Identifying High-ROI Opportunities in AI

The Path to Transformative AI: Focus on Core Principles

AI promises to reshape business, but many leaders struggle to cut through the hype and find truly valuable opportunities. Drawing on my experience helping hundreds of companies—from agile startups to global businesses—use AI for transformative products and services, I’ve seen what separates success from costly experiments. The winners focus on value, following five proven principles that deliver extraordinary results:

1. Target transformative gains (3-10X improvements), not incremental tweaks
2. Build success systematically: first Quality, then Efficiency, finally Innovation
3. Turn proprietary data into competitive advantage
4. Choose stable technology that’s ready for tomorrow
5. Design systems where humans enhance AI, not slow it down

By following these principles, you’ll move beyond AI experiments to create lasting, step-change improvements for your company. This article offers a clear framework to find and carry out high-ROI AI projects that give a real competitive edge.

1. Aim for Transformation, Not Tweaks: Pursue 3x to 10x+ Gains

When evaluating potential AI projects, the main idea is to seek transformative impact, not just small tweaks. Businesses should aim for improvements that are three times, ten times, or even more than current performance. This focus on big gains matters for several reasons.

AI projects often need a lot of upfront investment—not just in technology, but also in data preparation, talent, and process changes. Companies naturally resist change, so you need a strong reason to overcome it. Small improvements, like a 20% or 30% efficiency gain for one task, are positive. But they may not create enough drive or show enough benefit to justify the cost or overcome company inertia. Such projects can easily stall after the proof-of-concept stage or fail to become part of core business operations.

Big, bold projects, though sometimes more complex, are often the better opportunities. They get leadership’s attention, energise teams, and provide a clear story for change. Ambitious projects can unlock more creativity and resourcefulness. Industry analysis shows that leading AI adopters focus their investments on a few high-impact projects. Boston Consulting Group (BCG) found that companies focusing on about three to four major AI use cases expect more than double the ROI compared to firms that spread their efforts across many smaller projects¹. These leading companies usually put over 80% of their AI budget into bold projects designed to reshape whole functions or launch new offerings².

Order-of-magnitude improvements are realistic. Many case studies show AI can deliver such results. For example, JPMorgan Chase transformed its contract analysis process using AI, cutting review time from 360,000 hours a year to seconds—a huge leap in productivity³. In pharmaceuticals, generative AI has shortened drug discovery cycles from five years to as little as twelve months. A large manufacturer used AI-driven predictive maintenance to cut unplanned downtime by 68%, saving $4.2 million per year. These examples show AI can deliver not just small gains, but big shifts in what companies can do and how well they perform.

To find these transformative opportunities, businesses can use structured assessment frameworks. One common approach is the Impact vs. Feasibility Matrix, which helps rank potential AI projects by business impact (like revenue, cost savings, or strategy) and feasibility (like technical complexity, data readiness, and time or cost to implement). This helps you pick “quick wins” (high impact, high feasibility) and strategic bets (high impact, lower feasibility but with a clear path to value). Another useful lens is ROI Category Assessment, which sorts AI opportunities by the type of value they create: cost savings, revenue growth, or new revenue streams. This ensures a balanced portfolio that meets strategic financial goals. The most successful AI strategies aim high, targeting breakthroughs that truly move the needle for the company.

2. The Strategic Sequence: Quality First, Efficiency Second, Innovation Third

Maximising AI’s strategic impact often means tackling goals in a clear order: first, use AI to raise Quality; once quality is strong, go after Efficiency; and finally, with quality and efficiency in place, drive real Innovation. This QEI flywheel – Quality, Efficiency, Innovation – is a solid path to lasting AI value.

Quality First: The Bedrock of Trust and Value

First, focus AI projects on making existing products, services, or processes better. “Quality” here means accuracy, reliability, consistency, fairness, robustness, and security. Using AI just to do more, faster, or cheaper—without focusing on quality—risks lowering value, hurting your brand, and losing customer trust. But when the main goal is better, higher-quality results—like 99+% accuracy in a key process, much lower defect rates, or far better customer retention—AI becomes a powerful tool for excellence.

Improving quality with AI matters to both your team and your customers. Customers always value higher quality, and better internal processes mean happier employees and less rework. High-quality AI builds trust, which broader adoption needs. Reliable, high-quality data and AI-driven insights are the foundation for real efficiency gains and new ideas.

Efficiency Second: Capitalising on a Quality Foundation

Once you have strong quality, chasing efficiency is the next step. Often, better efficiency is a direct and happy result of higher quality; processes with fewer errors and less rework are simply more efficient.

Even if efficiency needs a dedicated effort, starting from established quality has a much bigger impact. With quality in place, AI can streamline workflows, automate repetitive tasks, cut costs, and speed up decisions—without hurting results. For example, after using AI to make demand forecasts much more accurate, Danone could use those insights to improve its supply chain, cut waste, and keep products in stock. At StarKist Foods, an AI system cut a complex planning process from 16 hours to under one by unifying and smartly processing demand and production data. These efficiency gains mean better margins, faster service, and freed-up resources for innovation⁴.

Innovation Third: Building on Experience and Strength

Radical innovation—creating new capabilities, products, services, or even business models with AI—is tempting. But chasing big innovation too soon, before you have quality and efficiency, is risky. Companies that skip these steps can face project failure, wasted investment, and damage to their core business.

But when a company builds quality and efficiency first, it gains real strengths:

• Deeper Expertise: Teams get hands-on experience building and running AI systems.
• Better Data: Focusing on quality and efficiency means better data collection, management, and understanding.
• Proven Abilities: The company learns what AI can really do in its world.
• Confidence: Success with quality and efficiency builds trust in AI and momentum for more.

This foundation of experience, better data, proven abilities, and confidence makes real innovation possible. New ideas for using AI often come naturally from what teams learn during the quality and efficiency phases. Teams are better able to spot and deliver truly valuable innovations. So, the strategic order of Quality, then Efficiency, then Innovation offers a structured, lower-risk way to realise AI’s full potential.

3. Your Data, Your Edge: Start with Proprietary Data Sets

In an era where AI models and algorithms are becoming increasingly accessible, even common, a company’s unique, proprietary data is a main source of lasting advantage. While many companies can use similar AI tools, the data they feed into these tools ultimately sets apart the insights they gain and the value they create. So, a key principle for high-ROI AI is to base projects on your company’s own rich data.

Powerful generative AI models can perform well with less training data than older models. Even so, starting with high-quality, relevant, unique data gives a big advantage. Proprietary data—reflecting your company’s specific operations, deep expertise, unique customer interactions, history, and market position—acts as a powerful lever for building effective, different AI systems. This is especially true if competitors cannot easily get this data.

With your own data, you can:

• Build and Evaluate with Precision: Train and test AI models using data that matches your real-world scenarios, leading to more reliable and relevant results.
• Uncover Unique Insights: Find patterns and links specific to your business and market that generic models or public data might miss.
• Create Defensible Solutions: Build AI tools tailored to your unique strengths and customer needs, creating a moat that’s hard for others to cross. As IBM Consulting’s AI leader said, “Every AI vendor has access to public data… What they don’t have is your company’s data.”⁵
• Enhance Customisation and Personalisation: Fine-tune AI models to deliver highly personalised experiences for your customers or highly optimised solutions for your internal processes. A B2B software company, for instance, integrated its CRM emails, support chats, and call transcripts into an AI lead-scoring model, resulting in a 28% increase in qualified lead conversion—a clear benefit from its own, proprietary data.

Using proprietary data doesn’t always mean you need the huge data lakes common in earlier AI. While more data can help, the quality, relevance, and uniqueness of the data matter most. Finding specific, high-value data—like customer support logs, detailed transactions, sensor readings, or curated internal knowledge—can give a strong base for good AI projects. An organisation may not yet have the perfect data set. But if it has a clear plan to gather, manage, and use such data, it’s well-set for AI success.

Investing in data readiness means building a data-aware culture throughout your company. Instead of focusing only on huge data lakes or big data science teams, successful companies help everyone spot valuable data. Engineers see chances in log files, sales teams notice patterns in customer talks, and support staff spot recurring issues in ticket data. When teams see how their daily work creates useful data, they naturally improve data quality, suggest new sources, and find ways to use that data in AI. Your data becomes a living asset that grows stronger as your whole team learns to nurture and shape it. When many people in the company work with data, this creates lasting competitive advantage, more so than any technology.

4. Build Smart: Use Stable Technology Poised for the Future

Navigating the fast-changing AI technology world takes a careful balance between using current, reliable tools and getting ready for future breakthroughs. The best strategy is to build AI systems that deliver real value today using proven, generally available technologies, while also designing them so they can easily benefit from, and adapt to, even more powerful AI as it arrives. This means avoiding the twin traps of chasing the latest hype and building systems that quickly become outdated.

The Perils of the Bleeding Edge vs. The Value of Proven Tech

While the appeal of the very latest AI models and techniques can be strong, building core business solutions on these new technologies carries big risks. Bleeding-edge AI is often:

• Poorly Understood: Its capabilities, limitations, and potential failure modes may not yet be fully characterised.
• Less Reliable: It may exhibit unexpected behaviours, lack robustness, or have “rough edges” in terms of performance and integration.
• More Expensive: Early access to novel technologies often comes at a premium cost, both for the technology itself and for the scarce talent required to implement it.
• Lacking a Clear ROI Path: It’s hard to see how unproven technologies will pay off or make business sense.

Instead of chasing the hype, a more practical approach is to focus on the “last generation of generally available AI capabilities” – technologies that may have been cutting-edge three to six months ago but have since matured, become well-documented, are offered at better prices, and have a clearer track record of success. As Deloitte’s research suggests, focusing on “demonstrated use cases in proven areas accelerates ROI” and helps spread AI across the organization. This lets companies build reliable, affordable systems with predictable performance, making their AI investments safer.⁶

Designing for Future Compatibility

Building on stable technology does not mean ignoring future advances. A smart AI strategy looks ahead and gets ready for them. To “future-proof,” consider these key steps:

• Build Modular Systems: Design AI with modular parts and clear interfaces (APIs). You can then upgrade or swap out pieces—like a specific AI model—without rebuilding the whole application.
• Focus on Fundamentals: Invest in things that keep their value even as models change. Build high-quality, well-managed proprietary data, make easy-to-use interfaces, and understand the business processes AI will help.
• Manage Data Well: Focus on data pipelines, data quality, and data management. As models change, the data stays a key asset.
• Avoid Short-Lived Fixes: Don’t spend a lot on complex workarounds for current AI model limits, especially if new platforms will soon fix those limits. For example, building tricky prompt engineering to handle a model’s small context window (how much text the model can handle at once) might become pointless when the next model has a much bigger one. These fixes add future maintenance problems and only give short-lived benefits.
• Adopt a “Late Beta” Mindset: Some cloud providers, like Google Cloud, suggest using AI services after they’ve matured through internal use and pilot programs. This helps ensure more stable and reliable systems before you roll them out widely.

The goal is to create AI solutions that work well today and can adapt as AI quickly advances.

5. Human Ingenuity, AI Power: Enhance Systems, Don’t Create Bottlenecks

A key to success in AI is how well you blend in human expertise. Always aim to design systems where people enhance AI’s abilities and results, not systems that always need people and create bottlenecks. The most popular consumer AI products, like chatbots and digital assistants, often involve constant human interaction. But applying this model to business AI systems can be problematic. It can lead to systems that are expensive to run, inefficient at scale, and unreliable under pressure.

Humans as Strategic Enhancers of AI

Human intelligence and domain expertise are essential for making AI systems truly effective, trustworthy, and aligned with business goals. People play key roles:

• Curate and Label Data: Experts prepare high-quality training data, label complex data types, and make sure datasets reflect the business. This “teaches” the AI well.
• Check and Improve Models: Evaluators check AI performance, spot biases, validate outputs for accuracy and relevance, and stress-test models against edge cases or ambiguous scenarios. This repeated feedback improves model quality.
• Build Knowledge Structures: Experts create and maintain knowledge bases, rule sets, or ontologies (knowledge structures) that help AI understand context and reason more accurately, especially in specialised fields.
• Oversee Ethics and Governance: People oversee the ethical side of AI, ensuring fairness, transparency, accountability, and compliance with rules. This includes watching for unintended consequences and making value-based judgments that AI alone cannot.
• Handle Exceptions and Complex Problems: While AI can automate routine tasks, experts handle new, complex, or sensitive situations that AI can’t yet manage or that need human judgement.

When people focus on these valuable tasks, AI gets much stronger. The goal is to build AI systems that, once enhanced and guided by human expertise, can work on their own and handle most of the workload.

Avoiding Human Bottlenecks in AI Systems

Conversely, organisations should be careful with AI projects that make routine tasks newly reliant on people, if these projects don’t offer enough value in return. If an AI system cannot perform its core functions well without a human constantly guiding it or correcting its outputs, it may not be a worthwhile investment. Such systems can fail to scale, prove more costly than alternatives, and may not deliver the efficiency or quality improvements that justify AI adoption.

Consider the different models of human-AI collaboration:⁷

• Tiered Review Systems: AI handles tasks, humans monitor and intervene for exceptions (e.g., financial trading oversight).
• Human-in-the-Loop (HITL) for Critical Decisions: AI makes preliminary assessments, humans review/approve every output (e.g., medical image analysis).
• Hybrid/Centaur Model: AI as a specialised assistant for subtasks, humans maintain strategic control (e.g., AI for research, human for strategic report).
• Hybrid/Cyborg Model: Continuous, fluid partnership between human and AI (e.g., AI copilots for coding or writing).

The model you choose depends on the task’s importance, complexity, and how much you want to automate. But the main idea is to use AI to automate and scale processes well. This frees people for tasks that truly need human insight, thought, creativity, and ethical judgment.

JPMorgan Chase achieved something remarkable with AI. They can now parse 12,000 commercial loan agreements in seconds—work that once took 360,000 human hours per year. This shows how AI can eliminate bottlenecks and boost productivity³.

When humans and AI work together thoughtfully, businesses can build systems that are not only powerful but also sustainable and scalable.

Conclusion: Realising AI’s Promise Through Disciplined Strategy

The journey to harness the full potential of Artificial Intelligence is a strategic necessity for any organisation aiming for sustained growth and leadership in the modern economy. But ad-hoc experiments or chasing short-lived tech trends won’t lead to transformative impact. It demands a disciplined approach focused on value, based on proven principles.

By focusing on initiatives that promise transformative, order-of-magnitude gains rather than mere incremental improvements, businesses can generate the drive and ROI needed to make real change. Sticking to the strategic order of Quality first, then Efficiency, and then fostering Innovation ensures that AI solutions are built on a foundation of trust and reliability, leading to lasting operational excellence and, ultimately, groundbreaking new possibilities.

Anchoring AI development in rich, proprietary data gives a unique and defensible competitive edge in an era where AI algorithms are becoming increasingly accessible. At the same time, making smart technology choices—building on stable, proven AI capabilities while designing for future adaptability—lets organisations capture value today without losing readiness for tomorrow’s advances. Crucially, optimising the human role in AI systems, making sure that human ingenuity enhances and augments AI rather than creating bottlenecks, unlocks the true combined potential of human-machine collaboration.

The challenges on this journey—from ensuring data readiness and bridging skill gaps to managing cultural change and navigating ethical questions—are clear. But they can be overcome with clear vision, strong leadership, and a steady commitment to these core principles. Organisations that embrace this strategic framework will move beyond the hype, turning AI from a complex challenge into a powerful, reliable engine for innovation, efficiency, and lasting competitive strength. The future will be shaped by those who not only adopt AI but do so wisely.