For all the talk of AI revolutionizing business, a statistic remains stubbornly telling: only about 1 in 10 AI proof-of-concept projects actually scales to enterprise value. The promise is there: operational efficiency and predictive insights, but most companies remain stuck in experimentation. So what separates fleeting experiments from true transformation?
The gap, as it turns out, is organizational, cultural, and strategic. The task confronting many leadership teams today is whether they can make it work for their company at scale, and whether the investment will justify itself.
In a recent panel at MachineCon New York, moderated by Asif Ghatala, Vice President and Business Unit Head at Tiger Analytics, this challenge was unpacked by senior leaders with firsthand experience navigating that journey. The speakers included Arys Nogueron, Campaign Director of Data Analytics, Reporting, and Governance at NYU Langone Health; Manav Misra, Chief Data and Analytics Officer at Regions Bank; and Rama Donepudi, SVP and Global CIO at Mead Johnson Nutrition. Together, they represented a cross-section of industries: finance, healthcare, and consumer nutrition, but voiced common themes about how to unlock AI’s true ROI.
From Concept to Confidence
Getting past the POC stage takes organizational trust, speed, and a willingness to fail productively. Several noted that early AI efforts were fast and scrappy, with short feedback loops, numerous failures, and small wins that helped build credibility. In regulated industries like healthcare and financial services, even small experiments often require diplomatic stakeholder engagement and stringent risk assessments.
The critical pivot from isolated success to broader value comes when AI projects are tied to real business problems and sponsored by leaders who are accountable for results. Success wasn’t just measured in model accuracy or predictive power, but in adoption, by branches, clinicians, or campaign teams, and the impact on topline or bottom-line performance.
One strategy that proved effective: have business stakeholders co-invest in AI initiatives, financially or with dedicated resources. This not only ensures buy-in but helps weed out low-priority ideas early. Others embraced a product-oriented mindset: treating AI initiatives not as tools but as end-to-end solutions that solve concrete problems and can be iterated on and scaled like any digital product.
What Works and What Doesn’t
There’s no single recipe for scaling AI, but a few patterns emerged. First, successful organizations avoid over-indexing on either experimentation or control. Too much of the former results in a graveyard of demos; too much of the latter, and innovation stalls.
In healthcare, for instance, even with strong POC results, clinicians may remain skeptical of AI unless engaged early and shown clearly how it enhances their decision-making. Meanwhile, financial institutions, despite strict regulation, are among the most advanced in AI deployment, with systems already embedded in fraud detection, underwriting, and customer analytics.
Across sectors, the gating factor has shifted. It’s no longer about data access or executive interest, if anything, there’s an oversupply of ideas and enthusiasm. The challenge is now prioritization. Teams are flooded with proposals from business units and C-suites alike, many backed by budget and urgency. The question has become which use cases to bet on, and how to say no effectively.
Governance Is the Missing Layer
As interest in agentic AI and generative applications grows, governance has emerged as a pressing concern. Companies are wary of recreating past mistakes from the robotic process automation (RPA) wave, where a proliferation of bots led to unmanageable complexity. Without strong visibility and oversight, even helpful automation can turn into a liability.
While some were enthusiastic about empowering employees across the organization to build and test AI-driven tools, others sounded a note of caution. Allowing too many unsanctioned agents into workflows, especially in regulated contexts, could create chaos. Guardrails and lifecycle management are essential.
Interestingly, a few organizations are already embracing this controlled decentralization. By providing company-wide access to generative AI tools with strict constraints, they hope to unlock grassroots innovation while maintaining security and compliance. In one case, an annual “Discovery Day” will now feature an internal competition where employees showcase the most valuable AI tool they’ve built. The goal is to scale through distributed experimentation.
From Constraints to Creativity
Even heavily regulated or compliance-heavy industries found ways to innovate. Some AI products were explicitly designed to help organizations navigate red tape faster, not skirt it. In consumer nutrition, one use case centered on automating the campaign approval process by training generative models on years of legal and regulatory feedback. The result was the ability to run more campaigns with the same resources, directly driving revenue.
Others are exploring AI to streamline notoriously bureaucratic workflows, such as insurance authorizations for healthcare procedures. Here, agentic AI could reduce approval times from weeks to seconds, transforming patient outcomes.
Still, excitement was tempered by realism. Leaders were frank about the need to pace innovation with organizational capacity. While boards may be pushing hard for ambitious AI-fueled transformations, execution teams are now the bottleneck, fielding more ideas than they can evaluate, let alone implement.
From Top-Down Demand to Bottom-Up Discipline
Perhaps the most striking shared insight was this: enthusiasm is no longer the problem. If anything, senior leaders are more bullish on AI than their own data teams. Budgets are flowing, use cases abound, and stakeholders want in. What’s needed now is disciplined filtering, robust governance, and strategic focus.
To move from scattered experiments to transformative impact, companies need to reframe the AI journey as a series of business bets and operational shifts
