Industry Deep-Dives: AI-Native Modernization in Insurance and Manufacturing

Insurance core systems are, in McKinsey’s description, a “living socio-technical system” that may comprise decades of sparsely documented business rules, batch windows, custom interfaces, and data semantics. That’s a more precise way of saying what insurance CIOs already know: nobody currently at the company fully understands what the policy administration system does anymore, and finding out is most of the work.

That creates a specific financial trap McKinsey calls the “double-bubble” problem: insurers pay to keep running the legacy stack while simultaneously funding the modernization program meant to replace it, and the harder the migration is to complete cleanly, the longer that parallel-running period stretches, extending the timeline for actually decommissioning the old platform. A related McKinsey analysis on P&C core modernization found that underdocumented product logic, actuarial settings, and semantic gaps that surface late in a project are what repeatedly drive schedule slippage and force programs back to the drawing board.

The nuance most business cases miss: rewriting code or configuring the new platform is a small fraction of the actual effort. The disproportionate share of time and cost sits in understanding and configuring rules, data conversion, quality control, reconciliation, operational readiness, and postmigration stabilization. Platform migrations disappoint most often when teams try to recreate legacy complexity on the new core rather than deciding what not to rebuild.

McKinsey’s April 2026 research breaks out productivity improvements by domain step, and the range is wide because the steps themselves are different kinds of work. Discovery and reverse engineering of legacy policy logic shows 20 to 50 percent productivity improvement. Target configuration of the new platform shows 15 to 40 percent. Data mapping, conversion, and quality work shows 20 to 60 percent. Testing, reconciliation, and defect cycle compression, the step McKinsey identifies as carrying the most schedule risk, shows the widest range: 15 to 90 percent. Cutover and operational readiness shows 10 to 40 percent, and program management and governance shows 25 to 50 percent.

The specific unlock is language and documentation. Insurance core systems often run on programming languages few current engineers can read fluently. Agents can decode archaic code, reverse-engineer the embedded logic, and translate it into plain English, work McKinsey found an agent can complete in days that would otherwise take a subject matter expert months or years.

The economics compound from there. Once a set of core agents is built and governed for one modernization effort, the marginal cost of reusing them on the next product line, billing integration, or satellite system drops sharply. That reframes modernization from a single high-stakes migration into a coordinated portfolio of opportunities across the technology estate, the same reuse dynamic that makes sequencing decisions so consequential across any modernization program. McKinsey is explicit that this requires governance built for agentic execution from the start: human-in-the-loop approval at stage gates, full traceability from requirement to configuration to test evidence, and new roles it calls “product definers” and “product builders” replacing purely transactional ones.

Manufacturing’s core system rarely gets the dramatic “legacy mainframe” framing insurance and banking systems get, but it carries the same economics. Enterprise resource planning traces its roots directly to manufacturing: McKinsey’s own history of the category starts with 1980s-era material requirements planning and manufacturing resource planning systems, decades before ERP became the generic enterprise term it is today. Most manufacturers are still running some descendant of that system at the center of production, inventory, and supply chain operations.

McKinsey’s own ERP cost benchmarks put the price of a large enterprise ERP migration between $100 million and $1 billion, with a payback period of four to five years when the migration is used as a genuine business transformation catalyst rather than a lift-and-shift. And the honest context worth sitting with: across large technology programs generally, McKinsey research conducted with Oxford University found that only 25 to 35 percent hit their targeted EBITDA and cash-flow impact, while 65 to 80 percent exceed their planned budget or timeline. Manufacturing’s core system carries all the same risk mainframe and insurance core modernizations do. It just doesn’t get talked about the same way.

McKinsey’s research finds AI reducing ERP program cost and duration by at least two times. Design and build work that normally takes six to nine months compresses to two to three months. Testing effort drops by roughly 80 percent. Training preparation, traditionally one of the most manual parts of any ERP rollout, drops by roughly 90 percent. Early adopters of AI-integrated ERP are already reporting EBIT improvements of 5 percent or more.

The constraint that remains is not technical. McKinsey’s research states it plainly: change management becomes the major constraint in ERP transformation roadmaps going forward, since the people who use the system still have to be brought along regardless of how fast the technical migration runs. That mirrors the same point Gartner’s own analysts make about mainframe modernization: the unsolved problem is organizational, not technical.

The opportunity extends past the core ERP migration itself. Deloitte’s 2026 Manufacturing Industry Outlook found that 80 percent of manufacturing executives plan to invest at least 20 percent of their improvement budgets in smart manufacturing initiatives, with agentic AI identified as a way to capture institutional knowledge from a retiring workforce, generate shift handover reports and work instructions autonomously, and identify alternative suppliers in response to disruption, use cases that extend the same reverse-engineering and reuse economics from the core system out to the plant floor.

The same two principles that hold across this entire series apply directly to insurance and manufacturing. Reuse economics matter more than raw code-generation speed: once a governed agent stack exists, whether it’s built for insurance policy logic or manufacturing process logic, the marginal cost of the next system in the queue falls sharply, which is why sequencing a modernization portfolio matters as much as executing any single migration well. And reverse engineering has to come before forward engineering: both insurance core systems and manufacturing ERP are described by McKinsey’s own research as sparsely documented, decades-old logic that has to be understood before it can be safely rebuilt, the same principle that separates working mainframe modernization programs from failed ones.