Making Workforce Training Affordable with Tiered Storage - with Aaron Demory of Fearlus

**Daniel Faggella** (0:12)
Welcome everyone to the Emerge AI in Business Podcast. Today's guest is Aaron Demory, Senior Partner at Fearlus and Chief of Information Technology and Security at the FDIC. Aaron joins Emerge's Matthew DeMello to discuss how large organizations can manage the infrastructure demands of AI adoption while controlling costs and mitigating risk. Throughout their conversation, Aaron explains why starting small with AI initiatives helps align technical ambition with realistic budgets, the role of data tiering in balancing costs and compliance, and how framing infrastructure investment through risk mitigation supports sustainable AI strategies. Just a quick note for our audience that the views expressed by Aaron on today's program do not reflect that of Fearlus, FDIC or the respective leadership at either organization. In AI, we see a lot of skepticism and for good reason. The challenge isn't excitement, it's execution. Data readiness, integration complexity, security concerns, proving ROI, it all has to align. Our sponsor, SHI, has been a major global IT solutions provider for over 35 years, helping 17,000-plus organizations navigate exactly these challenges. SHI guides enterprise leaders through their proven framework. Imagine the right AI strategy for your business context. Experiment in their AI and cyber labs to validate solutions with your actual data and workload, typically in 2 to 6 weeks. And adopt solutions that deliver measurable outcomes in production. If you're looking for guidance on moving from AI ambition to real results, check out shi.com. That's shi.com. Click for their AI and cyber labs or schedule a consultation with their teams. For our solutions partners, position your brand alongside the Fortune 500 leaders defining the enterprise AI roadmap. For the opportunity to showcase your solution to the executives currently funding and scaling global initiatives, partner with Emerge. Secure your partnership at go.emerge.com.
That's go.emerge.com.

**Matthew DeMello** (2:08)
Welcome back to the program, Aaron.

**Aaron Demory** (2:09)
Thank you, Matthew. Great to be back.

**Matthew DeMello** (2:11)
Absolutely. We're talking about a slightly different topic today, diving into especially what these very large systems look like on the data side and how comprehensive that view needs to be to really make changes going into the future, what executives need to know. Leaders, of course, are under immense pressure to prove the ROI of AI initiatives while containing infrastructure costs, maintaining compliance, and too often organizations underestimate the compute storage and governance demands of large scale data projects, leading to budget overruns, stalled implementations and eroded stakeholder trust, aligning technical ambition with a realistic understanding of infrastructure constraints is critical to avoid strategic missteps and deliver sustainable value. But just from that high level, why do ambitious AI roadmaps fail when storage and compute needs are underestimated?

**Aaron Demory** (3:14)
One of the major reasons I see is fighting off more than we can chew at first. I think that incremental advancement is really a prudent move that we need to be thinking about at the enterprise level. We need to temper that with value and ROI. So if we're starting small and we're kind of gaining confidence as we move along, we'll be able to keep our associated costs relatively low. Obviously, when you start an initiative, you're going to hit a spike and that's to be expected. You want to keep your expectations low, though, so you can start to move strategically. The issue that I've seen is we'll start off and say, well, we need to have a generative AI capability in-house and how quickly can we get it? How much is it going to cost? We all know IT projects notoriously go over budget, over schedule. What I see with AI adoption is a couple of different things. One, we'll underestimate the amount of compute this actually takes to pull off in terms of the type of infrastructure, the cost for that infrastructure, and the scale that we're looking at with our large enterprise initiatives. Even a scale back initiative with a large enterprise is still quite large. We're not talking about the garage startup. We're talking about using the funds in the millions of dollars, and that is going to be a huge amount of compute. If our ambitions are too big for the capability, even though we have the funds to support it, it's going to eat through them very, very quickly.
Something that we saw with the push to cloud, a lot of organizations assumed that moving to cloud was going to be cheaper, so we kind of packed up our data centers and we moved to cloud right away. For some, it definitely is a little bit cheaper. For others, they realize they're running servers 24-7, and it is not cheap at all. The cost actually exploded. We're seeing the same thing with AI adoption. If we're going after something that's using heavy infrastructure for too long, it's going to be very expensive. And the return on investment that you're going to get from that is not going to be immediate. A lot of these are building capabilities for the long term. Models have to be trained. We've got to go through ethical reviews of what those models are doing. The neural networks that are running underneath take a huge amount of resources to actually build and train on a data set. So strategically, from a business standpoint, we have to be thinking about, when do we need that ROI and what needs to happen in order for us to capture that. The cycle is going to be quite large for some very ambitious objectives. And for those reasons, I see that the AI adoption failing, not because of the technology, and it's not because of the organization's readiness for it. It's simply because of not right sizing the IT effort to begin with and being comfortable with the associated cost and the business cycle for what adopting that particular type of AI is. So, those are all questions that you'd start off with from a strategic standpoint that when done well in mature organizations, you see them starting small, doing small pilot projects, putting more of an incubator together where people are learning the technology and learning how to get it into their budget cycle, more so than going for a generative AI, for example. We need to get something like that or we need to do something like the popular products out there. That's where the cost balloon and your CFO ends up cutting it before it has a chance to give ROI.