confirm the person opening the account is the same as the ID holder.

Use Case: Banks and financial institutions can use face detection to enable touchless access control for secure areas such as vaults or data centers. Employees scan their face at a camera, and the system quickly detects and verifies their identity, improving security without relying on physical keys or cards.

For more information see these links:

Describe features of Natural Language Processing (NLP) workloads on Azure (15–20%)

Identify features of common NLP Workload Scenarios

Identify features and uses for key phrase extraction

Key phrase extraction is a natural language processing technique that automatically identifies the most important words or short phrases from large amounts of unstructured text, such as customer comments, emails, or reports.
Features include fast processing of large datasets, support for multiple languages, and integration with tools like Power BI, allowing users to extract actionable insights without manual reading.
In finance, key phrase extraction helps organize and summarize customer feedback, highlight frequently discussed products or concerns, and quickly surface important topics for decision-makers.

Example: A bank collects thousands of customer reviews each month. By using key phrase extraction in Power BI, analysts can quickly see terms like ‘mobile app’, ‘account fees’, or ‘customer service’ appear most often, helping them understand which issues are discussed most and require attention.

Use Case: A new-to-AI financial analyst uses Power BI’s key phrase extraction feature to analyze customer feedback. They easily visualize common concerns using a word cloud, allowing them to identify and address recurring topics such as ‘late transactions’ or ‘interest rates’, improving customer satisfaction.

For more information see these links:

Identify features and uses for entity recognition

Named Entity Recognition (NER) is a Natural Language Processing technique used to automatically find and classify key entities—such as people, organizations, locations, dates, and amounts—in unstructured text data.
NER works by analyzing text and identifying words or phrases that match predefined categories (like financial instruments, companies, or transaction dates). This helps turn unstructured information into structured, searchable data.
NER is widely used in finance to improve efficiency and accuracy in tasks like document analysis, news monitoring, regulatory compliance, and data extraction from reports. Azure AI Language’s NER feature supports multiple languages, making it suitable for global financial operations.

Example: A financial analyst receives a news article containing information about company mergers, stock price changes, and key executives. Using NER, the analyst’s application automatically highlights and extracts names of the companies, executives involved, and relevant dates, providing a quick summary without manual searching.

Use Case: A bank uses NER to process thousands of incoming emails and contracts daily. The system automatically identifies and extracts customer names, account numbers, transaction dates, and monetary amounts. This structured data is then used to monitor transactions for compliance, detect suspicious activities, and improve customer service by responding more quickly to queries.

For more information see these links:

Identify features and uses for sentiment analysis

Sentiment analysis is a Natural Language Processing (NLP) technique that helps identify whether a piece of text expresses a positive, negative, or neutral feeling. This is useful for understanding public opinion or customer feedback at scale.
Opinion mining, often called aspect-based sentiment analysis, is a feature of sentiment analysis. It goes further by detecting opinions about specific parts or aspects of a product or service, such as ‘customer service’ or ‘fees.’
In finance, sentiment analysis is commonly used to process social media posts, news articles, and customer reviews to quickly gather insights about market mood or client emotions regarding financial products.
The results of sentiment analysis typically include a label (positive/neutral/negative) and a confidence score, which indicates how certain the system is about its assessment. This helps users gauge how reliable the sentiment classification is.
Finance professionals can use automated sentiment analysis tools (such as those in Azure AI Language services) via APIs. With minimal technical setup, they can process large amounts of feedback or news, saving time compared to manual review.

Example: A bank collects thousands of online reviews from customers every month. By using sentiment analysis, the bank can quickly identify if overall customer satisfaction is rising or falling and spot issues that may need urgent attention (for example, a sudden increase in negative reviews mentioning slow transaction processing).

Use Case: A financial services company uses sentiment analysis to monitor social media and online forums for mentions of their mutual funds. If there is a sudden surge in negative sentiment tied to specific fund features, the company can proactively investigate, communicate with investors, or adjust their offerings to address common concerns.

For more information see these links:

Identify features and uses for language modeling

Language models use machine learning to understand and generate human language. They work by predicting what word or phrase comes next in a sentence, based on the patterns they learn from large amounts of text.
There are small and large language models. Small models are quicker and less expensive, but large models can handle more complex tasks and provide more accurate results because they have been trained on more data.
In finance, language models help process, categorize, and analyze large volumes of documents, such as financial reports, customer communications, or news articles, enabling faster decision-making and reducing manual work.
Language models can be used for tasks like extracting key information (such as names, dates, and amounts), classifying documents (for example, as sensitive or not), and understanding customer sentiment in emails or social media posts.

Example: A financial services company uses a large language model to automatically read and summarize news articles about stocks and market events. This helps analysts quickly understand important developments without reading every article themselves.

Use Case: A bank uses a pre-trained language model to screen incoming customer emails. The model classifies each message as a complaint, a query, or a new application request, and then routes each email to the appropriate team. This reduces response times and improves customer service.

For more information see these links:

Identify features and uses for speech recognition and synthesis

Speech recognition allows financial professionals to convert spoken words into text, making it easier to fill out forms, dictate emails, or trigger commands hands-free.
Speech synthesis (text-to-speech or TTS) enables applications to read out financial summaries, alerts, or transaction details to users, improving accessibility and helping those with visual impairments.
Modern speech technologies support multiple languages and customizable voices, allowing financial applications to serve a diverse client base and enhance user experience.
Speech recognition can identify the intent behind spoken commands, such as checking account balances or initiating transfers, streamlining common banking tasks.
Text-to-speech features can be tailored using Speech Synthesis Markup Language (SSML), which adjusts pronunciation, pitch, and emphasis for clear communication—important when reading out figures or account details.

Example: A banking app uses speech recognition to let users dictate payment instructions, such as ‘Pay $100 to Electric Company,’ and then uses text-to-speech to confirm the payee and amount before processing the transaction.

Use Case: A customer service chatbot for a financial institution uses speech recognition to transcribe customer questions about loan options, and text-to-speech to deliver personalized, clear responses, making support accessible even for customers with limited reading skills or disabilities.

For more information see these links:

Identify features and uses for translation

Translation in Natural Language Processing (NLP) automatically converts written or spoken content from one language to another, helping organizations communicate with customers and colleagues across different languages.
Finance teams can use translation tools to quickly convert documents such as reports, contracts, compliance notices, and customer communications, making it easier to work with multinational clients or branches.
Modern translation features like those in OneDrive support many formats (like DOCX, PDF, PPTX) and can translate a document into multiple languages at once, while preserving the original formatting and structure.
Translation tools often offer enterprise-level security, ensuring financial data and sensitive information remain protected during and after translation.
Automated translation reduces delays, minimizes manual effort, and helps financial professionals avoid costly mistakes caused by language misunderstandings.

Example: A finance analyst working for a global bank receives a quarterly financial report in Spanish from the company’s South American branch. Using OneDrive’s document translation feature, the analyst translates the report into English with a few clicks, keeping the original tables and formatting intact.

Use Case: A compliance officer in a multinational finance firm needs to distribute updated regulatory policies to branches in several countries. They upload the policy document to OneDrive, select up to 10 target languages, and instantly generate translated copies, ensuring all teams receive accurate information in their local languages.

For more information see these links:

Identify Azure tools and services for NLP workloads

Describe capabilities of the Azure AI Language service

Sentiment analysis: Azure AI Language can automatically determine whether customer feedback or financial news is positive, negative, or neutral. This helps financial organizations quickly gauge public opinion or customer satisfaction.
Key phrase extraction: The service can identify the most important terms and topics from large volumes of financial documents, emails, or customer chats. This makes it easier to summarize and organize key information.
Named entity recognition (NER): Azure AI Language can pick out specific entities such as company names, amounts, dates, and financial products within text. This is useful for extracting structured data from unstructured sources.
Language detection: The service can automatically identify which language a document or message is written in. This enables financial teams to route inquiries or documents to the right language experts or workflows.
Text summarization: Azure AI Language offers both extractive and abstractive summarization. This allows financial professionals to quickly get the gist of long reports or news articles without reading everything in detail.

Example: A bank receives hundreds of customer emails every day. Using Azure AI Language, the bank can automatically detect the language of each email, extract key phrases and named entities (like account numbers or product names), and analyze the sentiment to see if the customer is happy or upset. This allows staff to prioritize and route emails more efficiently.

Use Case: A financial advisor wants to track important news about client companies. By using Azure AI Language’s named entity recognition and sentiment analysis, the advisor can set up an automated system to scan financial news feeds, extract mentions of client companies, and flag articles that either have a negative sentiment or discuss significant events. This helps the advisor keep clients informed and respond proactively.

For more information see these links:

Describe capabilities of the Azure AI Speech service

Speech-to-text: Azure AI Speech can accurately transcribe spoken words from audio or phone calls into written text. This helps automate tasks like creating meeting notes, call transcripts, or compliance records.
Text-to-speech: The service can convert written content—such as reports or policy updates—into clear, natural-sounding speech in multiple languages and voices. This is useful for creating accessible content or automated alerts.
Speech translation: Azure AI Speech allows for real-time translation of spoken language, enabling communication with customers or colleagues who speak different languages, and making it easier to serve a global workforce.
Speaker recognition: The service can identify or verify speakers based on voice characteristics, helping to enhance security (for example, voice-based authentication for customer service hotlines).
Batch/audio file processing: Azure AI Speech can process and transcribe large volumes of recorded calls or interviews efficiently, which is especially helpful for audit trails or analyzing customer interactions in the finance sector.

Example: A bank wants to keep accurate records of customer support calls. By using Azure AI Speech, they can automatically transcribe every recorded phone conversation into searchable text, making it easy to review, archive, or analyze calls for compliance and quality assurance.

Use Case: A financial customer service center uses Azure AI Speech for real-time transcription of support calls. This allows agents to see live transcripts during a call, making it easier to record important information, follow compliance policies, and later review or audit conversations without listening to entire recordings.

For more information see these links:

Describe features of generative AI workloads on Azure (20–25%)

Identify features of generative AI solutions

Identify features of generative AI models

Generative AI models create new content (such as text, images, or audio) based on patterns learned from large amounts of existing data. This means they can respond to your prompts or questions with original outputs that are relevant to your input.
These models use advanced neural network architectures, like transformers, which allow them to understand, process, and generate human-like language. In finance, this enables natural conversations or the generation of readable reports and summaries from raw data.
Outputs from generative AI are unique each time, even with similar inputs, because the model combines learned patterns in dynamic ways. This makes them useful for tasks like drafting emails, summarizing documents, or generating personalized responses for customer service.
Generative AI models require large and diverse data sets for training so they can accurately reflect patterns in the domain they are used in (such as finance). Fine-tuning can adapt a general model to financial language or regulatory requirements.
For finance, security, privacy, and data compliance are essential when using generative AI. It’s important to use models and platforms that ensure data is protected and regulatory standards are met.

Example: A financial advisor chatbot powered by a generative AI model can answer customer questions using natural language—like explaining investment options or clarifying recent account transactions—by generating clear, tailored responses based on the customer’s query.

Use Case: A bank integrates a generative AI solution to automatically draft personalized quarterly portfolio summaries for clients. The AI reviews transaction data, identifies key activities, and generates a readable summary that advisors can send directly to clients after a quick review.

For more information see these links:

Identify common scenarios for generative AI

1. Automating Customer Communication: Generative AI can power chatbots and virtual assistants to handle routine customer questions, such as account inquiries or transaction details, allowing human agents to focus on complex issues.
1. Personalized Financial Recommendations: Generative AI can analyze transaction history and spending patterns to generate tailored budgeting tips, investment suggestions, or alerts to help customers make smarter financial decisions.
1. Intelligent Document Summarization: Generative AI can quickly summarize lengthy financial reports, statements, or regulatory documents, helping professionals and customers extract key information without reading every detail.
1. Fraud Detection Support: By generating descriptions and explanations for suspicious activity or unusual transactions, generative AI can help investigators or customers better understand potential fraud cases.
1. Enhanced Semantic Search: Generative AI enables users to search financial data using natural language, returning results based on meaning rather than exact keyword matches, making it easier to find relevant information.

Example: A bank uses generative AI to power an online chatbot that answers customer questions about their account balances, transaction history, and helps them quickly report a lost credit card—all in natural language, 24/7, without waiting for human assistance.

Use Case: A financial advisor uses a generative AI tool to automatically summarize long quarterly market reports into short, actionable insights for clients, saving time and improving client communication.

For more information see these links:

Identify responsible AI considerations for generative AI

Fairness and Bias Prevention: Generative AI should be trained on diverse and representative financial data to avoid unintentional discrimination. Regular checks and updates are needed to ensure the AI treats all users, regardless of background, fairly.
Transparency and Explainability: It is important to clearly communicate when and how generative AI is being used in financial services. Users should understand what the AI is doing, how decisions are made, and why certain outputs are generated.
Accountability and Compliance: Assign clear responsibility for monitoring AI outputs and ensuring the system meets industry and regulatory standards. Regularly audit the AI for accuracy, privacy, and ethical compliance to protect the organization and its customers.

Example: A financial advisor chatbot uses generative AI to suggest investment options for clients. If the training data is biased towards certain demographics, the AI might unintentionally favor one group over others. By regularly reviewing outputs and updating data, the company ensures all clients receive fair and balanced financial advice.

Use Case: A bank deploys a generative AI tool to help draft personalized loan recommendation letters for customers. The bank’s compliance team reviews and approves the AI-generated content, applies content filters to prevent sensitive data leakage, and documents all AI decisions. This workflow ensures the tool operates transparently, fairly, and in line with financial regulations.

For more information see these links:

Identify generative AI services and capabilities in Microsoft Azure

Describe features and capabilities of Azure AI Foundry

Unified Platform: Azure AI Foundry provides a single workspace for building, testing, deploying, and managing generative AI applications. This makes it easier for teams to work together and reduces the need to switch between multiple tools.
Collaboration and Security: The platform offers shared workspaces, version control, and enterprise-grade security features such as role-based access control. This allows financial organizations to work safely and efficiently, managing sensitive data and compliant workflows.
Responsible AI Tools: Azure AI Foundry includes built-in capabilities for bias detection, model interpretability, and privacy-preserving machine learning. These tools help financial institutions ensure their AI solutions are fair, transparent, and meet industry regulations.
Integration and Scalability: The platform seamlessly connects with other Azure services, open-source frameworks, and the Microsoft Fabric data ecosystem. This allows financial companies to develop solutions using familiar development tools and scale them to production as their needs grow.
Easy Model Management: Users can access, compare, and deploy over 1,600 AI models using serverless APIs and hosted fine-tuning options. This simplifies experimentation with generative AI and helps organizations make informed choices based on their requirements.

Example: A bank wants to automate customer support using an AI-powered chatbot. With Azure AI Foundry, the team can build the chatbot using prebuilt natural language models, ensure its responses are unbiased, and deploy it securely, all from a single platform.

Use Case: A financial advisory firm uses Azure AI Foundry to quickly develop a generative AI system that helps analyze client financial documents, summarize key points, and recommend investment strategies. The platform’s responsible AI tools enable them to ensure recommendations are free from bias and comply with industry regulations, while collaboration features let multiple advisors contribute and review outcomes together.

For more information see these links:

Describe features and capabilities of Azure OpenAI service

Access to Powerful AI Models: Azure OpenAI Service lets users access advanced generative AI models such as GPT-4, GPT-3.5, Codex, DALL-E, Whisper, and GPT-4o. These models can generate human-like text, create images from descriptions, transcribe or translate audio, and process both text and spoken inputs.
Enterprise-Grade Security and Compliance: Unlike standard OpenAI APIs, Azure OpenAI is built on top of Microsoft Azure’s cloud infrastructure. This means financial organizations benefit from private networking, regional data residency, compliance with strict industry regulations, and built-in responsible AI safeguards, including content filtering.
Integration with Azure Ecosystem: Azure OpenAI Service can be linked easily with other Azure services like Azure Speech, Azure Data, and Power BI. This allows finance professionals to create multi-modal solutions (e.g., voice-enabled chatbots, document analysis tools) and automate workflows within the secure Azure environment.

Example: A bank uses Azure OpenAI to build a natural language chatbot accessible on their website. Customers can ask questions about loan products, review their recent transactions, or get personalized advice on savings, all in real-time using both text and voice. The chatbot leverages secure Azure infrastructure to protect sensitive customer data while generating accurate, context-aware responses.

Use Case: A financial advisor uses an Azure OpenAI-powered assistant to automatically generate client portfolio summary reports. By connecting Azure OpenAI to the company’s secure databases, the assistant can read financial performance data and produce tailored summaries, risk analyses, and investment recommendations quickly, helping advisors spend less time on routine documentation and more time on personalized client engagement.

For more information see these links:

Describe features and capabilities of Azure AI Foundry model catalog

The Azure AI Foundry model catalog is a centralized hub for discovering a wide range of AI models from leading providers like Microsoft, OpenAI, Meta, and more. This means you don’t need to search different platforms—everything is in one place.
You can easily filter and search for models based on specific criteria, such as industry (e.g., finance), capabilities (reasoning, tool calling), deployment options (serverless API, managed compute, batch), and supported tasks. This helps beginners select a model that best fits their needs without overwhelming choices.
The catalog provides clear performance benchmarks and comparison tools so you can review how well different models perform on real-world tasks. Model cards also offer quick facts, detailed descriptions, and licensing info, making it easy to understand model features before deciding to use them.
Deployment is flexible: models can be deployed as serverless APIs, provisioned for high-volume needs, or on managed compute for customized hosting. This lets you choose the best way to integrate AI models into your finance applications based on cost, speed, and scale.
Responsible AI, security, and integration are built-in. Models sold directly by Azure follow Microsoft’s standards for reliability, privacy, and scalable enterprise support—important for compliance in finance.

Example: A finance team wants to analyze customer spending patterns to offer personalized savings advice. They use the Azure AI Foundry catalog to filter for models specifically trained on financial data, compare performance, check licensing terms, and deploy the model using a serverless API. This allows the team to quickly integrate the model into their customer app without needing deep AI expertise.

Use Case: A beginner financial analyst uses the model catalog to find and deploy a pre-trained model for automated fraud detection. By filtering models by industry and inference task, the analyst selects a model already validated for finance. Through the Azure portal, the analyst deploys the serverless API within their company’s transaction system, reducing fraud risk without needing to build or train a custom model.

For more information see these links: