A CIO’s Framework for AI-Accelerated Legacy Modernization

One of the largest integrated healthcare companies in the US hired a leading systems integrator to move its data platform off Hadoop. A year and $10 million later, not a single table had migrated. The original plan called for three years and $30 million. Ascendion finished the migration in nine months, with more than 85 percent of the work automated, and cut total cost of ownership by 60 percent.

The technology was never the reason that program stalled. The sequencing was.

That is the pattern behind most failed modernization efforts, and it is the central problem in legacy modernization for CIOs. Target architectures are usually sound. What breaks is the order of operations: leaders scope the entire estate at once, commit to a multi-year plan built on estimates nobody can defend, and stall before any value reaches the business.

This piece lays out a working framework for scoping, prioritizing, and sequencing a modernization program so early moves fund the later ones. The step that changes the math is AI-accelerated reverse engineering, because it removes the largest unknown in the program: the code nobody understands anymore.

Three forces put a modernization program on the rocks, and a CIO will recognize all of them.

The first is that maintenance consumes the budget modernization needs. The GAO reported that roughly $83 billion, about 79 percent of planned federal IT spending for fiscal 2025, went to operating and maintaining existing systems. The federal estate is an extreme case, but the ratio is directionally familiar in any large enterprise. When four of every five dollars is committed before the year starts, modernization competes for a sliver, and any overrun kills it outright.

The second is estimation risk. Programs are scoped against systems that no one fully understands. The original engineers have moved on, documentation is stale or wrong, and the business logic exists only in the code. Estimates built on that foundation are guesses wearing a spreadsheet, which is why overruns tend to appear halfway through rather than at the start.

The third is the all-at-once instinct. Scoping the full estate produces a program too large to fund, too slow to show value, and too fragile to survive a change in leadership or priorities. Value arrives, if at all, years after the spend begins.

The framework below addresses all three in order.

Three forces put a modernization program on the rocks, and a CIO will recognize all of them.

The first is that maintenance consumes the budget modernization needs. The GAO reported that roughly $83 billion, about 79 percent of planned federal IT spending for fiscal 2025, went to operating and maintaining existing systems. The federal estate is an extreme case, but the ratio is directionally familiar in any large enterprise. When four of every five dollars is committed before the year starts, modernization competes for a sliver, and any overrun kills it outright.

The second is estimation risk. Programs are scoped against systems that no one fully understands. The original engineers have moved on, documentation is stale or wrong, and the business logic exists only in the code. Estimates built on that foundation are guesses wearing a spreadsheet, which is why overruns tend to appear halfway through rather than at the start.

The third is the all-at-once instinct. Scoping the full estate produces a program too large to fund, too slow to show value, and too fragile to survive a change in leadership or priorities. Value arrives, if at all, years after the spend begins.

The framework below addresses all three in order.

Three forces put a modernization program on the rocks, and a CIO will recognize all of them.

The first is that maintenance consumes the budget modernization needs. The GAO reported that roughly $83 billion, about 79 percent of planned federal IT spending for fiscal 2025, went to operating and maintaining existing systems. The federal estate is an extreme case, but the ratio is directionally familiar in any large enterprise. When four of every five dollars is committed before the year starts, modernization competes for a sliver, and any overrun kills it outright.

The second is estimation risk. Programs are scoped against systems that no one fully understands. The original engineers have moved on, documentation is stale or wrong, and the business logic exists only in the code. Estimates built on that foundation are guesses wearing a spreadsheet, which is why overruns tend to appear halfway through rather than at the start.

The third is the all-at-once instinct. Scoping the full estate produces a program too large to fund, too slow to show value, and too fragile to survive a change in leadership or priorities. Value arrives, if at all, years after the spend begins.

The framework below addresses all three in order.

Sequencing is where programs live or die, and the rule is simple: front-load the moves that free capital or reduce run cost, so the savings from phase one help pay for phase two.

This does two things at once. It produces visible financial results early, which protects the program through budget cycles and leadership changes. And it converts modernization from a pure cost line into something closer to self-funding, which is a materially easier conversation with a CFO than a multi-year request with value promised at the end.

The delivery method matters as much as the order. Incremental modernization, working through a modularize, modernize, refactor approach, lowers both risk and cash outlay compared with a single large cutover. Systems are decomposed into parts that can be modernized independently, each change is bounded and reversible, and the business keeps running while the estate changes underneath it.

The alternative is the big-bang rewrite, which concentrates every risk in one event and asks the business to hold still for years. Sequenced programs finish. Rewrites stall.

Sequencing also changes what a CIO can promise. A program structured in funded phases can be reported on quarterly, with each phase producing a result the business can see. That is a different governance conversation from a multi-year commitment whose only checkpoint is completion, and it is usually the difference between a program that survives a leadership change and one that does not.

A CIO carries personal accountability for a program that touches production systems, so it is worth being precise about where the risk actually concentrates. It sits in two places: the cutover, and the code no one fully understands.

Both are containable. Reverse engineering addresses the second directly by producing documentation, dependency maps, and test baselines before any code is replaced. Phased sequencing addresses the first by ensuring no single cutover carries the whole program.

The governance layer is what holds it together. In a regulated environment, agentic workflows need human-at-the-wheel checkpoints, audit trails, and validation built into the workflow rather than applied afterward. Agents do the volume work. Engineers make the calls that carry consequences.

The healthcare recovery above shows containment in practice. The program was already failing when Ascendion took it on, having consumed $10 million and a year with nothing migrated. Recovering it in nine months, with more than 85 percent automation and a 60 percent reduction in total cost of ownership, was possible because the approach automated the repeatable migration work while keeping engineering judgment on the decisions that could break production.

Scope, priority, and sequence decide the outcome of a modernization program. Most partners can supply a target architecture. Fewer can execute all three as one operating model, which is where programs actually come apart.

That is the combination Ascendion delivers: AAVA supplies the platform and the agent library, agentic workflows supply the governance, and experienced engineers supply the judgment that keeps AI-generated work accountable in production. This is the Carbon + Silicon model, and in modernization it is the difference between a plan and a delivered system.

The proof is in the sequence working. More than 900,000 lines of 1980s code reverse-engineered in three weeks, and the modernization delivered at a third of the cost of traditional methods. A stalled migration recovered in nine months with a 60 percent reduction in total cost of ownership. These are production outcomes in environments where failure was already on the record.

For a CIO, the practical next step is to see how the framework applies to your own estate, starting with the systems you understand least.

See how AI-accelerated modernization works. Explore Ascendion’s Legacy Modernization Services →