Blog

Are you a team leader looking to expand into AI but unsure where to start? Your software engineering team has proven their excellence, and now clients are asking about AI capabilities. With frameworks like PyTorch and TensorFlow, approaches from fine-tuning LLMs to building RAG systems, and other endless technical choices, selecting the right first project is crucial.

This article presents a battle-tested framework that helps you identify winning AI projects before investing significant resources. By following these five criteria, you’ll be able to focus on technical execution with confidence, knowing you’re building something valuable.

Worry about these five criteria before thinking about GH100 vs A1000

As a team leader with a strong software engineering track record, you’re facing mounting pressure to incorporate AI into your solutions. Your competitors are making moves, your clients are asking questions, and your team is eager to apply their newly acquired AI skills. But with the overwhelming array of technologies and approaches available, choosing the wrong starting point could disappoint clients and damage team morale.

I’ve advised teams considering various AI initiatives, potentially saving substantial effort and resources by identifying fundamental issues early. For instance, a proposed real-time gaze detection system for sports advertising seemed technically fascinating but addressed a non-existent problem. Similarly, trading bot projects often fail because they demand impossible levels of accuracy, while anti-cheating AI solutions frequently create more problems than they solve.

In AI initiatives, selecting the right project is far more crucial than choosing specific technical tools. While your team can master any framework – be it TensorFlow or PyTorch, Rust or Python – even perfect technical execution cannot overcome poor strategic choices. Making the wrong project selection can waste months of effort and substantial resources, regardless of your engineering excellence.

After 15+ years in the field and numerous successful AI implementations, I’ve developed a beginner-friendly 5-point checklist that can validate any AI project in minutes. This framework helped guide successful projects, including a medical screening tool where we invested seven years and approximately 1M euros in developing non-invasive diagnostic solutions – an investment justified by clear clinical needs and well-defined success metrics.

1. Pain Level Assessment Verify that you’re addressing a real, measurable business pain point. For instance, in medical applications, the “business pain” often translates to actual patient discomfort from invasive procedures. Ideally, there should be a solution in place for this particular problem. If not, at least ensure you have performance measurements for it. Without baseline metrics, you’ll struggle to demonstrate value, no matter how sophisticated your AI solution becomes. Organizations rarely fail to measure what truly matters to their operations. If you can’t quantify the pain point, you might be building a solution in search of a problem.
2. Perfection Requirements Be wary of use cases demanding 100% accuracy. The trading bot scenario illustrates this perfectly – markets punish even minor prediction errors. Successful AI projects plan for and accommodate acceptable margins of error.
3. Adversarial Environment Assessment Consider whether your system will face active opposition. Anti-cheating systems often fail because they operate in highly adversarial environments where users actively work to defeat them. These scenarios require specialized approaches and carry higher risks.
4. Human Oversight Capability The best AI implementations maintain human oversight. For instance, in medical applications, doctors remain the final decision-makers, with AI serving as a supportive tool rather than a replacement.

The secret to success? Look for projects that raise no red flags across all five criteria. When you find such a “green” project, you can confidently focus on technical execution, knowing you’re building something valuable.

Want to learn more about selecting and executing successful AI projects? Follow me for insights from my experience implementing AI solutions across many verticals.