Author(s): Elangoraj Thiruppandiaraj Originally published on Towards AI. Introduction: The Problem With AI Ideas As a Data Scientist, I often find that every second person I talk to has a potential AI use case in mind. I get questions like: Can AI solve this? Can this be automated? Can I create an agent for my daily tasks? Will AI take over my job? I completely understand the curiosity — the excitement to bring AI into every aspect of life is real. But before diving in, we need to ask ourselves a few important questions: Does this idea have real-world impact? Will it save time or effort? Will it genuinely make life better? Too often, these questions are overlooked — and that’s exactly why, according to Gartner, up to 30% of generative AI projects will be abandoned after the proof-of-concept (POC) stage by 2025. Common reasons include poor data quality, unclear business value, and spiraling costs. Informatica also notes that most failed AI initiatives can be traced back to weak planning at the initiation phase. In this article, I want to highlight what I believe is the most important and often overlooked phase of an AI project: the beginning. From brainstorming and requirement gathering to early scoping — these steps help you determine the true value of your idea and lay the groundwork for writing a clear, realistic proposal. That’s what gives your AI project a real shot at turning into something useful, and more importantly, successful. Photo by Jason Goodman on Unsplash 1. Brain stroming AI ideas that actually matter These days, it feels like anything and everything can be automated with AI — from generating Excel formulas and managing energy usage to scanning barcodes and booking tickets with AI agents. The possibilities are exciting, but here’s the catch: not every task that can be automated needs AI. In many cases, what people really need is a simple, rule-based solution or a custom program — not a complex AI system. For example, automating an Excel workflow might just require someone to write efficient formulas or a well-structured script with proper test cases. Throwing AI at it might add unnecessary complexity without real benefit. That’s why choosing the right kind of problem to solve with AI is the first step. You need to separate shiny ideas from those that offer real, measurable value. When I brainstorm ideas for AI projects, I ask a few grounding questions to filter out the noise: What’s the return on investment? Is the value created (time saved, insights gained, errors reduced) greater than the time, effort, and cost of building the AI solution?What cost are we saving by doing this? Is it cutting down hours of manual effort? Reducing licensing costs? Avoiding mistakes?Will this help us sell more or improve profits? AI should contribute to business growth — either by improving sales, enhancing customer experience, or enabling new capabilities.Who is the primary user or consumer of this solution? Understand their workflow, needs, and whether they’d actually use what you’re building.Is there an existing solution that can be reused or adapted? Using AI or building something from scratch isn’t always the best starting point — in many cases, you can achieve your goal faster by leveraging existing tools, services, or solutions. A good AI idea starts with a real problem — one that’s painful enough to solve, and valuable enough to justify the effort. Brainstorming with this lens keeps the project grounded and increases the chances of it making a meaningful impact. Photo by Mika Baumeister on Unsplash 2. Gathering Data and Understanding Requirements Once you’ve nailed down your idea, the next — and often most critical — step is gathering the right data and defining clear requirements. This is where your AI project really begins to take shape. The reality? Accessing relevant, high-quality data is tough. Around 70% of manufacturers report data issues — from outdated and incomplete records to poor formatting. While clean data is ideal, what matters most is relevance, diversity, and coverage. A diverse dataset reduces bias and improves generalisation. For example, DeepSeek — a Chinese open-source LLM — faced backlash for restricted responses due to training on narrow, region-specific data. You also need to consider data privacy regulations like GDPR. Legal constraints can limit how data is used, even if it’s valuable — so understanding those boundaries early is key. Just as important as the data is having clear, well-communicated requirements. Without knowing exactly what you’re solving, even the best model or dataset won’t deliver results. Good requirements align teams, reduce confusion, and lead to better outcomes. Whenever I’m gathering data and shaping requirements, these are the questions I try to answer: Do we have relevant data to address this problem? Without relevance, even the cleanest dataset won’t move the needle.What’s the quality of that data? Outdated or inconsistent data leads to poor model performance and wasted effort.Is the data internal or external? Where is it stored? Knowing the source and location helps assess accessibility, ownership, and integration challenges.Are the project requirements clear and aligned with business goals? Clear goals ensure the solution solves the right problem and delivers real value.Are there any data protection rules or legal restrictions we need to consider? Legal and compliance constraints can affect whether and how the data can be used in your AI solution. Spending time on this phase can feel slow, but it’s what sets you up for success — or failure. No data, no AI. No clarity, no direction. Photo by UX Indonesia on Unsplash 3. Scoping the Project: Drawing the Line By now, you know why you’re building this AI solution and what data and requirements you have. The next step is to clearly define what the project will — and won’t — deliver. This is where scope comes in. A well-defined scope avoids confusion, prevents scope creep, and keeps the project realistic. I usually split it into two buckets: In Scope: These are the deliverables you can commit to based on the current data, time, and resources — for example, a working API, a WebApp prototype, data preprocessing scripts, documentation, or a codebase. Only include items that are feasible and agreed upon for this phase. Out of Scope: Just as important is stating what’s not included. For instance, a WebApp might be in scope, but full deployment or enterprise integration may not be. Defining this early — often in terms of an MVP (Minimum Viable Product) — helps set expectations and avoid confusion later. When I define scope, these are the questions I ask: What are the confirmed deliverables for this phase? Be specific — prototypes, reports, models, APIs, etc.Are we committing to deployment or just a POC/MVP? Clarifying this upfront prevents misaligned expectations.What features or tasks are explicitly out of scope? Listing them helps avoid scope creep and future conflict.Are roles and responsibilities clearly defined? Knowing who owns what avoids last-minute surprises.Is the scope aligned with available resources and time? Your scope should match your team’s bandwidth and capabilities. 4. Writing a Project Proposal That Actually Gets Read Once the scope is defined, the next step is to write the project proposal — your project’s elevator pitch. This document introduces the project to key stakeholders by outlining the objectives, scope, deliverables, timelines, and estimated effort or cost. Unlike a project charter, which is typically created during the planning phase, the proposal comes earlier — during the initiation phase — and serves as the foundation for alignment. It doesn’t need to be overly technical or long. The goal is simple: clearly explain what you’re building, why it matters, and what’s needed to make it happen. There are many great templates and resources out there, but I focus on answering a few core questions when I draft a proposal: What is the problem we’re trying to solve? A clear problem statement gives the project purpose and urgency.What is the objective of this AI solution? This shows the intended outcome — whether it’s cost savings, automation, or improved insights.Who are the key stakeholders? Identifying the decision-makers, users, and contributors ensures alignment from the start.What is the scope and what are the agreed deliverables? Outline exactly what will be delivered and what’s excluded.What are the next steps or outline of the proposed solution? Set expectations for how the work will proceed and what’s needed to get started. A well-written proposal doesn’t just inform — it builds confidence. It helps everyone understand the value of the work and ensures you’re not building in isolation. Photo by 2H Media on Unsplash Conclusion This phase of project planning might not be the most glamorous, but it’s without a doubt the most essential. When you take the time to ask the right questions, set realistic goals, and strip away jargon and assumptions, you lay the foundation for a project that’s actually built to succeed. A clear proposal, backed by well-defined scope and agreed-upon acceptance criteria, doesn’t just get sign-off — it gives you a way to track measurable impact and keep the project moving in the right direction. In the end, strong planning isn’t about slowing things down — it’s about making sure what you build is worth building. Keep in Touch I’d love to hear your thoughts on this approach to planning AI projects. How do you go about brainstorming ideas, gathering requirements, or writing proposals? Have you faced challenges turning a great AI concept into something real? Let’s connect and share experiences — drop a comment, share the article, or reach out to me on LinkedIn. Follow me on Medium for more insights on AI and project execution. Let’s keep the conversation going! 🚀 #ArtificialIntelligence #ProjectManagement #AIProjectPlanning #TechLeadership #MachineLearning Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor. Published via Towards AI