Welcome to the 4th Part of AWS Machine learning associate Series where we explain key topics to prepare for the exam..In this post, let's explore Whether your business problem needs ML solution or regular solution which is the first step required for building Machine Learning Model. As usual, i tried to keep it beginner friendly in such a way, even a non tech user or beginner can understand.So, even if you are non tech user, feel free to join our journey..
The Rice Shop, the Rain, and the Cloud
A story about how a small-town dream met Machine Learning — and how you can too..
Chapter 1 – The Problem Nobody Could Name (Defining the Business Problem for ML)
Imagine you are in a deep forest, where the smelled of cardamom tea and the monsoon clouds rolled in like slow-moving giants, you see a rice shop run by a man named Ravi.
Ravi wasn’t a tech guy. He didn’t own a laptop. His world was measured in bags of rice, customer smiles, and the weight of coins in his pocket at the end of the day.
But Ravi had a problem.
Some months, he bought too much rice. The extra bags sat in the corner, growing stale, attracting weevils. Other months, he bought too little, and customers left disappointed.
He didn’t know it yet, but in the language of machine learning problem formulation, Ravi’s struggle was called “Defining the business problem” — the first step in the ML project lifecycle and the CRISP-DM framework. It’s the same kind of thinking a US store uses to plan turkey sales before Thanksgiving, an Australian surf shop uses before summer, a coffee exporter in Brazil uses before harvest, an e‑commerce giant in China uses before Singles’ Day, or a stock trader uses to anticipate share prices before earnings announcements.
📘 What We Learned
Machine learning problem formulation – Spotting a real‑world issue (like Ravi’s unpredictable rice sales) and realising it could be turned into a question for a computer to answer.
Business understanding in ML – Knowing why the problem matters before thinking about how to solve it; for Ravi, it’s about avoiding waste and keeping customers happy.
Problem statement for ML – A short, clear description of the challenge; here, it might become “Predict next month’s rice demand.”
ML project lifecycle – The journey an ML project takes from problem to solution; Ravi is standing right at the starting line.
CRISP‑DM – A common roadmap for ML projects; Ravi’s story begins in its first phase, Business Understanding.
Problem framing ML – Deciding exactly what question the model should answer; Ravi hasn’t framed it yet, but the need is clear.
Chapter 2 – The Stranger with the Cloud (Business Problem vs ML Problem — The First Step in the AWS Machine Learning Workflow)
One rainy afternoon, a young woman named Asha walked into Ravi’s shop. She was visiting her grandmother in the village, but in the city, she worked as a data scientist.
She listened to Ravi’s woes and smiled.
“You know, this is exactly the kind of thing machine learning can help with,” she said.
Ravi frowned. “Machine… what?”
Asha explained:
“Machine learning is when we teach computers to learn from examples instead of giving them step‑by‑step instructions. But before we teach them, we have to frame the problem — like telling a student exactly what exam they’re preparing for.”
She pulled out her phone and scribbled:
Business Problem: Reduce rice overstock and shortages.
ML Problem: Predict the number of rice bags needed next month (a regression problem — predicting a number).
“This,” she said, “is the first step in the AWS machine learning workflow.
📘 What We Learned
How to define a business problem for ML – Start with the real‑world challenge, then translate it into an ML task; Ravi’s shortage/overstock issue became a prediction problem.
Business problem vs ML problem – The business goal is the “why” (reduce waste), the ML problem is the “what” (predict demand).
Translating business problem to ML – Converting a shopkeeper’s concern into a measurable prediction task.
Business problem to data science problem – Framing the issue so it can be solved with data and algorithms.
Defining a business case for ML – Explaining why ML is worth the effort; for Ravi, it meant saving money and keeping customers happy.
Identifying a business problem for AI – Spotting situations where AI can help; unpredictable demand was one for Ravi.
Business objective for ML model – The measurable target; e.g., “Reduce overstock by 20% in 3 months.”
Use cases for machine learning – Real‑world scenarios where ML adds value; from rice shops to stock price forecasting.
When to use machine learning – When patterns are too complex for simple rules; Ravi’s sales varied with weather, festivals, and prices.
Machine learning business value – The tangible benefit ML brings; for Ravi, less waste and more profit.
Chapter 3 – The Hunt for Data (Data Selection for Machine Learning — Finding the Right Ingredients)..
The next morning, Asha returned with a steaming cup of chai and a question:
“Ravi, do you have records of how much rice you sold each month?”
Ravi chuckled. “Records? I have this old notebook.”
Inside were scribbles — dates, quantities, prices, even notes like ‘Festival week – sold extra’ or ‘Heavy rain – fewer customers’.
Asha’s eyes lit up.
“This is gold. In ML terms, this is your training data — the raw material your model will learn from, just like a sports analyst studies past match stats before predicting the next game.”
She explained:
Structured data – neatly organised in tables (sales numbers, dates).
Unstructured data – free‑form text, images, or audio (customer comments, photos).
They decided to:
Enter the sales history into a spreadsheet.
Add weather data from a public source.
Mark festival dates.
“This,” Asha said, “is data selection for machine learning. In AWS, we’d store this in Amazon S3, clean it with AWS Glue, and maybe explore it in a Jupyter Notebook.”
📘 What We Learned
Data selection for machine learning – Choosing the right information to teach the model; Ravi’s sales, weather, and festival data were key.
Structured vs unstructured data for ML – Tables vs free‑form text/images; both can be useful.
Data quality in machine learning – Better data means better predictions; messy data leads to poor results.
Data sources for ML – Where your data comes from; Ravi used his notebook plus public weather records.
Data preparation for machine learning – Cleaning and formatting data so the model can understand it.
Chapter 4 – The Measure of Success (Identifying ML Success Criteria — Business Metrics vs Model Metrics)
By the third day, Ravi was curious.
“Asha, how will we know if this… machine learning thing… is working?”
Asha grinned.
“That’s the third step — identifying ML success criteria. We need to decide how to measure success before we start, the same way an investor decides what ‘good performance’ means before buying a stock.”
She explained:
Model Metric: Mean Absolute Error (MAE) — average difference between predicted and actual values.
Business Metric: Reduction in wasted rice bags per month.
“If the model predicts 50 bags and you sell 52, that’s good. If it predicts 50 and you sell 100, that’s bad. In Amazon SageMaker, we can track these metrics automatically.”
📘 What We Learned
Defining ML success criteria – Setting clear targets before training begins.
Business metrics vs model metrics – Real‑world impact vs statistical performance.
Evaluating ML model performance – Checking predictions against actual results.
Machine learning success metrics – Numbers like accuracy or MAE that show how well the model works.
Chapter 5 – The First Prediction
Weeks later, after entering data, cleaning it, and running it through a regression model in Amazon SageMaker, Asha showed Ravi a simple dashboard.
It read:
Next month’s demand: 48 bags (±3)
Ravi stocked 48 bags. By the end of the month, he had sold 49.
He didn’t understand all the technical magic — the ML pipeline, the model evaluation, the AWS cloud computing — but he understood this: For the first time in years, he had no stale rice in the corner.
It was like a farmer in Brazil seeing a coffee yield forecast come true, or a stock analyst watching their price prediction hit the mark.
📘 What We Learned
Regression – A type of ML that predicts numbers; Ravi’s model predicted bags of rice, just as a stock model predicts share prices.
ML pipeline – The sequence of steps from raw data to predictions; Asha’s process from notebook to dashboard was Ravi’s first.
Model evaluation – Checking how close predictions are to reality; Ravi’s 48 vs 49 bags was a strong result.
Next day, when Ravi asked magic She laughed, “You know, Ravi, your rice shop is basically a case study for the exam’s Problem Formulation domain — the same skills you’d need if you were designing a model to predict cricket match attendance in Melbourne or holiday shopping spikes in New York.”
Ravi smiled. “So I’m helping you study, and you’re helping me sell rice. Fair trade.”
The story ends there... Now, let's come to our exam point of view..
The key thing to know is which problem requires ML solution and which one doesn't need it. Which is the foundational for our exam.
Not every problem needs an ML solution. Sometimes, the simplest, most traditional approach is the smartest choice.
If Ravi’s sales had been steady year‑round, with only minor fluctuations, a simple spreadsheet and a bit of common sense might have solved his problem faster and cheaper than building a predictive model. In the same way, a shop in Sydney might just check last year’s summer sales before ordering surfboards, or a coffee exporter in Brazil might rely on decades of harvest records without spinning up an AWS SageMaker instance.
Machine learning shines when:
Patterns are complex and influenced by many factors.
The cost of being wrong is high.
There’s enough quality data to learn from.
But when the problem is straightforward, traditional analytics, domain expertise, and human judgment can be more efficient. The real skill — whether you’re a shopkeeper, a data scientist, or an AWS ML exam candidate — is knowing which tool to use for which job.
Ravi’s story worked out because the problem was a good fit for ML: multiple variables, seasonal changes, and enough historical data to train a model. But the lesson isn’t “always use machine learning.” It’s “always understand the problem first” — and then choose the right approach, whether that’s a cutting‑edge algorithm or a pencil and paper.
With that we will conclude the post, how you now learn whether your business problem requires ML Model or not, which is foundation before developing the ML Solutions.
Also Read:
What Is the Fediverse? - Welcome to New World!
The Ultimate Guide to Installing a Local LLM on Your Computer
