No Free Lunch Podcast – Building ROI on AI projects

In this episode of No Free Lunch my discussion with Sihle Letzolo, the Principal Machine Learning Engineer at Olostel AI, focusses on a practical question facing many businesses today: why do so many AI projects fail to deliver clear return on investment?

Our discussion highlighted a common frustration across industries – Regular large spending on AI without matching gains in cost savings, productivity, or bottom-line impact. Sihle argues that the problem is often not AI itself, but the way companies approach it.

Listen to the Podcast Here:

A major theme of the conversation is the danger of “selling novelty” instead of solving a business problem. Sihle explains that many technologists are drawn to flashy, cutting-edge solutions, but the most effective systems are often the simplest ones if they address the right operational pain point. He stresses that successful AI deployments start with a deep understanding of the business problem first, not with the technology. In his view, the value of any solution should be measured in four ways: whether it makes or saves money, makes or saves time, reduces risk, or improves status or capability within the organisation.

Real World Success Story

Our discussion then turns to a real-world success story: an AI automation project for a manufacturing company in Calgary, Canada. Sihle describes how the client was struggling with financial forecasting, inventory mismatches, manual production scheduling, fragmented procurement, and slow, reactive reporting. Rather than rushing into a generic AI package, his team began with discovery workshops, process mapping, and conversations with leadership, finance, operations, and quality teams.

That groundwork allowed them to identify bottlenecks and design a solution that delivered results within 90 days. This emphasises the value of this practical, metric-driven approach, notably that many technology vendors fail to clearly tie their solutions to business outcomes. Sihle concurs with this analysis and says his company does not sell “AI” for its own sake—it sells solutions to specific problems.

Challenge of Customised Solutions

Sihle explained one of the biggest challenges in his work: nearly every client needs a custom solution, which makes scaling difficult. His team is now working toward a more universal tools that can still adapt to different business needs.

Looking ahead, the conversation explores AI adoption across Africa. Sihle is optimistic about the continent’s long-term potential but says adoption remains slow, especially among small and mid-sized companies. He believes trust is the biggest barrier, along with the tendency for businesses to feel they are being sold technology rather than being helped with real operational problems. The episode closes with a clear takeaway: AI creates value only when it is grounded in business reality, measured by outcomes, and built around trust.