While executives across America desperately hunt for COBOL programmers and AI tools struggle with legacy code, Royal Caribbean is quietly celebrating AI wins that save tens of millions annually. The secret? They solved their COBOL problem before anyone knew it would become a crisis.
In our last piece, we warned that most AI Centers of Excellence stall because they launch before the foundations are ready.
Royal Caribbean flips that script.
Long before their CoE delivered operational savings or customer wins, they tackled the hard part: Extracting business logic from legacy COBOL systems. This sequencing wasn’t flashy, but it was critical. It’s proof that momentum in AI doesn’t come from more pilots, it comes from building on solid ground.
Royal Caribbean didn’t skip steps and that’s why we believe their CoE scaled where others stall. The stakes are high in this sector for a cruisline representing 20% of the global cruisers and serving close to 5 million guests worldwide annually.
The accidental genius
Back in 2014, Royal Caribbean quietly kicked off a major legacy overhaul. They dissected more than 7 million lines of RPG and COBOL, mostly buried in their monolithic reservation, inventory, and accounting systems.
They finished this unglamorous work around the time ChatGPT was still a research project. To our team, it’s a case study in strategic archaeology: extracting business rules, mapping code to data models and re-architecting systems into modular components. While nobody at the time called it AI readiness, that’s exactly what it became.
Royal Caribbean’s digital transformation has two powerful tracks.
The first, led by their CIO organization and delivered through partners, built the guest-facing foundation: a mobile app rolled out to 60% of ships, used by 95% of guests, and driving an 89% boost in satisfaction.
Royal Caribbean’s modernization wasn’t just a UI facelift, rather an architectural overhaul. They introduced reactive microservices between COBOL-based legacy systems and a ship-wide mobile app, ensuring fault tolerance and real-time responsiveness. With Apache Kafka, Spark, Flink, and MinIO, they built multi-modal data synchronization (event, file, DB) across ships and shore, a feat made harder by intermittent satellite bandwidth.
To operationalize this, they enforced full CI/CD pipelines, container orchestration with DC/OS from Mesosphere, and live telemetry via Prometheus, Grafana, and Splunk. The result was a self-healing, edge-resilient analytics architecture that made true “AI at sea” possible.
The second track is Royal’s AI Center of Excellence. This team focuses on the invisible layer of intelligence, applying AI to reduce food waste, optimize fuel, and standardize data across business lines. The CoE also governs the company’s data lake, sets internal standards, and helps business units adopt AI without relying solely on centralized teams.
Together, these two tracks — strong digital plumbing and applied intelligence — form the sequence many companies miss.
Royal Caribbean didn’t start with AI but they earned the right to scale it.
But here’s what doesn’t get mentioned often in conference talks: none of this would be possible without that earlier COBOL modernization.
What the X-Analysis project gave them was clean, structured, accessible data logic. Without it, the AI models—especially those built on real-time pipelines, operational forecasting, or cross-department integrations—would have been built on brittle, undocumented legacy code.
The sequencing wasn’t accidental. It was essential.
The crisis everyone else faces
While Royal Caribbean was quietly doing homework, the rest of the world kept procrastinating. Now they’re all scrambling:
- Morgan Stanley just spent millions building custom AI tools to translate old code because regular AI can’t read their ancient systems. They’re basically paying millions to fix what they could have fixed years ago.
- State governments embarrassed themselves during COVID when unemployment websites crashed under demand. Remember those governors begging retired programmers to come back? That was because their systems were built on 40-year-old code nobody understood anymore.
- IBM—the company that helped create COBOL—is now desperately running crash courses to teach it again because all the experts are retiring.
Here‘s the crazy part
COBOL still runs most of the financial world. Every time you use an ATM, you’re probably touching COBOL code. It processes $3 trillion in transactions every single day. But here’s the kicker: AI tools— which everyone thinks can solve everything—are terrible at understanding COBOL.
Why? Because most COBOL code is locked away in corporate systems that AI has never seen. It’s like asking someone to translate a language they’ve never heard. The business rules are buried in custom code that only a handful of aging programmers can decipher.
Most companies jump to AI before fixing their data foundations. At OpenNova, only 5% of the firms we see are AI-ready. That’s why our talent² strategy often begins with the unglamorous roles—data architects, governance leads, and systems analysts who can extract value from what’s already there. Whether that’s COBOL-based reservation logic or disconnected SQL reports, we build bridges first—models second.
The strategic silence
Interviews from team members reveal telling patterns around the AI CoE. They may talk about “traditional enterprises” and “heterogeneous systems”—corporate euphemisms for “we had a COBOL problem.” They emphasize the need to “deeply understand the business” and “make data more accessible”—code for “we had to extract logic from ancient systems.”
This isn’t accidental omission. Saying “our AI works because we modernized COBOL first” doesn’t make for inspiring transformation stories. But that’s exactly what happened.
The warning
Royal Caribbean’s story isn’t just about one company getting lucky. It’s about what happens when you do your homework versus when you don’t as reflected by comments made by the overhaul team back in 2014:
“…altering its 20-year-old applications and database development processes in preparation for the next 20 years of business goals. It is currently in the second phase of a program designed to migrate selected, monolithic, RPG applications to Java-based Web services that will continue to run on IBM Power servers and the IBM i operating system”
Right now, there are two types of companies: those who cleaned up their old systems early, and those who are now panicking because their ancient code can’t keep up.
The hard truth to accept is that you can’t use AI to magically fix decades of neglect. Royal Caribbean proves this by showing what’s possible when you tackle the boring stuff first.
Every company racing to build AI teams should ask themselves: are we building on solid ground or quicksand? Because the businesses that ignored their old systems are about to learn that even the smartest AI can’t work miracles on broken foundations.
Royal Caribbean didn’t set out to win the AI revolution. They just refused to lose the old-code cleanup game. Sometimes the smartest strategy is fixing what’s under the hood before racing toward the next shiny thing.
What this means for enterprise leaders
By doing the unglamorous work early and sequencing their transformation in the right order the results followed.
At OpenNova, that’s the lesson we see too many companies skip.
- We don’t lead with “AI wizards.” We sequence talent in layers, starting where the real work begins.
- We understand that data readiness is everything. In some sectors, that means unearthing forgotten COBOL.
- In others, it means creating data that never existed. (See: The future runs on fake data—and so do its jobs)
- We operate at the intersection of legacy, logic and leadership.
And while we don’t specialize in legacy coding roles like COBOL, we support modernization efforts by deploying governance experts, data architects, and systems thinkers—exactly the people needed to unlock value from legacy environments.
Because sometimes the smartest move isn’t building faster, it’s making sure you’re building on solid ground.
| A note to AI CoE leaders disrupting their industries: We see the strategic brilliance in your transformation sequence, whether intentional or not. The companies getting AI right often share this pattern: foundational work first, intelligence layer second. If you are involved in similar projects, we would love to hear your perspectives. |
