When Seagate’s IT team decided to replace the ITSM platform that had run its global IT operations for more than a decade, it had three months to do so.
It was a deadline imposed by a strict contract expiration. It will take three months to migrate 30,000 employees across Seagate’s global storage and infrastructure operations to an entirely new system. In such situations, most organizations do the obvious thing: unpack the existing configuration, drop it into the new environment, and reconcile the mess later. That’s the safer path. This also almost guarantees that the AI functionality the team was hoping for won’t be fully functional.
The team chose a more difficult path. They rebuilt it from scratch. We restructured our service catalog, established consistent SLAs across regions, and rewrote our category hierarchy so tickets can be automatically routed without agents having to guess where they belong. They did this because they didn’t want to intentionally bring forward the traditional process. One year later, the AI agent the team deployed on top of that foundation now avoids about a third of incoming tickets. First contact resolution rates are currently 27% higher than industry standards.
This decision—to rebuild rather than duplicate—is the real story that separates companies that embrace AI from those that don’t. And it has little to do with the model they’re running.
complex tax
An increasing proportion of enterprises’ AI investments are being consumed before the value reaches the business. MIT found that 95% of generative AI pilots fail to scale out to production. A September 2025 study by Boston Consulting Group found that 60% of companies are not creating real value from AI. This number was worse than the previous year, despite better tools and more experience. Freshworks’ upcoming Cost of Complexity study points out more reasons why. A quarter of AI budgets are spent on integration work, data cleanup, and the effort to force some kind of consistent conversation into systems that weren’t designed to talk to each other.
This pattern is consistent across industries. Programs may stop, reset, or silently disconnect. It’s not because the model doesn’t work. That’s because the underlying operating environment wasn’t ready to support them.
This is disproportionately true for a certain type of company, the type of company I’ve come to call the Agile Enterprise. These companies have between 500,000 and 20,000 employees, run lean IT teams, and have far less margin for failure on technology bets than companies with $500 million transformation budgets. If a company in this position loses a quarter of its AI spend to integration overhead, that’s not a rounding error. It’s a canceled effort.
What companies moving forward have in common
But smaller, more nimble groups of companies are producing very different results. They aren’t spending any more. They’re starting somewhere else.
Seagate is one version of this. New Balance is another. Nike operates with 80,000 employees. New Balance runs on 9,000. And New Balance isn’t gaining market share by getting bigger, but by getting faster and sharper. The company didn’t achieve its status by doing anything fancy. The company achieved this award by consolidating its fragmented IT stack into one platform with a single source of truth, freeing its teams from maintenance tasks and rewiring the way it runs its business.
This is the foundational work that will pay off well before AI arrives, and is the very foundation for AI to function when it arrives. Companies such as Nucor and Steel Dynamics, two of the top four U.S. steel producers, exhibit the same pattern on an industrial scale. Decades of operational discipline have created an operational environment that AI can actually optimize.
In all of this, AI is working where the operating model is ready. It’s not perfect. Ready. That means data was integrated, workflows were defined, and the system was able to pass information without manual intervention, giving us the clear, measurable results we wanted AI to improve upon.
How to start when starting from a messed up state
Most companies aren’t in the situation Seagate is in today. Most are somewhere in between. Legacy platforms that have been in place for too long, data scattered across inconsistent systems, and an IT team that has spent more time over the past five years keeping systems running than rebuilding them. The question is not whether AI will work in the environment, but where to start.
Robert Lyons, CTO of Katz Media Group, has one of the clearest answers I’ve heard. Katz is an 800-person division within a 10,000-employee parent company, and it’s a nimble company that doesn’t have the luxury of chasing every AI initiative that sounds appealing. Lyons maps every potential AI project onto what he calls a value-effort matrix. So one axis is ease of implementation and the other axis is business value. He starts with areas of high value and low effort and works outwards from there. “Don’t start with the worst problems first,” he said recently. “You can’t deliver value. Focus on ease of implementation for immediate impact.”
Before deploying AI tools, Lyons’ team did two things that most organizations skip. They cleaned and labeled the data. Because feeding messy data into AI and wondering why the results are disappointing is the most common failure mode in businesses today. We also conducted an introductory AI webinar for all employees, conducted by a neutral third-party research firm rather than by the IT department. “IT isn’t barking at you,” Lyons says. “If a neutral party socializes this, you end up in a different direction.”
This methodical, disciplined, and results-based approach distinguishes companies that have adopted AI from those that are still talking about it.
Where the benefits actually exist
All the agile companies I’ve seen succeed with AI consistently exhibit three operational characteristics. There is no mention of which model the company has selected.
- They reduced fragmentation before adding intelligence. Not by consolidating everything into one super platform (although that’s a different and usually more expensive story), but by allowing critical systems to exchange information without manual handovers. This is not a glamorous job. This doesn’t make for an exciting board presentation. But this is the single most leveraged action midsize companies can take before writing a check to an AI tool.
- They applied AI to the parts that improve execution, not the parts that create complexity. The best use case in an agile company is not a moonshot. It’s about speeding up your workflow. Faster ticket resolution, smarter demand planning, automated quality inspection, and predictive maintenance scheduling. Use cases where inputs are structured, outputs are measurable, and humans stay in the loop.
- They treated AI adoption as an operational discipline rather than a technology project. Leading companies didn’t hand AI over to innovation teams and wait for reports. They built it into the daily operations of the teams closest to the customer, the production line, or the revenue cycle, and measured it the same way they measure other operational investments, by whether it moves important numbers.
agile enterprise moment
AI is often discussed as if it is a capability that only the largest and best-resourced companies can deploy at scale. That framework is wrong and risks becoming a self-fulfilling prophecy for agile companies that believe in it.
Agile companies make up the majority of businesses around the world. If the productivity promise of AI is real, it will be proven or disproven in these organizations, not the multi-trillion dollar companies running bespoke foundational models.
Part of a CEO’s job now is to live in the future and understand where technology is going. But the other part, the more difficult part, is bringing that vision back to the present and delivering something that solves real business problems today. The companies that I see doing both of these things don’t have very big budgets. They are the ones who, somewhere along the way, made an intentional choice to stop dragging on the past and start building for what comes next.
This is an option that agile companies can take now.
The opinions expressed in Fortune.com commentary articles are solely those of the author and do not necessarily reflect the author’s opinions or beliefs. luck.
