Guest: Scott Morgan, EVP of Data and AI at Marlabs

Enterprises have the models, the data, and the talent. Yet the overwhelming majority of AI initiatives never survive contact with the real world. According to research from MIT, 95% of AI pilots fail to produce measurable returns, and the cause is not the technology. It is the organizational gap between the teams building AI and the teams expected to run it, a gap that is widening as pressure from executives and boards intensifies and patience runs out.

Scott Morgan, EVP of Data and AI at Marlabs, has spent years at the center of this problem. In a recent conversation with TFiR, he laid out exactly why enterprises are stalling in pilot purgatory and what it actually takes to industrialize AI at scale.

The Real Reason AI Pilots Stall

Morgan is direct about where the fault line sits. “Technology is not really the problem here,” he said. “The AI tech that exists out there is outstanding. It only gets better, exponentially better by the day. It’s not a technology problem. It’s a business problem.”

What he means is a lack of organizational maturity and strategic discipline. Morgan points to four pillars that consistently separate enterprises that are succeeding with AI from those that are not: aligning AI initiatives with core business outcomes, building a data-ready foundation, operationalizing AI across the enterprise through mature governance, and investing in organizational readiness including change management and AI literacy.

The data quality issue in particular continues to trip up organizations that rush to deploy AI before their data house is in order. “Garbage in, garbage out. You’re not going to do AI well if you still have bad data quality problems,” Morgan said.

From Pilot to Production: What It Actually Takes

One of the most persistent misconceptions Morgan encounters is treating AI deployment as if it were purely an AI engineering problem. In reality, he estimates that AI engineering represents only 30 to 40% of a full AI implementation. The rest, including architecture, design, infrastructure, validation, testing, and long-term maintenance, gets neglected in POC-focused projects.

“A lot of these AI POCs were really just focused on the AI technology part, and we neglected those other components,” he said. The enterprises making real progress are the ones that have put mature governance structures in place, whether through a Chief AI Officer, a center of excellence, or some other centralized function that standardizes infrastructure, tech platforms, policies, and use case approval.

Agility AI: Closing the Gap with Pre-Built Accelerators

To help clients move faster without starting from scratch, Marlabs developed Agility AI, a framework built around three components: AI strategy alignment, a catalog of pre-built agentic accelerators, and an enterprise governance model.

The accelerator catalog is where the framework delivers its most tangible speed advantage. Marlabs has built more than 100 agentic accelerators spanning technical and industry-specific use cases. A Databricks migration accelerator, for instance, uses eight to nine specialized agents to execute platform migrations and has delivered 50 to 60% reductions in time, effort, and cost for clients. In financial services and lending, Marlabs has developed 15 to 20 accelerators covering eligibility prediction, dynamic pricing, automated income verification, and trigger-based sales outreach. Similar catalogs exist for healthcare providers and life sciences.

The governance layer is what Morgan considers the most important piece. “That’s always the big struggle right now. It’s the ability to operationalize consistently across the enterprise, and not do it in this siloed fashion.”

Shadow AI and the Platform Fragmentation Problem

Beyond governance gaps, Morgan is watching another risk grow. Shadow AI, where teams deploy their own AI tools without central oversight, is becoming increasingly common as the number of AI point solutions on the market explodes. “There’s an AI point solution for almost every business process at this point,” he said. Without clear ownership, organizations end up with fragmented tech stacks, competing platforms, and mounting technical debt that makes scaling to production exponentially harder.

In response, Morgan is seeing a growing number of enterprises take a build-over-buy approach, creating centralized AI platforms built on open source technologies that integrate with 20 to 30 enterprise systems through a single interface. “You can basically do anything you want from an AI point of view, getting all of your data from your ERP, CRM, HRIS, and ITSM systems, as well as your proprietary data lakehouses.”

What C-Suite Leaders Should Do Differently

For executives who feel stuck, Morgan’s advice is consistent: stop chasing technology and start enforcing strategic discipline. Focus AI investment on high-ROI workflows tied to margin expansion, customer retention, and speed to market. Bake AI into all workflows, not just isolated pilots. And commit to a well-defined, well-governed infrastructure and platform strategy. “I see way too many clients that have built multiple AI solutions with completely different tech stacks, completely different platforms. That makes it exponentially harder to operationalize AI throughout the enterprise.”

As for what comes next, Morgan sees specialized agentic AI as the defining shift of the next 12 to 18 months, with agent capabilities advancing faster than most enterprises are prepared for. The organizations that invest in governance and operationalization now will be the ones positioned to capture that value first.