A proven joint venture model generated billions across Europe, but when it reached the US market, it failed. The technology worked and the business case was fully established, yet the operating model simply wasn’t portable.
This article draws on insights from Elaine Barsoom, who led innovation partnerships at both American Express and Nike. It explores how organizational behavior and fragmented operating models undermine AI adoption long before the technology itself fails.
The companies seeing meaningful returns from AI recognized early that the technology was never going to be the hardest part. Redesigning how people work around it is.
At American Express, Elaine Barsoom watched a joint venture model that had generated billions across Europe arrive in the US market with every advantage in place. The brand relationships were established, the business case was proven, and the commercial logic was clear, yet it didn’t translate.
The US market was more sales-driven, the commercial dynamics were entirely different, and a playbook that had worked at enormous scale in one region required a completely different approach in another.
The assumption that proven meant portable became an expensive lesson.
It is also a lesson that keeps repeating across enterprise AI. According to BCG’s 2025 research, the true value of AI is being captured by a smaller group of organizations that go beyond deployment to fully redesign how work gets done, while roughly 60 percent of AI transformation efforts are delivering limited or no material value despite sustained investment. For most organizations, the gap between adopting AI and seeing any return on it continues to widen.
The adoption problem that investment cannot fix
As Nike’s Global Head of Tech Innovation Partnerships, Elaine Barsoom built and scaled the company’s first AI Center of Excellence across product, retail, marketing and HR. When her team deployed GitHub Copilot across the engineering function, the setup was textbook, with a proven use case and sound technology behind a fully approved business case. Two months in, almost nobody was using it.
“Innovation doesn’t stall because the teams aren’t capable,” Elaine said on a recent episode of the Phrase podcast In Other Words. “It stalls when we treat the launch as the destination.”
The tool worked as promised, but nobody had rethought how engineers actually operated before asking them to adopt something new. And without best practices or peer support there was no reason for anyone to move away from what was already familiar. The fix was organizational rather than technical, combining a change management program with a champions network and structured peer-to-peer sharing. Within months, Slack channels started filling up with use cases nobody had been asked to share, and the tool went from something that had been deployed to something people genuinely used.
Elaine’s measure of success was not the adoption dashboard. It was the behavior change underneath it. “Value doesn’t show up in the dashboards,” she said. “It shows up in behavior.” When engineers stopped asking how to use the tool and started asking whether it could solve specific problems, the technology had finally become operational rather than experimental. If AI is not changing how work gets done, as Elaine puts it, “it might be running, but it isn’t working.”
BCG’s global survey found that when leaders actively support AI adoption, the share of employees who feel positive about the technology jumps from 15 percent to 55 percent, but only about a quarter of frontline employees say they receive that kind of support. The confidence gap between the leadership team and the people who have to use the technology every day is where most AI programs lose their momentum. More investment won’t fix what is fundamentally a change management failure.
Customers see the fragmentation leadership ignores
The Amex experience illustrates a broader pattern that becomes even harder to manage when organizations operate globally. Every market wants to move fast and every function wants to build its own approach. Without clear decisions from leadership on what is standardized and what belongs to the local market, the brand starts to fragment from the inside out.
“Fragmentation happens when no one defines which is which. The best leaders define what’s the brand standard, what’s the trust and governance boundary, and then they give freedom on everything else. You always find out when the customer does.”
Customers never see the vendor architecture or the integration roadmap underneath. As Elaine puts it, “customers don’t experience your ecosystem, they experience the moment.”
When the joins between systems, teams and markets start to show, the cost is not just operational but commercial, eroding the trust and consistency that global brands depend on. The best ecosystems disappear entirely in use, and that should be the standard rather than the aspiration.
For organizations managing content and communication across multiple markets and languages, this is where operational infrastructure becomes a strategic issue. The systems underneath the customer experience often determine whether a global brand scales coherently or fragments market by market.
One global fashion retailer using Phrase was able to manage up to one million words of translation per week while expanding into 14 new regions, with content published within minutes rather than days. The advantage did not come from producing more content. It came from building a coordinated operating model capable of scaling consistently across markets while maintaining a coherent customer experience.
The long-term cost leaders fail to plan for
When organizations encounter these scaling challenges, the instinct is often to build internally through custom AI tools, proprietary infrastructure, or bespoke workflows designed around the business.
The logic feels reasonable since building creates a sense of control. What leadership teams often underestimate is how ownership escalates over time.
“When you build, you own every decision the software touches,” Elaine says.
“The governance, the compliance, the workflows that grow around it, and the people maintaining it years later without the context of why it was built that way.”
Most leaders plan carefully for what they are building but almost none plan for what maintaining it will cost them five years from now, when the original architect has moved on and the business itself has changed.
The same logic applies to partnerships. Organizations frequently treat partnerships as transactions, signing the deal and moving on to the next initiative, then wondering why the value never materialized.
“The real work starts after the deal is signed. Every time, without exception.”
Partnerships fail when they’re treated like transactions. They work when they’re treated like operating systems.
At Nike, Elaine saw what happens when you bring in something that runs on a completely different operating model, different culture, different cadence. At a certain point, you’re not running one company anymore but two, and the coherence between them becomes a full-time leadership challenge that most organizations underestimate until it’s too late.
Buying a platform does not automatically resolve this either. An organization that buys technology and sees no change in how people operate has simply moved the problem rather than solved it. What a good platform decision delivers is coordination, a shared system that removes fragmentation rather than adding more capability on top of a structure that is already struggling to hold together.
What the organizations making progress have figured out
The organizations making progress are not the ones with the biggest AI budgets or the most sophisticated tooling. They are the ones that recognized early that the technology was never going to be the hardest part.
The harder work is building an operating model that holds together as complexity increases. Markets need to stay aligned, partnerships need to function beyond implementation, and governance needs to scale alongside the technology. How an organization gets there, whether it builds, buys, or partners, matters less than whether the outcome is genuine coordination or just more tools on top of the same problem.
AI is extraordinary at pattern recognition and operating at scale, surfacing signals humans would miss and handling volume that would exhaust any team. But customer-facing work that depends on context, emotion, and judgment requires human ownership that cannot be automated without eroding the trust it is meant to protect.
The companies solving those questions are building systems that become stronger as they scale, while organizations still treating deployment as the finish line are often creating a more expensive version of what they already were.
Watch the full conversation
This article draws on Elaine Barsoom’s conversation with Jason Hemingway on In Other Words, the podcast from Phrase. Elaine is the former Global Head of Tech Innovation Partnerships and Strategy at Nike and now Venture Partner at Silicon Foundry.
