The number that gets a CFO's attention
Mead Johnson Nutrition originally projected its SAP ERP to S/4HANA transition would take 17 to 18 months, a timeline consistent with how most large enterprises plan these migrations. Rama Donepudi, the company's SVP and Global CIO, finished the project in 11 months instead, and the company estimates the compression saved between 5 and 6 million dollars. That reduction amounts to roughly a third of the original schedule on one of the most notoriously slow categories of enterprise IT work, a far larger cut than CIOs typically expect from any single modernization lever.
Donepudi did not achieve that by using one AI tool for everything. He describes AI as one tool in the toolbox, applying generative AI coding copilots to fix custom programs, UiPath AI bots to validate data as it moved between systems, and separate AI tools to produce training content and employee communications. The variety matters: different parts of an ERP migration bottleneck in different ways, and Donepudi matched specific AI capabilities to specific bottlenecks rather than expecting one platform to compress the whole program.
Why ERP migrations are the perfect target for this
ERP transitions bog down for a predictable reason: years of custom code and undocumented business rules accumulate inside the old system, and migrating to a new platform requires someone to manually decipher what that custom code actually does before deciding what to keep, rewrite, or retire. That deciphering work has historically consumed a large share of migration timelines, and it is exactly the kind of pattern-matching, code-reading task that current AI coding tools handle faster than human engineers working through legacy code line by line.
Ensono, a managed services provider, reports using AI for code discovery and legacy code cleanup on behalf of its customers, confirming this is not a one-company anecdote. Avaya makes the same argument on the contact center side rather than ERP specifically, saying AI tools built on frontier models can compress rebuild timelines from months down to days by automating the manual work of deciphering years of accumulated workflows and business rules, the same underlying bottleneck Mead Johnson faced in its SAP environment.
The honest limits Gartner is pointing at
Gartner analyst Arun Chandrasekaran sees real potential in three specific areas: gathering requirements, mapping legacy code, and code quality assurance. Those are precisely the categories where Mead Johnson and Ensono report their gains, which lends some independent credibility to the vendor and customer claims rather than taking them purely on faith. But Chandrasekaran is equally direct about where the technology is not ready: models are getting better at these tasks, he says, but they are still often hindered by hallucination rates high enough to require human verification at every step.
That caveat is the difference between AI accelerating a migration and AI running one unsupervised. Donepudi's own description of the approach, calling it risky and explicitly taking full ownership of that risk, reflects the same reality from the practitioner side. The gains Mead Johnson realized came from AI handling specific, bounded tasks faster, with humans still validating outcomes, not from removing human judgment from the process entirely. Any CIO repeating this playbook should budget explicit review time at each handoff rather than assuming a faster first pass also means a safer one.
Avaya's limitless framing versus the governance Chandrasekaran wants first
Avaya's Vijay Thangella describes the potential here in near-unbounded terms, saying he does not want to call it limitless but that it essentially is. That kind of framing is common from vendors selling AI-accelerated modernization tools, and it is worth holding against Chandrasekaran's more measured view that many CIOs still need stronger AI governance and better data infrastructure in place before they can responsibly capture gains at this scale. Vendors are incentivized to describe the ceiling, while analysts and practitioners who have actually run the project tend to describe the floor, and a CIO evaluating either pitch should ask explicitly which one they are hearing.
The practical reconciliation is that both things are true simultaneously. The technical potential for timeline compression is real and demonstrated, Mead Johnson's 5 to 6 million dollar savings is evidence of that, but realizing it safely depends on governance maturity that not every organization has built yet. A CIO without solid data infrastructure and clear AI oversight structures should expect to capture less of this upside, and capture it less safely, than Mead Johnson did. Treat Avaya's framing as the ceiling worth aiming for and Chandrasekaran's checklist as the floor a program needs before it starts climbing toward it.
Why midsize companies are moving first
Chandrasekaran notes that midsize companies show particularly high interest in AI-accelerated modernization, largely because they tolerate risk differently than large multinationals do. A midsize company has less legacy complexity to navigate, fewer regulatory jurisdictions to satisfy simultaneously, and often a more concentrated decision-making structure that can approve a riskier approach faster than a global enterprise's governance layers typically allow. That combination lets a midsize CIO move from pilot to production on an AI-accelerated migration in a fraction of the approval cycles a comparable global enterprise would need to clear first.
Donepudi's own observation about his peers reinforces this pattern from the other direction: some large-company CIOs hesitate and continue relying heavily on big consulting firms for program and change management rather than leaning into AI-accelerated approaches themselves. That caution can be the right call for a given organization's risk profile, yet it also means the timeline and cost gains Mead Johnson achieved will likely reach midsize companies faster than they reach the largest, most risk-averse enterprises.
What this means for the next migration on your roadmap
If an ERP migration sits on your roadmap in the next 12 to 18 months, Mead Johnson's experience is a useful benchmark to pressure-test your own vendor and integrator timeline estimates against. Ask specifically where AI coding copilots and automated validation tools could compress your legacy code discovery and data validation phases, since those are the exact categories where real, measured gains already exist, rather than asking vendors for a vague AI acceleration promise with no specific mechanism attached.
Pair that ambition with Chandrasekaran's caution: build or confirm your AI governance and data infrastructure foundation before leaning on AI for a program this consequential, and plan for human validation at every handoff rather than assuming the tools will catch their own errors. Mead Johnson's 5 to 6 million dollar saving came from disciplined application of multiple specific tools with a CIO who explicitly owned the risk, not from treating AI as a single button that compresses an ERP migration on its own.
