3 strategic AI frameworks to move from experiment to execution

Reading Time: 9 minutes

Remember the halcyon days of artificial intelligence, when your coworker asked ChatGPT to write his out-of-office vacation message as a limerick? Or when your teenager prompted it to write a rap battle between a revolutionary hero and a current chart-topper for school?

Those experimental days are over.

AI is now a core business driver that requires a new operational framework to lead your organization into the future. Successfully scaling AI across the enterprise means revisiting your infrastructure, shoring up your governance, and reassessing your value indicators.

At our recent SHI Summit — Scaling smarter: Infrastructure for the AI era — leaders from SHI and beyond shared real-world insights to help you build the operational AI frameworks needed to move your visionary projects from experimentation to execution.

1. Build specialized, scalable AI infrastructure

The infrastructure your organization painstakingly built a few years back may still be well-positioned for a pre-artificial intelligence world. But AI’s compute and cooling demands are forcing innovative organizations to rethink their infrastructure strategies entirely.

On-prem versus cloud versus hybrid infrastructure

Ian Fisk, Chief Technology Officer of the Simons Foundation, has found that the best way for his organization to afford AI projects is to bring its infrastructure on-prem. He likened the economies of scale to obtaining a car; while the cost of purchasing a car initially exceeds that of a rental, it would be far more expensive over time to rent indefinitely because the car rental agency needs to make a profit — as do cloud providers.

Of course, there’s no one-size-fits-all approach to infrastructure. But we’re already seeing some shifts. In a poll of SHI’s summit attendees, 23% indicated they’re already running production AI workloads on-prem, while only 8% are cloud-only for AI workloads. The rest are still developing their AI strategies and evaluating how they want to shape their infrastructure for 2026 and beyond.

AI-ready cooling strategies

Beyond infrastructure location, organizations must build cooling strategies and consult unlikely sources for assistance. Concerns about the amount of particulate in water lines and filter systems for water safety mean IT leaders must now check in with plumbers on a regular cadence. When asked about their primary cooling strategies for high-density AI workloads, 44% of summit attendees admitted they’re still hoping for the best with traditional air cooling, while 27% have pivoted to hybrid air and liquid cooling solutions. But even these modernized strategies bring their own questions.

Liquid-cooled servers require a pipe connecting the cooling system to your server. Who owns the pipe? Who owns the server? If one of the servers gets clogged with particulate, who owns the fix? Your contractor? Facilities? You need an ecosystem of partners involved to ensure you have the right temperature, flow, purity of water, and worker safety measures in place to automatically shut down rack power in the event of a leak.

Upgrading your power system is still a hidden ‘gotcha’ in these nascent days of artificial intelligence. It requires planned outages, which is often an unpopular conversation — you must sell leadership on downtime for upgrades to infrastructure no one sees. But it’s critical to shift this conversation. Building AI projects requires simultaneously growing your power capacity and cooling strategies.

We recommend that you:

• Assess your spare power and cooling capacity.
• Understand what you need to successfully launch your AI project.
• Build a realistic path between points A and B prior to investing in AI infrastructure.

2. Establish governance frameworks for cost, compliance, and security

As organizations rush to demonstrate value from AI, new risk factors are materializing. Last year, just 43% of organizations believed they had full visibility across their software assets, according to Flexera’s 2025 State of ITAM Report. This is largely because of shadow AI or ML tools in their environment that IT was not previously aware of, which is putting data at risk every day. Even well-meaning employees are feeding confidential information to public generative AI tools. It’s time to lock these practices down.

Building AI software governance

Governing your AI software stack requires three phases: discovery, management, and optimization. If you haven’t established rules yet, start by defining scope, ownership, and functionality; for instance, are your employees using generative or predictive AI? Talk to your stakeholders to figure out what tools are being used, for what purposes, and what goals. We recommend conducting an employee survey to gather this information, but you can also shore up your investigation by reviewing procurement records to see what’s being expensed. Look into formal AI governance frameworks as guardrails, like NIST AI Risk Management Framework or the ISO/IEC 42001 AI Management system.

Once you understand what your employees are using and what their goals are, integrate with FinOps and start reviewing policies and setting controls. According to Flexera’s 2025 State of ITAM Report, 38% of organizations have ITAM teams working closely with FinOps teams. Get ahead of the curve and start small and iterative. Is that public chatbot approved in your organization? Are your AI tools providing discriminatory results? Determine which tools you are comfortable with and set boundaries to ensure employees cannot use unapproved solutions. These policies are the foundation of your AI software governance plan.

Post-rollout, treat AI like your other software assets by implementing lifecycle management controls. Who is using it? How much does it cost? Who is paying? When will it be retired? Conduct real-time monitoring and reporting on usage and compliance. Don’t wait for the monthly bill to explore spikes; be proactive and understand your data across the organization. And continue to optimize and improve upon your investment — AI will only become more complex over time. But embracing the rush to AI doesn’t have to mean chaos.

Meet compliance mandates

Section 508 of the U.S. Rehabilitation Act requires federal agencies, state and local governments, and their contractors to ensure all websites, applications, and digital services are accessible to individuals with disabilities via features like alt text and captions. The Americans with Disabilities Act also uses WCAG to establish web accessibility compliance. State and local governments are required to meet these standards by April 2026 or April 2027, depending on headcount.

To assist public sector organizations with this directive, SHI has released a comprehensive Section 508 compliance solution developed in collaboration with HPE, NVIDIA, and Kamiwaza. This solution “addresses accessibility barriers across visual, auditory, motor, and cognitive needs, ensuring compliance with Web Content Accessibility Guidelines (WCAG) standards.”

Are you embedding accessibility into your AI and cloud initiatives? These steps are foundational for your data, allowing more complex autonomous systems to be built on top of it.

Contain identity sprawl

Traditional identity models are not equipped to handle AI projects. CyberArk estimates that non-human identities in AI-powered enterprises outnumber humans 82:1 — and will soon skyrocket. And with 29% of our summit attendees planning to use agentic AI within the next 12-18 months, it’s clear that now is the time to bolster identity and access management (IAM) strategies before a catastrophic breach.

But in this case, the problem and the solution are aligned — intelligent AI-driven IAM is capable of analyzing billions of events in real time to spot threats. Already, AI agents, bots, and IoT devices outnumber human users in many organizations. We’re seeing organizations add model context protocol (MCP) interfaces to their IAM solutions to handle this influx of agentic AI. These open-source frameworks standardize how large language models interact with external data, applications, and services.

According to the 2025 IBM Cost of a Data Breach report, organizations using AI-driven security save an average of $1.9M per breach. By operationalizing your IAM solutions, aligning with zero-trust principles, and reducing identity sprawl, your organization can build more secure user experiences. Our modern IAM playbook can help you build a cohesive strategy across lifecycle and governance, authentication, privileged access management, cloud identity management, customer IAM, and identity threat detection and response.

3. Quantify and communicate AI business value

Without measurable ROI, AI initiatives stall out. That exhilarating look at a time-saving automation or efficiency driver fades away when you can’t measure success. To avoid this corporate heartbreak, we recommend embracing FinOps as a strategic enabler of AI that can offer winning functions like cost modeling for generative AI to help support your business case.

When it comes to quantifying AI value, many of the principles you establish in your governance framework can be leveraged again:

• Inform — Do you have visibility into what tools you’re using and how they are allocated?
• Optimize — What does your total utilization look like?
• Operate — Are you using data to continuously improve and optimize your AI toolshed?

Adopting a standard methodology for your CIO to review business value and establish accountability goes a long way towards empowering future data-driven decisions. Quantifying the business value of AI requires more than theory. It demands accurate data, targeted frameworks, and strategic calibration across people, process, and technology.

How to create the future in the era of AI

The path to the future isn’t paved with AI use cases, incremental improvements, quick wins, and low-hanging fruit. It must start with a bold vision — and the resources to make it happen. AI strategist Brian Evergreen likens this sentiment to building the Duomo di Firenze in Florence, Italy. This UNESCO World Heritage site wasn’t built by handing every citizen a shovel and telling them to get to work. Filippo Brunelleschi was a visionary architect, designer, and sculptor who conceived of the brilliant dome and led a team in bringing it to life.

“We’re all going to the future,” Evergreen noted. “The difference is how we’re going to get there. Are you going to back into it, or are you going to create it?”

If you’re reading this piece, you likely already believe in the transformational power of enterprise AI. And you may already have an idea to improve a practice or two within your organization. What do you want to accomplish? Define your vision and make your strategy visible. What systems across your organization will be impacted by this project, and how do they interact with each other?

Evergreen argues that as humans, our goal should always be to create new value. In his Total Addressable Value Creation Framework (TAVC), he points out what’s addressable by humans and what’s addressable by AI agents and implores us to shift our focus. Rather than focusing on using AI to make what humans can already do slightly faster, consider what could be.

“Your career is filled with people telling you what to do next,” Evergreen said. “But that’s not where vision comes from. Pick up the mantle, be the visionary leader; data won’t tell us and it won’t come from trend analysis. Move beyond that to something that would actually be valuable.”

AI is a strategic imperative

Organizations that master these operational AI frameworks will lead us into the future. AI is no longer a lab experiment — it’s a strategic imperative. Here’s your to-do list:

1. Build your future-ready AI infrastructure, primed to handle compute and cooling needs.
2. Govern your AI stack, stop the sprawl, and secure your data.
3. Show your work by quantifying value and gaining C-level and public approval.

SHI offers proven, actionable AI frameworks for CIOs and technology leaders. Our team has deep expertise in FinOps, ITAM, and AI strategy, and our data-driven calibration helps organizations like yours unlock hidden value. Still not sure what solutions are right for your team? You can access, test, and validate technology at our AI & Cyber Labs prior to investing, with guidance from our staff.

It’s time to be bold and bring your AI vision from experiment to execution.

“If not you,” Evergreen asked, “Then who?”

NEXT STEPS

Ready to modernize your AI infrastructure and operationalize secure, scalable AI systems — and prove their value? Contact AI@shi.com to connect with our AI experts.