Defining the Problem: The First Step to Building a Successful AI Model
In the world of AI and Machine learning, most attention goes to models, algorithms and data science techniques. However, there’s a crucial, often overlooked step that happens long before any code is written: Defining the problem. AI model are usually built to solve a particular problem, without a well-defined problem, even the most sophisticated AI models can produce irrelevant or misleading results.
How do we go about defining a problem?
Defining a problem is the process of translating a real-world situation or business challenge into a clear, structured problem that can be addressed using AI. We need to break down the challenge into measurable objectives, defining success criteria, and ensuring that the AI model’s goals align with the business objectives.
For example, a retail company might start with a vague goal of “improving sales.” However, a well-defined problem would look more like, “How can we use AI to predict which products are most likely to be purchased by customers during the holiday season, based on their browsing history and purchase patterns?” The second version is clear, measurable, and gives direction for model development.
Why is defining a problem so important?
There have been multiple case studies, where data teams started working on a problem without understanding it, which eventually failed. We need to sit down and understand what are we trying to achieve, before we start rushing towards data analysis, and building machine learning models.
Focus and clarity: A well-defined problem helps focus the AI team on what truly matters. It ensures that every step of the project—from data collection to model deployment—aligns with the intended goals. Without clear problem definition, teams risk building models that answer the wrong questions.
Efficient Resource Allocation: Defining the problem early helps allocate resources—time, data, and computational power—more effectively. A vague problem can lead to wasted effort, with teams chasing after irrelevant data or building overly complex models that don't serve the end goal.
Measurable success: Defining the problem includes criteria for success. Suppose a team is trying to figure out “how many people are likely to continue their subscription”, here success might be defined as “Achieving a 90% accuracy in identifying the customers that eventually continue their subscription”.
The key elements of a well-defined problem..
Understand the Business Objective: Before diving into technical details, it is important to understand the business goal behind an AI project. Example questions to think about - “What is the overall objective of the project?”, “What do we want to achieve with this project?”, “How AI will help us finish this project?". Sometime you also need to ask whether AI is actually needed for this task? Some instances we might not need AI, but just a simple automation that will solve our problem. Therefore understand what we need is very important.
Breaking the problem down into specific and measurable objectives: Once you are done understanding the business objective, think about how it can be broken down into specific questions that AI can help us solve.
Broad problem - Suppose your broad problem is “Increase customer satisfaction.”
Defined problem - "Can we predict customer service ratings based on interaction history and agent response times?"
Define the Output: Consider what kind of output you expect from the model. Is it a classification (e.g., predicting which category a customer belongs to?), a regression (predicting sales figures), or a recommendation system (suggesting products)? Defining this upfront prevents confusion down the line and ensures the AI model delivers the right kind of insights.
Determining Success Criteria: A model’s success is measured using different methods, the most common you might have heard of is “accuracy”. Success criteria should also consider business outcomes, not just model performance. Suppose a model predicts some credit card frauds with a 90% accuracy rate, you also need to how this affects the business outcome?
Constraints and assumption: Every AI project has constraints—whether it’s limited data, computational resources, or time. Identifying these constraints early helps the team make realistic decisions. Similarly, documenting any assumptions (such as assuming that customer data is reliable) is vital to avoid misalignment later in the project.
Common mistakes while defining a Problem..
Being too vague: One of the most common mistakes is starting from a poorly defined problem like “improving customer experience”. This leaves too much room for interpretation, which can lead to misaligned efforts.
Specific - Reduce average customer wait times by 15% in the next quarter by predicting peak call volumes using AI.
Ignoring Stakeholders: The stakeholders are very important. Failing to communicate and engage with stakeholders can lead to AI solutions that do not meet real world applications.
Skipping the Success Criteria: Without defining what success looks like, it becomes impossible to judge whether the AI solution is working. Clearly establish metrics like accuracy, precision, recall, or even financial KPIs to track progress.
How to frame an AI problem correctly?
Start with the business problem: We always have to find what we are trying to solve? After finding a business problem we need to work backwards and see how we can go about solving it.
Identifying what data is available: Data is the first step after defining the problem. If we don’t have enough data available, we won’t be able to progress towards solving our problem. Any AI model however advanced it may be, it won’t do well if the input data is not good quality.
Collaborate across teams: Defining the problem should be a collaborative effort between business stakeholders and technical teams (data scientists, engineers). Business teams understand the problem, while technical teams understand what’s feasible. This collaboration ensures that the problem is both important to the business and solvable using AI.
Real world Examples
Netflix’s recommendation system: Netflix’s problem was to improve user retention by recommending movies and shows that match users’ preferences. They clearly defined the problem: “How can we predict which titles users are most likely to enjoy based on their previous viewing behavior?” The success metric was user engagement and viewing time, and the solution resulted in personalized recommendations that have become a hallmark of their platform.
Amazon’s product search algorithm: Amazon needed to ensure customers find relevant products quickly. The defined problem was: “How can we rank products in the search results that are most likely to be purchased by the user, based on their search query and past behavior?” The success metric was an increase in the click-through rate on search results and the number of purchases per search session.
Defining the problem is often the most critical step in an AI project, yet it is frequently overlooked or rushed. Without a clear understanding of the problem, AI teams may spend time building solutions that don't address the business's real needs. By focusing on problem definition, you set a strong foundation for the entire AI process, ensuring that your models are relevant, impactful, and aligned with your business objectives. Always ensure you have defined the problem before starting on an AI journey, it will save time, energy and resources of the organisation.
