Every week, I see another announcement: a company launching an “AI-powered” solution, a government unveiling an “AI transformation strategy,” a consultancy promising that machine learning will revolutionise operations. And every week, I watch many of these initiatives quietly fail — not because the AI wasn’t sophisticated enough, but because the foundations underneath it were never solid.
Here’s what I’ve learned from building and operating digital systems across Nigerian state governments, UK telecommunications infrastructure, event ticketing platforms, and healthcare technology: AI is not a transformation strategy. It’s a layer. And layers only work when what’s beneath them is sound.
The conversation about AI has become disconnected from the reality of how complex systems actually function. We talk about algorithms and models while ignoring data governance. We celebrate automation while neglecting operating models. We deploy cutting-edge technology on top of decade-old processes and wonder why the results disappoint.
After 18 years building technology that has to work — not in demos, but in production, at scale, for years — I’ve developed a different perspective on what “AI transformation” actually requires.
Lessons from Long-Living Systems: What Operating RomFlex Taught Me
In 2014, I founded RomFlex, a civic revenue infrastructure platform for Nigerian state governments. The system handles tax collection, payment processing, and receipt verification across six states. It’s been in continuous operation for over a decade.
That last sentence matters more than any technical specification I could share. In an industry obsessed with launches, we rarely talk about what it takes to keep systems running — and evolving — for ten years.
Here’s what RomFlex taught me about building systems that last:
Infrastructure before intelligence. Before we could think about analytics or optimisation, we had to solve fundamental problems. How do you process payments when internet connectivity is intermittent? How do you prevent receipt fraud in an environment where paper receipts were routinely forged or reused? How do you build trust with government officials who have seen technology projects fail repeatedly?
We built real-time bank API integration so that no receipt could be generated without confirmed payment. We designed QR-code verification so any citizen could validate a receipt’s authenticity. We created offline-capable interfaces for field agents working in areas with poor connectivity. These weren’t exciting innovations — they were necessary foundations.
Data governance is not optional. When you’re handling government revenue, every transaction must be auditable, every access must be logged, and every data point must be traceable. We built comprehensive audit trails not because regulations demanded it, but because the system’s credibility depended on it.
Treating data governance as core infrastructure rather than compliance overhead is something I see missing in many AI initiatives. You cannot build trustworthy AI on untrustworthy data.
Systems must evolve without breaking. Over ten years, RomFlex has adapted to new states, new tax categories, new banking integrations, and new government requirements. The original architecture had to accommodate changes we couldn’t anticipate. This required building for modularity, maintaining clean interfaces between components, and resisting the temptation to take shortcuts that would create technical debt.
Enterprise Reality: Applying These Principles at Scale
These principles have shaped every system I’ve built since — whether in emerging markets or UK enterprise.
UK Telecommunications Infrastructure
I served as a Senior Digital Transformation Project Manager on Virgin Media O2’s NetEx programme — a £4.5 billion initiative to expand fibre-to-the-premises infrastructure to seven million UK homes. The scale was different from Nigerian state government systems, but the principles were remarkably similar.
Large-scale infrastructure programmes fail for predictable reasons: unclear data ownership, disconnected systems, processes that evolved organically rather than by design, and the assumption that new technology will somehow fix structural problems. The questions I brought from emerging markets proved directly relevant: What happens when this component fails? How do we maintain data integrity across system boundaries? Who owns this process, and do they have the tools to manage it?
Building Lybertine: Complex Systems for the Creator Economy
Currently, as CTO of Lybertine, I’m applying these same principles to a very different domain: event ticketing and social commerce for the creator economy. The platform connects event organisers, artists, and fans through integrated ticketing, real-time social features, and commerce capabilities.
The technical complexity is substantial — over 150,000 lines of code across a microservices architecture, handling everything from real-time @mention systems and notifications to comprehensive finance dashboards and payment reconciliation. We process ticket sales, manage event logistics, handle creator payouts, and facilitate social interactions, all within a single integrated platform.
What makes Lybertine work isn’t any single clever feature — it’s the underlying architecture. Clean service boundaries. Consistent data models. Robust event-driven communication between components. The ability to scale individual services independently. These foundational decisions, informed by years of building systems that have survived real-world conditions, determine whether a complex platform thrives or collapses under its own weight.
The lesson transfers across domains: whether you’re building government revenue systems, telecommunications infrastructure, or creator economy platforms, the fundamentals don’t change. Get the architecture right. Treat data as a first-class concern. Design for evolution.
AI as a Layer, Not a Shortcut
I’ve built AI into systems. My healthcare platform, Rapid Capsule, uses dual AI integration — combining a medical diagnostic API with a large language model for patient communication. This AI adds genuine value: it helps patients understand symptoms, translates complex medical information into plain language, and assists with prescription verification.
The AI only works because of what sits beneath it. Structured data models. Clear user flows. Robust authentication. Comprehensive logging. The AI is a capability layer on top of solid infrastructure — not a substitute for it.
This is where I see organisations getting AI wrong. They treat AI as a shortcut: a way to leapfrog the hard work of fixing data quality, modernising processes, and establishing governance. They layer machine learning on top of fragmented data and expect insights. They deploy automation on top of broken processes and expect efficiency.
What they get instead is AI that amplifies existing problems. Garbage in, garbage out — but faster and at scale.
The organisations that will succeed with AI are those willing to do the unglamorous work first: consolidating data sources, establishing clear ownership, documenting processes, building the infrastructure that makes AI useful rather than just impressive.
Why Experience in Emerging Markets Matters
There’s a particular value in building technology under constraints. When resources are limited, when infrastructure is unreliable, when you can’t assume users have the latest devices or stable internet, you learn to build differently. You learn to design for the worst case, not the best case. You learn to make systems resilient rather than just functional. You learn to solve problems with architecture.
This constraint-driven thinking is increasingly relevant in enterprise contexts. As organisations face pressure to do more with less, grapple with technical debt accumulated over decades, and try to modernise systems while maintaining operations, the skills developed in resource-constrained environments become valuable.
I’ve sat in rooms where the assumption is that emerging market experience is somehow less rigorous than developed market experience. The opposite is often true. Building systems that work despite infrastructure limitations requires more rigour, not less. It forces you to think carefully about failure modes, to design for degradation, to build systems that bend rather than break.
When I approach any technical challenge, I bring this perspective of experience. What are the constraints we’re not acknowledging? What assumptions are we making about reliability that may not hold? How do we build something that will still work when conditions are less than ideal?
What Leaders Get Wrong About “AI Transformation”
The biggest mistake I see leaders make is treating AI as a destination rather than a capability. They announce AI strategies, fund AI initiatives, hire AI teams — all while neglecting the foundational work that determines whether AI will actually deliver value.
Here’s what actually works:
Start with the problem, not the technology. The question isn’t “how do we use AI?” It’s “what problem are we solving, and is AI the right tool?” Sometimes it is. Often, better data management or process improvement will deliver more value.
Invest in foundations before features. Data quality, governance, and integration are not exciting. They don’t make good press releases. But they determine whether your AI initiatives succeed or fail. Budget accordingly.
Design for evolution. The AI landscape is changing rapidly. The systems you build today need to accommodate capabilities that don’t exist yet. This requires architectural thinking — clean interfaces, modular components, clear data contracts.
Learn from constraints. The most innovative solutions often come from environments with the most limitations. Look beyond the usual sources of “best practice” to organisations that have had to solve hard problems with limited resources.
After nearly two decades of building systems that have to work in production, across diverse environments, through changing requirements, I’ve learned that sustainable digital transformation is less about the technology you deploy and more about the foundations you build.
AI is powerful. But power without infrastructure is just potential. The organisations that will lead in the AI era are those that understand this — and do the hard work accordingly.