DEV Community: Lucy Joan

Unlocking the Full Potential of AI-Powered ERP Systems

Lucy Joan — Thu, 23 Jul 2026 11:42:59 +0000

The finance function has continually evolved alongside technological innovation, from manual bookkeeping to spreadsheets, and from standalone accounting software to Enterprise Resource Planning (ERP) systems.

This technological advancement has transformed how financial information is recorded, processed, analyzed, and reported. Today, the finance profession is undergoing another significant transformation as Artificial Intelligence (AI) becomes an integral component of modern ERP systems.

AI technologies such as machine learning (ML), natural language processing and predictive analytics are integrated into ERP systems. These AI-powered systems can automate routine tasks, provide advanced data analysis and forecasting, and enhance decision-making so as to improve operational efficiency and streamline business processes(Hayes & Downie, 2024).

What This Article Covers:

The Evolution of ERP Systems: From Automation to Intelligence
Prompt Engineering: The Language of AI-Powered ERP Systems
Practical Applications of Prompt Engineering Across the Finance Function
Best Practices for Prompt Engineering in Finance
When Good AI Produces Poor Results: Common Prompting Mistakes in Finance
Governance, Ethics, and Human Oversight: Building Trust in AI-Powered Finance
The Future of Prompt Engineering in Finance
Industry Perspectives: Why Prompt Engineering Matters
Conclusion

The Evolution of ERP Systems: From Automation to Intelligence

Every month, finance teams spend considerable time preparing management reports, reconciling accounts, investigating variances, monitoring budgets, and providing decision support to management. Many of these activities can now be accelerated by AI embedded within ERP systems.

However, despite these capabilities, many organizations have yet to realize the full value of AI-powered ERP solutions. The challenge is no longer access to AI; rather, it is the ability to use AI effectively.

ERP systems have been at the heart of organizational operations for more than three decades. Originally they were developed to integrate business processes across departments such as finance, procurement, inventory, manufacturing, human resources, and sales.

Their primary role was to standardize business processes, improve data accuracy, and automate routine transactions, enabling organizations to operate more efficiently.

For finance professionals, ERP systems fundamentally transformed the accounting functions such as manual journal entries, spreadsheet-based reconciliations, fragmented reporting, and disconnected financial records gradually gave way to integrated financial management.

Month-end closing became more structured, financial reports could be generated in real time, and regulatory compliance became easier through standardized workflows and centralized data management.

The next phase in ERP evolution was emergence of cloud computing. Cloud-based ERP platforms provided organizations with greater scalability, real-time collaboration, automatic updates, and improved accessibility.

Finance teams could access financial information from anywhere, integrate data across multiple business units, and leverage dashboards that offered greater visibility into organizational performance.

Today, the integration of Artificial Intelligence represents the most significant advancement in ERP technology since its inception. Rather than functioning solely as systems for recording and retrieving information, AI-powered ERP platforms actively assist users by interpreting data, identifying patterns, generating insights, and supporting decision-making.

Embedded AI assistants can summarize financial reports, explain unusual transactions, recommend corrective actions, forecast future performance, automate repetitive tasks, and even generate narrative reports using natural language.

The progression from transaction processing to intelligent decision support reflects a broader shift in the role of ERP systems. They are no longer limited to storing organizational data; they are increasingly capable of transforming that data into meaningful knowledge.

Prompt Engineering: The Language of AI-Powered ERP Systems

While these interfaces remain important, AI introduces a more intuitive way of working natural language. Natural language interaction enables users to communicate with ERP systems in much the same way they would engage with a colleague.

AI does not possess an inherent understanding of an organization's financial policies, reporting objectives, industry-specific requirements, or management expectations. Its responses are shaped by the instructions it receives. A vague request is likely to produce a generic response

This practice of designing effective instructions is known as prompt engineering. According to Glover and Peters (2025), Prompt engineering is the process of designing and refining everyday language text/voice prompts for use in Large Language Models.

In the context of AI-powered ERP systems, prompt engineering enables finance professionals to transform business questions into structured instructions that AI can interpret effectively.

An effective finance prompt is more than a simple question. It provides sufficient context for the AI to understand the business problem, the financial objective, the reporting period, and the expected format of the response. Consider the difference between the following prompts:

Prompt 1

Analyze this financial report.

Prompt 2

Analyze the attached income statement for the second quarter of 2026. Compare actual results against budget, identify all expense variances exceeding Z%, explain the possible business drivers, calculate the gross and operating profit margins, and present your findings in a concise executive summary for senior management.

The second prompt is significantly more effective because it defines the reporting period, specifies the analytical tasks, provides the business context, and clearly states the desired output.

By reducing ambiguity, it enables AI to generate responses that are more relevant and aligned with the user's needs.

Practical Applications of Prompt Engineering Across the Finance Function

The following examples illustrate how prompt engineering can be applied across key finance functions.

1. Financial Reporting and Management Reporting

Preparing financial reports often involves consolidating information from multiple sources, analyzing performance, and communicating findings to management. AI-powered ERP systems can assist by summarizing financial statements, explaining significant movements, and drafting management reports.

Example Prompt

Analyze the June 2026 Income Statement and Balance Sheet. Compare actual performance against budget, identify all material variances exceeding X%, explain the possible business drivers, and prepare a one-page executive summary for the Board of Directors. Present the findings in a table showing the variance, financial impact, and recommended management action.

2. Budgeting and Forecasting

Budget preparation and forecasting require finance professionals to evaluate historical trends, consider business assumptions, and assess future scenarios. AI can accelerate this process by analyzing historical performance and modelling different business outcomes.

Example Prompt

Using the last three years of revenue and expenditure data, prepare a cash flow forecast for the next six months under optimistic, expected, and pessimistic business scenarios. Clearly explain the assumptions used for each scenario and highlight potential liquidity risks.

3. Variance Analysis

Variance analysis is essential for monitoring financial performance and identifying operational issues. Rather than manually reviewing numerous reports, finance professionals can use AI to explain the underlying causes of significant deviations.

Example Prompt

Review the Budget versus Actual Report for the second quarter of 2026. Identify all revenue and expense variances greater than Y%, categorize them as operational or strategic, explain the likely causes, assess their financial impact, and recommend corrective actions.

4. Internal Audit and Risk Management

Internal auditors spend significant time reviewing transactions, identifying anomalies, and assessing compliance with internal controls. AI-powered ERP systems can assist by rapidly analyzing transaction data and highlighting unusual patterns that warrant further investigation.

Example Prompt

Review all journal entries posted outside normal business hours during the month-end closing process. Identify unusual transactions based on value, timing, account combinations, and user activity. Rank each exception according to its potential audit risk and explain why it requires further investigation.

5. Cash Flow and Treasury Management

Effective cash flow management is critical to maintaining organizational liquidity. AI can analyze historical payment behaviour, forecast cash inflows and outflows, and identify potential liquidity constraints.

Example Prompt

Analyze the organization's projected cash inflows and outflows over the next 90 days. Identify periods where liquidity may fall below the minimum operating threshold, explain the underlying causes, and recommend practical strategies to improve cash availability.

6. Compliance and Financial Governance

Maintaining compliance with accounting standards, internal policies, and regulatory requirements is a fundamental responsibility of every finance function. AI can assist by reviewing financial records, identifying inconsistencies, and highlighting areas requiring management attention.

Example Prompt

Review the attached lease accounting schedules and assess whether they comply with IFRS 16 recognition and measurement requirements. Identify any inconsistencies, explain the relevant accounting treatment, and recommend corrective actions where necessary.

7. Financial Analysis and Executive Decision Support

Finance professionals play a strategic role in helping management interpret financial information and evaluate business performance. AI can support this role by transforming complex financial data into concise and actionable insights.

Example Prompt

Analyze the organization's financial performance for the first half of 2026 using profitability, liquidity, efficiency, and solvency ratios. Identify the three most significant business risks, the three strongest performance indicators, and recommend strategic priorities for the next quarter.

Best Practices for Prompt Engineering in Finance

The following best practices can help finance professionals maximize the value of AI embedded within modern ERP systems.

1. Be Clear and Specific

Ambiguous prompts often result in broad or generic responses. AI performs more effectively when the task is clearly defined and the expected outcome is explicitly stated.

Instead of writing:

Analyze this report.

Consider:

Analyze the June 2026 Income Statement, identify all operating expenses that exceeded budget by more than x%, explain the likely business drivers, and recommend corrective actions.

Specific prompts reduce ambiguity and produce outputs that require less revision.

2. Provide Adequate Business Context

Financial data rarely exists in isolation. Business performance is influenced by industry conditions, organizational strategy, accounting policies, and operational circumstances. Including this context enables AI to generate more meaningful insights.

For example, specifying that the organization is a manufacturing company reporting under IFRS provides information that influences the interpretation of financial data.

The more relevant context provided, the more tailored the response is likely to be.

3. Define the Desired Output

Finance professionals communicate with diverse stakeholders, including senior management, boards of directors, auditors, regulators, lenders, and investors. Each audience requires information presented differently.

When prompting AI, specify how the results should be presented.

Examples include:

Executive summary
Board report
Financial commentary
Variance analysis table
Risk assessment matrix
Dashboard narrative

Defining the output format improves consistency and reduces the need for manual editing.

4. Break Complex Tasks into Smaller Steps

Complex financial analyses often involve multiple stages. Rather than asking AI to perform every task in a single prompt, divide the work into logical steps.

For example:

Step 1: Summarize the financial statements.

Step 2: Identify material variances.

Step 3: Explain the causes.

Step 4: Recommend management actions.

This iterative approach generally produces more accurate and transparent results while allowing finance professionals to review and validate each stage of the analysis.

5. Verify AI-Generated Outputs

AI should support not replace professional judgment. Although AI can accelerate analysis, it may occasionally misinterpret information, overlook important context, or generate incorrect conclusions.

Finance professionals should therefore:

Validate financial calculations.
Confirm compliance with accounting standards.
Review assumptions.
Cross-check figures against source data.
Exercise professional skepticism before relying on AI-generated outputs.

The responsibility for financial accuracy remains with the finance professional.

6. Protect Confidential and Sensitive Information

Financial records often contain confidential information relating to customers, suppliers, employees, pricing strategies, and organizational performance. Before sharing data with AI systems, finance professionals should understand their organization's data governance policies and ensure that confidential information is handled appropriately.

Where possible:

Remove personally identifiable information (PII).
Mask sensitive financial information.
Use approved enterprise AI solutions integrated within the organization's ERP environment.
Comply with applicable data protection and privacy regulations.

Responsible AI usage begins with responsible data management.

7. Refine Prompts Through Iteration

Prompt engineering is an iterative process. The first response generated by AI is not always the final answer.

Finance professionals should build on previous responses by asking follow-up questions such as:

Explain the largest variance in greater detail.

Present the findings as a dashboard narrative.

Compare these results with the previous financial year.

Highlight only the risks requiring immediate management attention.

Each refinement improves the quality and usefulness of the analysis while enabling AI to produce increasingly focused outputs.

8. Combine AI with Professional Expertise

AI excels at processing information quickly, identifying patterns, and generating summaries. Finance professionals contribute business understanding, ethical judgment, regulatory knowledge, and strategic insight.

The greatest value is achieved when these capabilities work together.

When Good AI Produces Poor Results: Common Prompting Mistakes in Finance

1. Asking Vague or Ambiguous Questions

One of the most common mistakes is providing AI with insufficient direction. Broad instructions leave too much room for interpretation, increasing the likelihood of generic responses.

Poor Prompt

Analyze this financial report.

The AI has no information about the reporting period, the business objective, the audience, or the aspects of the report that require attention.

Improved Prompt

Analyze the June 2026 Income Statement, identify operating expenses that exceeded budget by more than x%, explain the likely business drivers, and prepare a one-page executive summary for senior management.

Providing a clear objective and scope enables AI to deliver focused and actionable insights.

2. Failing to Provide Business Context

Financial data does not exist in isolation. Industry conditions, accounting policies, organizational strategy, and operational circumstances all influence how financial information should be interpreted.

Without sufficient context, AI may produce technically correct but commercially inappropriate recommendations.

For example, recommending aggressive cost reductions without understanding that the organization is investing in a strategic expansion could lead to misleading conclusions.

Providing relevant background allows AI to align its analysis with the organization's operating environment.

3. Ignoring Accounting Standards and Organizational Policies

Finance operates within well-defined regulatory and professional frameworks. AI cannot automatically determine whether an organization follows IFRS, GAAP, IPSAS, or internal accounting policies unless this information is explicitly provided.

For example, a prompt requesting advice on lease accounting should specify the applicable accounting framework to ensure that the response reflects the relevant reporting requirements.

Including accounting standards and organizational policies improves the reliability and compliance of AI-generated outputs.

4. Attempting to Solve Too Many Problems in One Prompt

Finance professionals sometimes expect AI to perform multiple unrelated tasks simultaneously, such as analyzing financial statements, forecasting cash flows, assessing investment opportunities, preparing board reports, and identifying compliance risks within a single request.

Although AI can manage complex instructions, combining too many objectives often reduces the clarity and quality of the response.

A more effective approach is to divide large assignments into smaller, sequential tasks. This allows each stage of the analysis to be reviewed and validated before proceeding to the next.

5. Accepting AI Outputs Without Verification

Perhaps the most significant risk is assuming that AI-generated responses are automatically accurate. While AI can produce impressive analyses and summaries, it may occasionally misinterpret financial data, overlook important information, or generate unsupported conclusions.

Finance professionals remain responsible for verifying:

Financial calculations.
Accounting treatments.
Regulatory compliance.
Supporting assumptions.
Source data.

Professional skepticism remains just as important in the age of AI as it has always been in the finance profession.

6. Sharing Sensitive Financial Information Inappropriately

Financial data often includes confidential information relating to customers, suppliers, employees, pricing, and organizational performance. Uploading sensitive information into AI tools without considering organizational policies or data protection requirements can expose organizations to significant legal, regulatory, and reputational risks.

Finance professionals should ensure that:

Only approved AI solutions are used.
Organizational data governance policies are followed.
Sensitive information is anonymized or masked where appropriate.
Confidential financial information is protected throughout the prompting process.

Responsible prompt engineering begins with responsible data stewardship.

7. Treating AI as a Decision Maker Rather Than a Decision Support Tool

AI is designed to assist finance professionals—not replace them.

While AI can rapidly analyze financial information and generate recommendations, it cannot fully understand organizational culture, strategic priorities, stakeholder expectations, or emerging business risks.

Critical financial decisions continue to require human expertise, professional judgment, ethical reasoning, and accountability.

The role of AI is to enhance decision-making by providing timely insights, not to make decisions on behalf of the organization.

Governance, Ethics, and Human Oversight: Building Trust in AI-Powered Finance

The integration of AI into ERP systems has introduce new governance responsibilities. As organizations increasingly rely on AI to assist with financial reporting, budgeting, forecasting, auditing, and compliance, maintaining trust in AI-generated outputs becomes just as important as improving productivity.

One of the most important principles of AI adoption in finance is that accountability cannot be delegated to technology.

While AI can generate analyses, recommendations, and forecasts, the responsibility for financial decisions remains with finance professionals and organizational leadership. AI should therefore be viewed as a decision-support tool rather than an autonomous decision-maker.

Organizations should establish clear governance frameworks that define how AI is used within financial processes. These frameworks should specify approved AI applications, data access permissions, validation procedures, documentation requirements, and approval workflows.

Data governance is equally important. AI systems often process highly sensitive financial information, including payroll data, customer records, supplier contracts, pricing strategies, and confidential management reports. Organizations should implement appropriate controls to ensure that data shared with AI complies with internal security policies and applicable data protection regulations.

Ethical considerations should also guide the adoption of AI within finance. AI-generated recommendations must be transparent, explainable, and free from inappropriate bias.

Finance professionals should understand the assumptions underlying AI-generated outputs, challenge recommendations where necessary, and avoid overreliance on automated analyses.

Internal controls should evolve alongside AI adoption. Organizations should regularly monitor AI performance, document significant AI-assisted decisions, review prompt quality, and periodically assess whether AI-generated outputs remain accurate and aligned with organizational policies.

Integrating AI governance into existing risk management and internal control frameworks helps ensure that innovation does not compromise accountability or financial integrity.

The Future of Prompt Engineering in Finance

Future ERP platforms are expected to move beyond responding to prompts and towards proactive intelligence. Instead of waiting for users to request analyses, AI will continuously monitor business performance, detect emerging risks, identify unusual transactions, recommend corrective actions, and alert management to significant changes in real time.(AFP Staff, 2026)

Prompt engineering will therefore evolve from asking isolated questions to guiding ongoing collaboration between finance professionals and intelligent systems.

The role of finance professionals will also continue to evolve. Routine transactional work will become increasingly automated, allowing finance teams to devote more time to strategic planning, business partnering, performance management, and value creation.

Educational institutions and professional accounting bodies are already beginning to incorporate AI-related competencies into curricula and continuing professional development programmes, recognizing that effective collaboration with intelligent systems is becoming an essential aspect of modern financial management.

Industry Perspectives: Why Prompt Engineering Matters

As AI becomes embedded within enterprise software, organizations increasingly recognize that achieving value from AI depends not only on technological investment but also on developing the capabilities of the people who use it.

Research consistently shows that organizations adopting AI expect improvements in productivity, operational efficiency, and decision-making. In finance, this means combining accounting expertise with digital competencies such as data literacy, AI literacy, and prompt engineering.

Major ERP vendors have also repositioned AI as a core component of enterprise applications rather than a standalone technology.

Microsoft has integrated Copilot across the Dynamics 365 ecosystem to assist users with financial analysis, reporting, reconciliation, and workflow automation.
SAP has introduced Joule as an AI assistant across its enterprise applications.
Oracle has embedded generative AI capabilities within Oracle Fusion Cloud ERP to enhance financial planning, procurement, and operational decision-making.

These developments illustrate a broader industry trend toward conversational enterprise systems that enable users to interact with business data using natural language.

The future finance professional will require a combination of technical accounting expertise, business understanding, digital literacy, and the ability to collaborate effectively with AI.

Industry analysts also predict that AI will fundamentally reshape enterprise software over the coming decade. Rather than replacing ERP systems, AI is expected to enhance them by transforming operational data into actionable insights and enabling more proactive decision-making.

Conclusion

Artificial Intelligence is transforming Enterprise Resource Planning systems from platforms that primarily record transactions into intelligent systems that assist finance professionals in analyzing data, generating insights, and supporting strategic decision-making.

As AI becomes embedded within everyday financial processes, the nature of work in finance is changing. The question is no longer whether organizations should adopt AI-powered ERP systems, but how effectively finance professionals can use them to create value.

References

AFP Staff. (2026). Trending articles and topics in Treasury and finance. AFP.
Glover, J., & Peters, R. (2025). Prompt engineering for finance 101 | Deloitte US. In Deloitte.
Hayes, M., & Downie, A. (2024). Artificial intelligence in ERP | IBM. In Ibm.com.

Connect with me via:
LinkedIn | Email

Introduction to Supervised Machine Learning for Beginners

Lucy Joan — Sat, 02 Aug 2025 08:10:15 +0000

Introduction

Machine learning might sound complex, but at its core, it’s simply about teaching computers to learn from examples — just like humans do.

One of the most common and important ways machines learn is called supervised learning. This article will walk you through what supervised learning is, how it works, and why it matters — all explained in simple, beginner-friendly language.

What is Supervised Learning?

Supervised learning is a type of machine learning where the model learns from a labeled dataset — meaning each input comes with a correct answer (or label).

Supervised learning is like having a teacher guide you through a subject, giving you the questions and the correct answers so you can learn how to solve similar problems on your own.

The “supervised” part means the machine is guided by examples.

These examples come with labels (the correct answers), so the machine knows what the right outcome should be.

Imagine you’re learning to identify fruits.
- You see a picture of a red, round fruit labeled “apple.”
- You see another picture of a long, yellow fruit labeled banana.”
- Over time, you can correctly identify an apple or a banana on your own because you’ve seen labeled examples.

That’s exactly what supervised learning does — but with data instead of fruit pictures!

Let's use an example:

Suppose we want to teach a computer to distinguish between emails that are "Spam" and those that are "Not Spam."

The Data: We gather a large collection of emails.
The Labels: For each email, we manually mark it as either "Spam" or "Not Spam."

Our labeled dataset would look something like this:

Email 1: "Congratulations! You've won a free vacation! Click here!" – Label: Spam
Email 2: "Meeting notes from yesterday's team sync." – Label: Not Spam
Email 3: "Claim your prize now! Limited time offer!" – Label: Spam
Email 4: "Your order has been shipped. Tracking number: XYZ123." – Label: Not Spam

How the Computer Learns (The Magic Behind the Scenes)

The computer then uses special algorithms (think of them as learning strategies) to look at all these examples. It starts to identify patterns and relationships between the content of the email and its label.

For instance, it might notice that emails with words like "free," "win," "prize," "urgent," or a lot of exclamation marks are more likely to be spam. Conversely, emails with professional language, specific sender addresses, or order tracking information are more likely to be legitimate.

The algorithm tries to build a model – which is essentially a set of rules or a mathematical function – that can accurately predict the label (Spam or Not Spam) for new, unseen emails.

The goal is for the model to learn the relationship between input features and the target output so it can make predictions on new, unseen data.

Importance of Supervised Machine Learning

Learns from labeled data to make accurate predictions.
Widely used in real-world tasks like:
- Spam detection
- Image and voice recognition
- Medical diagnosis
- Financial forecasting
Easy to understand and implement — models learn like humans do, through examples.
High accuracy and performance when trained with quality data.
Gives control over the learning process since outcomes (labels) are known.
Scalable and adaptable across many industries and use cases.

Examples of Supervised Learning

Supervised learning is all around us, even if we don’t notice it. Here are some relatable examples:

1. Email Spam Detection

Input: Words in your email
Output: Spam or not spam

2. Loan Approval

Input: Your income, credit score, etc.
Output: Approve or reject your loan

3. Medical Diagnosis

Input: Symptoms and test results
Output: Disease present or not

4. Voice Assistants (like Siri or Alexa)

Input: Your voice command
Output: Translated into text or action

5. Image Recognition

Input: Photo of a cat
Output: Label “cat”

Two Main Types of Supervised Learning

Supervised learning problems generally fall into two main categories:

1. Classification:

Predicting a category or a discrete class.

The model learns to answer "What kind of thing is this?"

Examples:

Is this email Spam or Not Spam? (Two classes)
Is this picture a Cat, a Dog, or a Bird? (Multiple classes)
Will this customer churn (leave) or not? (Two classes) Think of it as: Sorting things into buckets.

Common Classification Algorithms:

Logistic Regression
Decision Tree Classifier
Random Forest Classifier
Support Vector Machine (SVM)
K-Nearest Neighbors (KNN)
Naive Bayes

Output Type:

Discrete

Example: ["Yes", "No"], [0, 1], ["Dog", "Cat", "Bird"]

Algorithm	Definition	When to Use	Python Import Example	Notable Features
Logistic Regression	Predicts probability of class membership using a sigmoid function.	For binary classification (e.g., spam vs not spam).	`from sklearn.linear_model import LogisticRegression`	Fast, interpretable, works well with linear data
Decision Tree Classifier	Creates decision rules in a tree structure to classify data.	When interpretability and simple logic are needed.	`from sklearn.tree import DecisionTreeClassifier`	Easy to visualize, can overfit
Random Forest Classifier	Uses many decision trees and averages results for better accuracy.	When accuracy is more important than interpretability.	`from sklearn.ensemble import RandomForestClassifier`	Powerful, reduces overfitting, handles non-linearity
Support Vector Machine (SVM)	Finds a hyperplane that best separates classes.	When classes are well separated or data is high-dimensional.	`from sklearn.svm import SVC`	Works well in small datasets, robust to outliers
K-Nearest Neighbors (KNN)	Classifies based on the majority of k nearest neighbors.	When the data is low-dimensional and relationships are local.	`from sklearn.neighbors import KNeighborsClassifier`	Simple, no training phase, memory intensive
Naive Bayes	Uses Bayes' Theorem assuming feature independence.	When working with text (e.g., sentiment or spam detection).	`from sklearn.naive_bayes import MultinomialNB`	Very fast, good baseline for text classification

2. Regression:

Predicting a continuous numerical value.

The model learns to answer "How much?" or "What’s the value?"

Examples:

What will the price of a house be based on its size, location, and number of rooms?
How many sales will a store make next month based on advertising spend and seasonality?
What will the temperature be tomorrow? Think of it as: Predicting a number on a scale.

Common Regression Algorithms:

Linear Regression
Ridge Regression
Lasso Regression
Decision Tree Regressor
Random Forest Regressor
Support Vector Regressor (SVR)

Output Type:

Continuous

Example: 23.7, 1500, -4.8

Algorithm	Definition	When to Use	Python Import Example	Notable Features
Linear Regression	Predicts a target value by fitting a straight line through the data.	When the relationship between variables is linear.	`from sklearn.linear_model import LinearRegression`	Simple, fast, interpretable
Ridge Regression	Linear regression with L2 regularization to reduce overfitting.	When you want to penalize large coefficients but keep all variables.	`from sklearn.linear_model import Ridge`	Adds stability by shrinking coefficients
Lasso Regression	Linear regression with L1 regularization for feature selection.	When you want to shrink some features to zero (ignore them).	`from sklearn.linear_model import Lasso`	Useful for sparse data or reducing model complexity
Decision Tree Regressor	Splits data into branches and predicts a value at the leaves.	When the data has non-linear relationships or clear decision boundaries.	`from sklearn.tree import DecisionTreeRegressor`	Easy to understand, can overfit without pruning
Random Forest Regressor	An ensemble of decision trees for regression.	When you want accurate predictions and want to avoid overfitting.	`from sklearn.ensemble import RandomForestRegressor`	Handles non-linearities well, robust and powerful
Support Vector Regressor (SVR)	Uses support vectors to fit a curve within a margin of tolerance.	When the data is high-dimensional or not linearly separable.	`from sklearn.svm import SVR`	Works well with complex, small datasets

The Machine Learning Recipe: Step-by-Step Guide to Supervised Learning

Supervised learning follows a structured process — just like following a recipe. Here's how you build a machine learning model from scratch:

1. Collect Data
Gather examples that include both input features and the correct answers (labels).

Example: A list of emails labeled as “Spam” or “Not Spam.”

2. Clean the Data
Fix missing values, remove duplicates, and correct errors to ensure data quality.

Clean data = better learning.

3. Split the Data
Divide your dataset into:

Training Set (usually 70–80%) – used to teach the model
Test Set (20–30%) – used to see how well the model learned

This prevents the model from just memorizing everything.

4. Choose the Right Model
Pick an algorithm that suits your problem:

Classification (e.g., Logistic Regression)
Regression (e.g., Linear Regression)

5. Preprocess the Data
Prepare the features so the model can understand them:

Normalize/standardize numeric values
Encode categories (like Yes/No → 1/0)

6. Train the Model
Feed the training data into the model so it can learn patterns between inputs and outputs.

7. Test the Model
Use the test data to check how well the model performs on unseen examples.

8. Evaluate Performance
Use metrics to measure accuracy:

For classification: Accuracy, Precision, F1-Score

Metric	What it Means	Use When
Accuracy	% of correct predictions	Classes are balanced
Precision	Of those predicted positive, how many were correct?	Cost of false positives is high (e.g., email spam)
Recall	Of all actual positives, how many did we find?	Cost of false negatives is high (e.g., disease)
F1-Score	Harmonic mean of precision & recall	When you want balance
Confusion Matrix	Shows TP, FP, FN, TN	To visualize classification mistakes

True Positive(TP) - Model correctly predicted Positive (and it actually is Positive).
False Positive(FP) - Model predicted Positive, but it's actually Negative (aka "False Alarm").
False Negative(FN) - Model predicted Negative, but it's actually Positive (aka "Missed it").
True Negative(TN) - Model correctly predicted Negative (and it actually is Negative)             |

For regression: Mean Squared Error (MSE), R² Score

Metric	What it Means	Use When
Mean Absolute Error (MAE)	Average of absolute errors	Easy to understand
Mean Squared Error (MSE)	Penalizes big errors more	Common and popular
Root Mean Squared Error (RMSE)	Square root of MSE	Same units as target
R² Score (R-squared)	How much variance is explained	1 is perfect, 0 is bad

9. Tune the Model (Optimize)
Adjust the model’s settings (called hyperparameters) or try different algorithms to improve results and fix:

Underfitting – Model is too simple
Overfitting – Model memorized training data too well

10. Deploy the Model
Once it performs well, integrate the model into a real system — like predicting spam emails or product prices in an app.

11. Monitor and Update
Track how the model performs over time. As new data comes in, you may need to retrain or update the model to keep it accurate.

Types of Errors in Supervised ML

Error Type	Meaning	How to Fix
Underfitting	Model is too simple	Use a more complex model or add features
Overfitting	Model memorized instead of learning	Simplify model or use more data
Bias	Model is unfair or always wrong in one way	Use fairer data, tune model
Variance	Model changes too much on small changes	Use regularization or simpler model

Summary

The main parts of machine learning include data, features, a model, and an algorithm. The process starts by feeding data into the model, which is trained using the algorithm. After training, the model is tested and evaluated. Once it's accurate enough, it can make predictions on new, unseen data.

Think of it like teaching someone to bake:
You gather ingredients (data), follow a recipe (algorithm), practice baking (training), test how good the cookies are (evaluation), and eventually, bake confidently (prediction).

Supervised learning is just one branch of machine learning. There’s also unsupervised learning, which finds patterns in data without labels (like grouping similar customers), and reinforcement learning, where an agent learns by trial and error (like teaching a robot to walk).

Check out my github: Github

Introduction to SQL Using PostreSQL

Lucy Joan — Sat, 26 Apr 2025 13:44:52 +0000

PostgreSQL for Absolute Beginners: Your First Step into Data Management

Imagine managing millions of Instagram posts or handling customer orders worldwide without confusion.
Behind the scenes, a powerful tool called SQL (Structured Query Language) makes it all possible. In this beginner’s guide, we’ll walk you through PostgreSQL, one of the world’s most popular database systems, and help you write your first SQL commands.

. What is SQL?
. How to set up PostreSQL
. Basic database concepts (schemas, tables, rows)
. Writing your first SQL queries
. A mini hands-on project
. Pro tips and next steps

Introduction to SQL?

SQL (structured query language) is a programming language for storing and processing information in relational database management systems (stores information in tabular form, with rows and columns).

It is used extensively for storing, manipulating and retrieving data in systems such as MySQL, PostgreSQL etc.

You can use SQL statements to store, update, remove, search, and retrieve information from the database - making it essential for tasks like data analysis, software development, and database administration.

You can also use SQL to maintain and optimize database performance.
Whether you're building applications, generating reports, or analysing big data, SQL provides the tools needed to interact with complex datasets quickly and efficiently. Thanks to its versatility and widespread adoption, SQL remains one of the most important skills in today’s data-driven world.

In this article, we will focus on PostgreSQL, a powerful and widely-used open-source relational database management system (RDBMS). You will learn what makes PostgreSQL a preferred choice for developers, data analysts, and businesses managing complex data. In our next article, we will provide a step-by-step guide on how to download, install, and set up PostgreSQL on your computer, so you can start building and managing your own databases.

How to Set Up PostgreSQL: Step-by-Step Guide for Beginners

Setting up PostgreSQL correctly is essential for creating secure and high-performing databases. In this guide, we’ll cover everything you need to know to download, install, and configure PostgreSQL on your machine, whether you are using Windows, macOS, or Linux.

Step 1: Download PostgreSQL
. Go to the official PostgreSQL website: https://www.postgresql.org/download/
. Choose your operating system (Windows, macOS, Linux).
. Download the latest stable version of PostgreSQL.

Step 2: Install PostgreSQL

For Windows:
. Open the downloaded installer.
. Follow the setup wizard.
. Choose an installation directory.
. Set a password for the PostgreSQL superuser (postgres).
. Select the default port (5432) unless you need to change it.
. Complete the installation.

For macOS:
Install using the PostgreSQL installer package or use Homebrew

For Linux:
Install using your package manager

Step 3: Set Up PostgreSQL
. Verify the installation.
. Open pgAdmin (the PostgreSQL management tool) or connect via command line:

psql -U postgres

Basic database concepts

Before you dive into writing SQL queries, it’s important to understand the core building blocks of databases. These basic concepts will help you feel more confident as you start creating and managing your own data structures.

Term	Meaning
Database	A collection of organized data stored electronically
Schema	A logical container that holds tables, views, and other database objects. Think of it like a folder that keeps everything organized
Table	A structured set of data organized into rows and columns, similar to a spreadsheet
Row	A single record inside a table, containing related data for one item
Column	A specific attribute or field that holds one type of data, such as a name, email, or date of enrollment

Visual Example of a Database Structure

Database: school
│
└── Schema: public
    │
    ├── Table: students
    │    ├── id (column)
    │    ├── name (column)
    │    ├── enrollment_date (column)

In this example:

Database: school is the main database.

Schema: public groups related tables.

Table: students stores information about students.

Columns: id, name, and enrollment_date are the fields capturing each student's data.

Your first SQL queries

Now that you understand the basic building blocks of databases, it’s time to write your first SQL queries! SQL (Structured Query Language) is the tool you’ll use to interact with your database — to create tables, insert data, and retrieve information.

Here are a few simple SQL commands to get you started:

1. Create a Database

First, you need a database to store your tables.

CREATE DATABASE hr_system;

Explanation:

CREATE DATABASE creates a new database named hr_system.

This is where all your data and tables will live.

✅ Tip: After creating it, make sure to connect to the hr_system database before proceeding.

\c hr_system

2. Create a Schema

A schema helps organize your tables inside the database.

CREATE SCHEMA hr_schema;

Explanation:

CREATE SCHEMA creates a container (folder-like structure) for your tables.

It’s especially useful when you have many tables or want to separate different areas of your system.

✅ Tip: You can also skip this if you want to use the default public schema.

3. Create a Table

To create a new table for storing student information:

CREATE TABLE hr_schema.HR_DATASET (
    employee_id SERIAL PRIMARY KEY,
    first_name VARCHAR(100),
    last_name VARCHAR(100),
    department VARCHAR(100),
    age CHAR(4)
    gender VARCHAR(10)
    enrollment_date DATE
);

Explanation:

CREATE TABLE hr_schema.HR_DATASET means you’re creating the table HR_DATASET inside hr_schema

employee_id is a unique, auto-incremented number for each employee.

first_name and last_name store the employee’s name.

department indicates which department the employee belongs to.

age is stored as a fixed-length character field.

gender and enrollment_date capture additional details about the employee.

4. Insert Data into the Table

INSERT INTO HR_DATASET (first_name, last_name, department, age, gender, enrollment_date)
VALUES ('John', 'Doe', 'Finance', '29', 'Male', '2024-09-01');

Explanation:

INSERT INTO tells SQL where to add the new record.

We provide values for first_name, last_name, department, age, gender, and enrollment_date.

5. Retrieve Data with SELECT

To view all the records in the table:

SELECT * FROM HR_DATASET;

Explanation:

SELECT * fetches all columns and all rows from the HR_DATASET table.

6. Update Data

If you need to update an employee's department:

UPDATE HR_DATASET
SET department = 'Human Resources'
WHERE employee_id = 1;

Explanation:

UPDATE modifies existing records.

SET changes the value of a specific column.

WHERE ensures only the correct record is updated.

7. Delete Data

To remove an employee from the table:

DELETE FROM HR_DATASET
WHERE employee_id = 1;

Explanation:

DELETE FROM removes specific data.

Using WHERE is important to avoid deleting all records accidentally.

Hands-on Project: Customer Order Management System

In this project, you will create a simple Customer Order Management System. This system will help manage customer details, the products they purchase, and the orders they place. The goal is to create a database that can handle customer and order data effectively.

Project Objective:
You will:

Create a database and schema.
Create tables for customers, orders, and products.
Insert records.
Query the data.
Update and delete records.

Step 1: Create the Database

Create a database called customer_orders_system:

CREATE DATABASE customer_orders_system;

Objective: This database will hold all the data for customer orders, including customer details, products, and orders.

Step 2: Create the Schema

Create a schema called sales_schema:

CREATE SCHEMA sales_schema;

Objective: Organize your data into a dedicated schema for the sales system.

Step 3: Create the Customers Table

Now, create a table to store customer information:

CREATE TABLE sales_schema.customers (
    customer_id SERIAL PRIMARY KEY,
    first_name VARCHAR(100),
    last_name VARCHAR(100),
    email VARCHAR(100),
    phone VARCHAR(20),
    address TEXT
);

Objective: This table will store customer details, such as name, email, phone number, and address.

Step 4: Create the Products Table

Next, create a table to store product information:

CREATE TABLE sales_schema.products (
    product_id SERIAL PRIMARY KEY,
    product_name VARCHAR(100),
    price DECIMAL(10, 2),
    stock_quantity INT
);

Objective: This table holds the product details — name, price, and available stock.

Step 5: Create the Orders Table

Create a table to store order details:

CREATE TABLE sales_schema.orders (
    order_id SERIAL PRIMARY KEY,
    customer_id INT REFERENCES sales_schema.customers(customer_id),
    order_date DATE,
    total_amount DECIMAL(10, 2)
);

Objective: This table holds the orders placed by customers. It includes references to the customer table and stores the total amount of the order.

Step 6: Create the Order Items Table

Since an order can contain multiple products, we need a table to link products to orders:

CREATE TABLE sales_schema.order_items (
    order_item_id SERIAL PRIMARY KEY,
    order_id INT REFERENCES sales_schema.orders(order_id),
    product_id INT REFERENCES sales_schema.products(product_id),
    quantity INT,
    subtotal DECIMAL(10, 2)
);

Objective: This table stores the relationship between orders and the products within each order, including the quantity and subtotal for each product in the order.

Step 7: Insert Customer Records

Now, insert some customer records into the customers table:

INSERT INTO sales_schema.customers (first_name, last_name, email, phone, address)
VALUES
('Alice', 'Johnson', 'alice.johnson@example.com', '123-456-7890', '123 Elm Street'),
('Bob', 'Smith', 'bob.smith@example.com', '234-567-8901', '456 Oak Avenue');

Objective: Add some customers to the system.

Step 8: Insert Product Records

Now, insert some products into the products table:

INSERT INTO sales_schema.products (product_name, price, stock_quantity)
VALUES
('Laptop', 1000.00, 50),
('Smartphone', 500.00, 200),
('Tablet', 300.00, 150);

Objective: Add products that customers can purchase.

Step 9: Insert Orders

Let’s now create some orders for the customers:

INSERT INTO sales_schema.orders (customer_id, order_date, total_amount)
VALUES
(1, '2025-04-01', 1500.00),
(2, '2025-04-02', 800.00);

Objective: Create orders for Alice and Bob.

Step 10: Insert Order Items

Add products to each order by inserting records into the order_items table. Let’s assume Alice bought a Laptop and a Tablet, while Bob bought a Smartphone:

INSERT INTO sales_schema.order_items (order_id, product_id, quantity, subtotal)
VALUES
(1, 1, 1, 1000.00),  -- Alice bought 1 Laptop
(1, 3, 1, 300.00),  -- Alice bought 1 Tablet
(2, 2, 1, 500.00);  -- Bob bought 1 Smartphone

Objective: Link the products to the orders, including the quantity and subtotal for each item.

Step 11: Retrieve Data

Let’s query the system to retrieve records for all tables:

SELECT * FROM sales_schema.customers;
SELECT * FROM sales_schema.products;
SELECT * FROM sales_schema.orders;
SELECT * FROM sales_schema.order_items;

Step 12: Update Product Stock

Let’s say Bob purchased the last smartphone. Update the stock quantity for the smartphone:

UPDATE sales_schema.products
SET stock_quantity = stock_quantity - 1
WHERE product_name = 'Smartphone';

Objective: Learn how to update product inventory after an order is placed.

Step 14: Delete an Order

If you need to remove an order, say Alice’s order with ID 1:

DELETE FROM sales_schema.orders
WHERE order_id = 1;

Objective: Practice deleting records from the database.

Step 15: Review and Reflection

Once you’ve completed this mini project:

Review the relationships between customers, orders, products, and order items.

Experiment by adding more customers, products, and orders.

Reflect on how your system is tracking orders and inventory.

Pro tips: Understanding SQL Data Types and Constraints

When working with SQL, it’s essential to understand the various data types and constraints to properly define your tables and columns. Here’s a quick summary of some commonly used SQL data types and constraints:

Term	Type	Purpose	Example	Range/Details
Integer	`INT`	Stores whole numbers	`age INT`	Typically -2,147,483,648 to 2,147,483,647
Large Integer	`BIGINT`	Stores large whole numbers	`employee_id BIGINT`	Up to 9,223,372,036,854,775,807
Character	`VARCHAR(n)`	Stores variable-length strings	`first_name VARCHAR(100)`	Max length varies (commonly up to 255 characters)
Fixed Character	`CHAR(n)`	Stores fixed-length strings (padded with spaces if shorter)	`gender CHAR(10)`	Max length typically 255 characters
Large Text	`TEXT`	Stores large strings of text (no predefined length)	`address TEXT`	Varies by database (e.g., 65,535 characters in MySQL)
Decimal	`DECIMAL(p, s)`	Stores fixed-point numbers, precision (`p`) and scale (`s`)	`price DECIMAL(10, 2)`	Max precision typically 65 digits
Floating Point	`FLOAT`	Stores floating-point numbers with approximate precision	`rating FLOAT`	Range typically -1.0E+308 to 1.0E+308
Date	`DATE`	Stores date values (year, month, day)	`enrollment_date DATE`	Format: `YYYY-MM-DD`
Time	`TIME`	Stores time values (hours, minutes, seconds)	`order_time TIME`	Format: `HH:MM:SS`
Datetime	`DATETIME`	Stores both date and time values	`order_timestamp DATETIME`	Format: `YYYY-MM-DD HH:MM:SS`
Timestamp	`TIMESTAMP`	Stores date and time with automatic updates when records are modified	`created_at TIMESTAMP`	Auto-updates on insert/update
Boolean	`BOOLEAN`	Stores true/false values	`is_active BOOLEAN`	Typically `TRUE` or `FALSE`
Primary Key	`PRIMARY KEY`	Uniquely identifies records in a table	`employee_id SERIAL PRIMARY KEY`	Must be unique, cannot be NULL
Foreign Key	`FOREIGN KEY`	Creates a relationship between tables by referencing another table's primary key	`customer_id INT REFERENCES customers(customer_id)`	Ensures referential integrity
Not Null	`NOT NULL`	Ensures that a column cannot contain a `NULL` value	`first_name VARCHAR(100) NOT NULL`	Column must always have a value
Auto-Increment	`SERIAL`	Automatically generates unique numbers for new records (commonly used for primary keys)	`employee_id SERIAL PRIMARY KEY`	Similar to `AUTO_INCREMENT` in MySQL
Unique	`UNIQUE`	Ensures all values in a column are distinct	`email VARCHAR(100) UNIQUE`	No duplicate values allowed

Where To Go From Here

Now that you've learned the basics of SQL, it’s time to take your skills to the next level. Here’s what you should focus on next:

Learn SQL JOINs — Master how to combine data from multiple tables using SQL JOINs like INNER JOIN, LEFT JOIN, and RIGHT JOIN. This is essential for working with relational databases.
Understand SQL Indexes — Learning how SQL indexes work will help you optimize your database performance and make your queries run faster.
Work on SQL Project Ideas — The best way to improve your SQL skills is through practice. Start by building small projects like:

- A blog database to manage posts, authors, and comments

- An inventory management system to track products, stock levels, and 
  suppliers

Lastly, Consistent SQL practice and real-world projects will help you master database management and prepare you for advanced topics like stored procedures and database optimization.

DEV Community: Lucy Joan

Unlocking the Full Potential of AI-Powered ERP Systems

The Evolution of ERP Systems: From Automation to Intelligence

Prompt Engineering: The Language of AI-Powered ERP Systems

Practical Applications of Prompt Engineering Across the Finance Function

1. Financial Reporting and Management Reporting

2. Budgeting and Forecasting

3. Variance Analysis

4. Internal Audit and Risk Management

5. Cash Flow and Treasury Management

6. Compliance and Financial Governance

7. Financial Analysis and Executive Decision Support

Best Practices for Prompt Engineering in Finance

1. Be Clear and Specific

2. Provide Adequate Business Context

3. Define the Desired Output

4. Break Complex Tasks into Smaller Steps

5. Verify AI-Generated Outputs

6. Protect Confidential and Sensitive Information

7. Refine Prompts Through Iteration

8. Combine AI with Professional Expertise

When Good AI Produces Poor Results: Common Prompting Mistakes in Finance

1. Asking Vague or Ambiguous Questions

2. Failing to Provide Business Context

3. Ignoring Accounting Standards and Organizational Policies

4. Attempting to Solve Too Many Problems in One Prompt

5. Accepting AI Outputs Without Verification

6. Sharing Sensitive Financial Information Inappropriately

7. Treating AI as a Decision Maker Rather Than a Decision Support Tool

Governance, Ethics, and Human Oversight: Building Trust in AI-Powered Finance

The Future of Prompt Engineering in Finance

Industry Perspectives: Why Prompt Engineering Matters

Conclusion

References

Introduction to Supervised Machine Learning for Beginners

Introduction

What is Supervised Learning?

How the Computer Learns (The Magic Behind the Scenes)

Importance of Supervised Machine Learning

Examples of Supervised Learning

Two Main Types of Supervised Learning

1. Classification:

2. Regression:

The Machine Learning Recipe: Step-by-Step Guide to Supervised Learning

Types of Errors in Supervised ML

Summary

Introduction to SQL Using PostreSQL

PostgreSQL for Absolute Beginners: Your First Step into Data Management

Table of contents

Introduction to SQL?

How to Set Up PostgreSQL: Step-by-Step Guide for Beginners

Basic database concepts

Your first SQL queries

1. Create a Database

2. Create a Schema

3. Create a Table

4. Insert Data into the Table

5. Retrieve Data with SELECT

6. Update Data

7. Delete Data

Hands-on Project: Customer Order Management System

Step 1: Create the Database

Step 2: Create the Schema

Step 3: Create the Customers Table

Step 4: Create the Products Table

Step 5: Create the Orders Table

Step 6: Create the Order Items Table

Step 7: Insert Customer Records

Step 8: Insert Product Records

Step 9: Insert Orders

Step 10: Insert Order Items

Step 11: Retrieve Data

Step 12: Update Product Stock

Step 14: Delete an Order

Step 15: Review and Reflection

Pro tips: Understanding SQL Data Types and Constraints

Where To Go From Here