Enterprise Strategy Group Analysts’ Perspectives
View All Series Articles
Opinion

Why managing GenAI models is like owning a paddleboard

Lessons on GenAI model management can come from unlikely sources -- even from buying a paddleboard for the first time.

Mark Beccue
By
  • Mark Beccue, Principal Analyst
  • Enterprise Strategy Group

    We provide market insights, research and advisory, and technical validations for tech buyers.
Published: 09 Sep 2025

I've been thinking about buying a paddleboard for a few years, and I finally decided to buy one last week. I researched options, visited a paddleboard store and consulted with experts about which board to pick. Once we settled on one, I was excited to start my first paddle with dolphins and sting rays in the clear waters of Tampa Bay off Saint Petersburg.

But I quickly discovered I didn't have everything I needed. I also had to consider car racks for transport, wet bags for storage, locations to paddleboard and methods to maintain and protect my board.

This whole process is a work in progress, much like the journey enterprises take with generative AI models.

Many enterprises are currently navigating the operationalization of generative AI. Feeding and caring for generative AI models is at the core of that effort. Sure, enterprises must make decisions to choose models, but the path to successful deployment is fraught with other decisions and obstacles. As with paddleboarding, it's not just about the models themselves; it's equally about the ecosystem of tools and support needed to reap the benefits of the models.

As companies venture deeper into this territory, several challenges have emerged that can hinder or accelerate their generative AI journey.

Technical hurdles

Perhaps the most immediate concern is the infamous hallucination problem -- when AI confidently produces inaccurate or fabricated information. This undermines trust and can lead to serious business consequences when decisions are made based on flawed outputs. How do enterprises address hallucinations? What are the tools and techniques required to keep it under control?

Model selection and customization present another significant barrier. Organizations must navigate a complex, shifting landscape of available models, each with different capabilities, limitations and resource requirements. Adapting these general-purpose models to specific business contexts often requires specialized expertise that many companies lack. What are the tools and approaches to optimize choices?

The computational demands of running generative AI workloads cannot be overstated. The high cost of compute resources necessary for training and inference can quickly drain budgets, especially when scaling beyond initial pilots. This hurdle is huge. Enterprises will not tolerate AI workload compute costs if they remain cost-prohibitive.

Governance and risk management

As generative AI becomes more integrated into business processes, governance concerns take center stage. Organizations struggle with establishing frameworks for model lifecycle management -- from performance monitoring to regular evaluation and updates.

Security vulnerabilities and compliance issues add another layer of complexity. Many regulatory frameworks weren't designed with generative AI in mind, leaving companies to navigate uncertain compliance waters while protecting against novel security threats.

Ethical considerations and bias mitigation require ongoing vigilance. Without proper oversight, AI systems can perpetuate or amplify existing biases, creating reputational and legal risks.

Strategic challenges

Perhaps most fundamentally, organizations struggle with quantifying the ROI of their generative AI investments. The transformative potential is clear, but measuring concrete business value remains elusive for many.

Data readiness -- having sufficient high-quality, well-structured data -- is a prerequisite that many organizations underestimate until they're deep into implementation.

The journey from successful pilot to enterprise-wide deployment represents a significant scaling challenge that requires thoughtful infrastructure planning and change management.

Organizations that proactively address these challenges will be better positioned to harness generative AI's transformative potential while minimizing risks and maximizing returns on their AI investments.

Research coming: The ecosystem to support generative AI models

Enterprise Strategy Group, now part of Omdia, a division of Informa TechTarget, is fielding a study to explore generative AI model trends. It's hard to name a technology that has evolved as quickly as AI models. Organizations are scrambling to understand and develop their strategies for AI models, but best practices for building strategic and technical plans are limited.

The study will focus on the ecosystem needed to support models, from the AI infrastructure required to the tools, platforms and services needed for AI model lifecycle management. It will cover drivers and barriers, buying personas and budgets.

One of the areas I'm particularly interested in seeing is the responses to our questions around which generative AI tools, platforms, products and services are most important to generative AI model initiatives. Related to that: Do enterprises prefer open source or proprietary models, tools and platforms? The answers will certainly be interesting.

Mark Beccue is principal analyst at ESG, now part of Omdia, covering artificial intelligence.

Enterprise Strategy Group is part of Omdia. Its analysts have business relationships with technology vendors.

Dig Deeper on AI business strategies