February 18, 2025
“A brilliant solution to the wrong problem can be worse than no solution at all: solve the correct problem.” – Donald A. Norman, The Design of Everyday Things
“We need an AI strategy!”
If you’re a business leader, you’ve heard this phrase in countless meetings. Perhaps you’ve said it yourself. And for a good reason – AI’s transformative potential is undeniable. Yet, for many organizations, there’s a wide, vacuous gulf between recognizing AI’s importance and implementing solutions that deliver business value.
The challenge isn’t a lack of enthusiasm or resources. Many companies have the budget, stakeholder support, and voracious desire to incorporate AI into their technology stack. What they often lack is a clear understanding of where to begin – how to bridge the gap between technical possibilities and practical business solutions.
And this disconnect is understandable. The fast-paced landscape of statistical modelling has grown increasingly complex, with terms like “Machine Learning,” “Large Language Models,” and “Computer Vision” often used ambiguously. Business leaders find themselves caught between high-level AI marketing and deep technical discussions, searching for a practical middle ground.
In this article, we’ll cut through the noise and provide a framework for turning AI aspirations into actionable plans. We’ll explore the practical differences between AI and Machine Learning to highlight an overlooked advantage of the former. Then, we’ll elucidate the trade-off between timeline to value and business impact for different kinds of AI projects. Finally, we’ll share a structured approach to identifying and evaluating AI opportunities in your organization.
Understanding the fundamental difference between AI and Machine Learning isn’t just an academic exercise – it’s paramount to identifying practical opportunities in your organization. While both technologies require training on data, their implementations differ significantly in ways that matter to your bottom line.
Traditional ML models must be trained on your organization’s specific dataset. This means you need substantial, clean, and relevant data before you can even begin. This step itself can be laborious even for mature organizations. It’s like teaching a new, entry-level employee by showing her examples from your company’s history – she can only learn from what you show her.
Modern AI models, particularly Large Language Models (LLMs) and Computer Vision Models (CVMs), flip this dynamic on its head. These models come pre-trained on vast amounts of data and are ready to help “out of the box.” Rather than requiring you to feed them more data, they can instead help you unlock and structure the data you already have. This is like hiring a new, high-level employee who’s already learned a lot from her experience at other companies – she’s already great at her job, you simply need to point her toward the correct resources with few instructions.
Here’s where it gets interesting: CVMs, while often overshadowed by the conversational and literary abilities of LLMs, are particularly powerful yet underutilized tools. Think about all the information trapped in your organization’s PDFs, scanned documents, and images. CVMs can extract specific content from these sources and structure it for use in downstream systems – whether that’s feeding an ML training pipeline or populating a database. You’re not just implementing AI; you’re using it to enhance your entire data ecosystem.
This distinction is crucial because it shapes how you approach AI implementation. You won’t be building your own LLM or CVM – instead, you’ll leverage pre-trained models from industry leaders like Google, OpenAI, and Anthropic. This means you can focus on applying these tools to solve specific business problems rather than tackling the massive challenge of training models from scratch.
When it comes to AI projects, understanding the timeline to value can be confoundingly complex. Unlike traditional software development, where you can often predict outcomes with reasonable certainty, AI implementations exist on a spectrum of empirical uncertainty – some solutions are straightforward to validate and implement, while others require extensive testing and infrastructure development.
Consider two contrasting examples: a CVM pipeline that extracts text from documents can be implemented and validated quickly – you’ll know almost immediately if it’s accurately pulling the right information. This is a “quick win” that can start delivering value within weeks of implementation. On the other hand, developing a chatbot powered by LLMs is a more intensive undertaking. While the core technology is powerful, there are several prerequisites: you need to build the underlying Retrieval-Augmented Generation (RAG) infrastructure, carefully design the user experience, and conduct significant end-user testing. The value here isn’t directly measurable until you’ve completed substantial development work.
To help you plan your AI initiatives, here’s how a few common implementations typically compare in terms of time to value and business impact:
The key is understanding that longer implementation times don’t necessarily indicate lower value – rather, they often reflect the complexity of the problem being solved and the thoroughness required for a robust solution. The most successful organizations typically start with quick wins to build momentum and expertise, then gradually tackle more complex implementations as their AI capabilities mature.
When it comes to evaluating AI opportunities, many organizations find themselves in a paradoxical situation: there’s no shortage of problems to solve, but determining which problems are well-suited for AI solutions can be challenging. The key isn’t finding problems – it’s about asking “Can AI effectively help me with this specific problem?”
Let’s break this down into a practical framework that will help you identify and prioritize AI opportunities in your organization.
Start by casting a wide net. Look around your organization for problems of any kind – inefficiencies in processes, accessibility issues, technical pain points, or even minor irritations that people have learned to “live with.” Don’t filter at this stage; collect at least ten problems, even if they seem unrelated to AI. Remember: automation opportunities often hide in plain sight, disguised as “that’s just how we’ve always done it.”
Now comes the critical evaluation phase. First, rank your collected problems by business value, regardless of whether AI seems like an obvious solution. This creates your priority list. Then, for each problem, write down potential AI-driven solutions. Don’t worry about feasibility yet – this exercise helps translate business problems into technical opportunities and creates a concrete starting point for internal discussions.
This is where technical reality meets business ambition. For each potential solution, map out what the final implementation might look like. Some important questions to address:
• Which cloud services would support this solution?
• What company data would we need to access?
• How would we integrate this with existing systems?
This is fundamentally a technical discussion, so involve your engineering teams or technical partners early. Their insights will be crucial in identifying potential roadblocks and opportunities.
By this point, you have a clear view of both problems and potential solutions, along with their relative complexity and impact. Some solutions will naturally appear more intricate than others, and some problems will seem more amenable to AI intervention. The point is to develop an execution plan that balances quick wins with long-term ambitions.
A strategic approach often starts with implementing simpler solutions, even if their immediate business impact seems modest. These “quick wins” serve multiple purposes:
• They build confidence in AI implementations
• They establish basic infrastructure that can be reused
• They create momentum for more ambitious projects
• They provide valuable learning experiences for your team
Remember: if your solution to a problem requires another solution, you haven’t solved the problem – you’ve just changed it. Keep your initial implementations focused and achievable, then build upon that foundation for more complex solutions.
A last piece of advice: a picture is worth a thousand words, but so are a thousand words. Look for opportunities in both images and text – these are areas where modern AI excels. Moreover, successful AI implementation often opens doors to cloud computing capabilities that can benefit your entire organization.
The journey from “We need AI!” to “we’re successfully using AI.” doesn’t have to be daunting. By understanding the practical distinction between AI and ML, you’re better equipped to identify which solutions require custom training and which can leverage existing models. By recognizing the varying timeline to value, you can set realistic expectations and plan accordingly.
Most importantly, armed with a systematic framework for evaluating AI opportunities, you’re now positioned to move beyond general AI enthusiasm to specific, actionable projects. The main idea in this article is to start small but think big. Begin with focused solutions that can demonstrate value quickly, while building the foundation for more ambitious implementations.
Remember that you’re not alone in this. The challenges you face in implementing AI are shared by many organizations, but the solutions don’t have to be identical. Your unique business context, existing processes, and specific pain points will guide you toward the most valuable applications of AI for your organization.
Ready to begin? Start by gathering your team and applying the framework we’ve discussed above. List out your organization’s pain points, evaluate them through the lens of AI capability, and begin mapping out your path to implementation. The future of AI in your organization isn’t about what the end solution looks like – it’s about developing a strategy that you can execute on today to solve the correct problem.