What would you say is the best way for a CIO to determine what AI can bring to a company as its highest and best function?

Trkay: My initial way of introducing it to my team has been focused more on: How do I extend key capabilities I’m looking for within my organization to a broader portion of my organization? I don’t look at it and go, ‘Ooh, how do I replace software engineers with AI?’ I actually say, ‘How do I take a larger number of my technically oriented staff and allow them to do developer-oriented tasks?’ Things that maybe their individual experience hasn’t been based on yet, but they have a technical foundation that AI becomes assistive, helps get them further down a path and extend into a higher level of functions than what they’ve traditionally been able to do. 

It’s now [been about] 24 months on this AI hype cycle. There was this throw-the-spaghetti-against-the-wall approach that a lot of companies started taking, where they were doing pilots and proof of concepts, and any sort of opportunity they saw, they took a swing at it. 

But one of the hardest parts about the utilization of AI is the operationalization of it—not just taking it from something cool and whizzbang that you’re doing on your desktop, but actually making it part of the operational backbone and character of an enterprise. To really make it enterprise class and enterprise worthy. There’s a lot of things that go into that. It has to do with making sure that it’s robust, resilient, reliable, safe against a wide range of conditions. Making sure it’s explainable, ethical, auditable. Then there’s also how to make sure I’m doing it at scale, and being able to handle the throughput that I need to. 

In that initial phase, what I saw industrywide is lots of pilots, lots of interest, lots of money thrown at all of that activity—but very little of it ending up returning value to the business because they couldn’t get it operationalized in a material or substantial way that fundamentally changed the way they were working. 

I do think that’s starting to pivot. I see and hear more and more from peers and colleagues. They’ve created a more focused portfolio of AI initiatives that, rather than starting with the demo and the ‘Isn’t this cool?’, they’re starting with the business objective and value that they’re trying to extract. They’re working their way back into what are the AI tools, technologies, approaches that help us address that and bring some of that value to the business.

I don’t know if that really has resulted in change in investment in AI in terms of absolute dollars,  but I do think it’s resulted in a change of investment where it’s become more focused dollars on a smaller subset of tools and technologies, or applications or use cases where they’re trying to get it applied.

Is there anything that CIOs should look for when they are evaluating potential AI platforms and applications?

There’s two types of data that I’m concerned about. One is the type of data that we’re managing for our customers, which tends to be privileged data—PII or PCI data—just data in general. How do you ensure that there’s no risk of data leakage? A lot of times when you are going to publicly available AI models, people don’t realize it’s capturing that data, utilizing [it] to further refine and train itself to kind of not only personalize its responses to you, but also to further get feedback on, ‘Am I giving the right sort of answers,’ and utilizing that for the next iteration of the model. [You need to be] making sure that you’re [trying out AI] in an isolated, data-fenced, and generically secure way.

We’re also a company. Our software is our IP. Even when looking at things like assistive development, if I’m providing my source code over to something to say, ‘Hey, help me work on this,’ I’m now exposing my IP. This still falls into sensitive or privileged data. Most CIOs should bring that perspective to the table because they’re used to understanding digital loss prevention from data leaks. 

Then, there’s a lot of risk around responsible AI stuff. A lot of the tools tend to hallucinate. They can give you kind of false answers. So anything that’s critical, even if we’re utilizing AI today, still a lot of times we have a human-in-the-loop approach to what we do. AI is making a suggestion. It’s doing some work, but in a lot of cases, especially when the blast radius of impact is large, there’s a human in the loop that is reviewing and being the approver, to make sure there isn’t a hallucination or AI doing something a little bit off the mark that would create a risk for us and how we operate.

AI adoption that I’ve seen within the industry, [you need to determine] truly how do you operationalize it? You need to be thinking about that from day one. How do I do this at scale? How do I do this low latency? Not just think about it as a pilot, because there’s a lot of stuff that you can see when it becomes lost dollars. This is where, as the CIO, you’re wearing your financial management hat, going, ‘How do I make sure that I’m spending our money wisely?’

What advice would you give to a CIO who, at this point, is trying to quickly find the right thing to do with AI?

Start with the business value that you’re trying to get out of it. Don’t start with the technology. Don’t start with the contractor. Don’t start with the tool. Understand what is the agent, the action, the activity, the task that allows for that business value. And then you work your way back, because the more specific that you can get about what you are trying to get out of AI, the better it’s going to lead you to what is the right tool for the right fit. There’s a lot of things that can do a lot of things, but they’re 80% or 70% okay at stuff. They’re not great with anything. 

To get real business value, [you need to start having] a little bit more focus. What are the [platforms] that are not just good at a lot of stuff, but great at doing incident correlation, predictive root cause analysis, fraud detection, decisioning. Pick tools that are made for purpose to the specific tasks that you’re looking to get value out of.