U.S. air traffic control still runs on COBOL and magnetic tape. What aviation's 40-year legacy infrastructure teaches every enterprise about migration as a data-lineage problem.
You cannot move what you cannot trace. Bronze layer thinking — preserve raw, immutable source records before you touch anything — is the only safe path through a legacy migration.
In January 2023, a corrupted file in a legacy FAA safety system called NOTAM — Notice to Air Missions — triggered the first nationwide ground stop of U.S. commercial aviation since September 11, 2001. Over 11,000 flights were delayed. Hundreds were cancelled. The cause was a single bad data record in a 30-year-old system that the agency had been trying to replace for over a decade.
The response from technology commentators was swift and predictable: why is the FAA still running on COBOL? Why hasn't this been moved to the cloud? The question reveals a fundamental misunderstanding of what makes legacy migration hard — and it is the same misunderstanding that causes enterprise data migrations to fail at a 70% rate.
The FAA's technology is old. But the technology is not the bottleneck. The bottleneck is data lineage — specifically, the absence of it. Over four decades of continuous operation, the air traffic control system has accumulated data that exists in formats no longer documented, was transformed by processes no longer understood, and encodes business rules that exist only in the institutional memory of engineers who retired years ago.
Moving this data to a modern cloud platform does not solve any of these problems. It relocates them. You end up with undocumented, untraceable data in a modern warehouse instead of an old mainframe. The failure mode has been modernized. The failure itself has not.
Every legacy migration must begin with a Bronze layer operation: create an immutable, complete, exact replica of the source data before touching anything. This is not a backup. A backup is designed to restore the system to a prior state. A Bronze layer is designed to preserve the raw truth of what existed so that every subsequent transformation can be traced back to it.
The Bronze layer operation answers a question that most migration projects never ask: what exactly do we have? Not what do we think we have. Not what the data dictionary says we have. What do we actually have, field by field, record by record, with every anomaly documented and every inconsistency catalogued.
This step takes time. In a large enterprise legacy migration, the Bronze layer documentation phase can take months. Organizations routinely skip it or abbreviate it because it feels like analysis paralysis. It is not. It is the only way to know what you are migrating, which is a prerequisite for migrating it correctly.
While the FAA's infrastructure has struggled, individual airlines — Delta, American, United — have made meaningful progress on legacy data modernization, with instructive lessons.
The airlines that succeeded treated their reservation and operations data migration as a data product problem, not a system replacement problem.
Instead of attempting a wholesale cutover from legacy to modern systems, they built parallel Gold tables — curated, validated, purpose-built datasets that served specific AI and analytics applications — while leaving the legacy transactional systems in place. The legacy system remained the system of record for operations. The Gold tables became the system of intelligence for AI. Over time, as confidence in the Gold tables grew and the legacy systems were gradually documented, the migration proceeded incrementally rather than catastrophically.
If your organization is facing a legacy migration, start with these three questions before evaluating any technology: What data do we actually have, and is it documented? What business rules exist only in application code, not in data? What is the migration sequence that allows us to validate outputs at every step before proceeding? Answer these questions first. Then choose your cloud provider.
The FAA is an extreme case, but the pattern it represents is present in every organization that has been operating for more than fifteen years. Undocumented data. Untraceable transformations. Business rules embedded in systems that were never designed to expose them. These are not technology problems. They are data governance problems that accumulated over years of treating data as a byproduct of operations rather than as a strategic asset.
The organizations that will win at AI over the next decade are not the ones that move fastest to adopt the latest model. They are the ones that invest now in understanding what data they have, documenting how it was created, and building the lineage infrastructure that makes it trustworthy. That investment is not glamorous. It does not make for compelling demos. But it is the only foundation on which AI that actually works in production can be built.
Follow OBT on LinkedIn