AI deployment is a business problem, not just an IT project.
CIOs are increasingly tasked with helping their organizations adopt AI. But handing that responsibility entirely to IT can create a disconnect between the technology and the business. IT may know how to build and deploy AI, but it may lack sufficient insight into day-to-day business operations to identify which use cases will deliver meaningful results. The result can be a significant investment in AI projects that fail to solve a problem the business actually cares about.
Instead, some organizations are creating dedicated AI enablement teams that work directly with business units to identify their challenges, prioritize AI use cases and help deploy technology that improves business outcomes. The approach can help organizations scale AI without simply treating it as another IT project.
TechTarget interviewed Jim Begley, CTO of ARG, a Virginia-based IT services and consulting firm, about how his company built its AI enablement team and what CIOs should consider when creating one.
Editor's note: The following transcript was edited for brevity and clarity.
What does your AI enablement team do at ARG?
Jim Begley: Internally, we first engaged a consulting firm to help us develop a good proof of value (POV). We said, "OK, how do we look across the ecosystem and understand where we could apply AI to get outcomes?" And when I talk about AI capabilities, I'm referring to the large language model (LLM) solutions. We also have agentic CX solutions that fall into a different category.
We took one of our top business analysts and applied thoughtfulness to the use cases that people identified. We did that based on what would be easiest to implement with the highest impact. We rated them across those capabilities and then stood up a separate team -- pulled people out of their group to deliver on this.
We had a group of two people who were responsible for sitting with the business teams, identifying their challenges and developing a workforce agent to help them do their jobs in small, incremental ways that would improve their outcomes.
In our first tranche, we drastically increased the scalability of the organization. We weren't necessarily looking for ways to eliminate head count, but were looking to create more scale and capabilities within the organization. So, in our first group, we strung together custom GPTs and agents to help them deliver faster while maintaining human-in-the-loop capabilities.
Could IT not handle this responsibility itself?
Begley: I don't know that IT couldn't handle it, but it requires someone focused on the business outcomes. That's where IT gets a little hung up. For example, one of our clients dedicated three or four DevOps engineers to delivering AI outcomes. They approached it from the perspective of "We know how to deliver IT," but they didn't engage with the business to identify which capability would yield the most successful outcome.
Part of the problem was that it was living within IT, and they don't necessarily understand the business operations impact.
They had ideas, engaged a couple of business units, got some use cases, invested quite a bit of money in both resources and people, and then found that the business was like, "Yeah, the juice isn't worth the squeeze."
It's not a bad story because you want to see that failure early and say, "OK, stop. What did we do wrong?" Part of the problem was that it was living within IT, and they don't necessarily understand the business operations impact.
We focus our clients on identifying the business operations impact they're trying to achieve. AI enablement teams should sit with the group they're trying to help, understand how they operate and then understand how to deliver something that will make their lives easier. Of course, they are IT people by and large, so it's not that we're having non-IT folks doing this. They're just not living in IT and getting IT outcomes.
What roles and expertise make up the AI enablement team?
Begley: It's in that business analyst kind of view where people are seeing good success. They have to be strategic thought leaders who can look at the business outcome and go, "OK, here's what they're doing operationally. How do I improve the business outcome?"
They have to be strategic thought leaders who can look at the business outcome.
Did you hire externally at all for this team?
Begley: We didn't, but we do use additional resources as needed. For instance, if there's some specific technical skills that we need, we will reach out and pull some people in for that.
How big should an AI enablement team be?
Begley: It depends on the organization and what you're doing. We're about a 100-person organization, and we have two people -- one and a half, really -- focused on it. The other organization I was talking about has about 1,500 people, and they dedicated five engineers and a manager to developing their POVs. That was just POVs, not through to implementation.
It will also depend on the organization's capabilities and where it fits within its portfolio. If you're a development organization, your adoption curve and capabilities will be very different from those in a traditional knowledge work environment.
Does every big organization using AI need a dedicated AI enablement function?
Begley: If an organization has developed a good -- and this typically lives within the IT team -- training and methodology to get end-user adoption of capabilities, then it may not need a dedicated AI enablement team. If we think about how AI is being used by different organizations, the chat functions of LLMs and their adoption may be simple and driven through the normal IT organization.
Securing that may be done by a different group. Then the development side of that, and how they're integrating any code development into their organization, has a different adoption curve than just the chat function of the LLMs with end users, so it really depends on what they're specifically doing.
Can you give me an example of a real AI initiative at ARG that the enablement team was involved in?
Begley: We conduct analysis for clients to help them understand their billing capabilities. They may give us their invoices, and we've built a private LLM -- a small model where we have control over the data. We use it to analyze hundreds, even thousands, of line items and consolidate them into a single bill format.
They may have 100 different invoices from all over the world for their telecom spend. That gives us the ability to take that information in and do a quick analysis. Sometimes that would take us weeks. Now it can be done in less than a day. We can consolidate it into a simple format and use it to price out other solutions for them, or they can review it to understand where their costs are and how they're being tracked.
One example involved converting foreign currency into U.S. dollars. In the past, we would have just exceptioned that out and said we couldn't do that because we didn't have anyone who could read the language.
Now, with the LLM, we're able to translate it into English, run it through, get that conversion in, and evaluate whether the dollar conversion rate was correct. Those are capabilities that, frankly, we just didn't have before. Now we're able to execute on that much faster and much cleaner.
Are there any drawbacks or challenges to creating an AI enablement team as a separate function?
Begley: There are. One challenge has been keeping everybody abreast of what's happening, because they are siloed. They tend to be running quickly in directions -- and good directions -- but there's people off to the side who aren't aware of where we're going.
One challenge has been keeping everybody abreast of what's happening, because they are siloed.
Every once in a while, somebody within the organization will get a piece of information about something that's been going on without the full picture, and either take them down a path we don't want them to go down or not understand what's happening within the ecosystem.
How do you measure whether the AI enablement team is successful?
Begley: We measure the teams that they're impacting. When we look at our normal KPI reporting, we can see the actual net positive impact of what they're doing. For instance, we might notice teams doing more with the same.
Some of it isn't necessarily about having ROI. We're still in the early stages and willing to experiment and invest. Of course, over time, we want that and will get that, but we're willing to invest because we're ahead of the curve.
Tim Murphy is a site editor and writer for the IT Strategy team at TechTarget.