Inside Highmark Health’s AI-Powered “Sidekick”: An Interview with the VP, AI COE

Highmark Health built Sidekick to do something many AI programs are striving to achieve: move past scattered pilots and make everyday work better for the masses in a large enterprise. Sidekick started as a secure way for employees to use generative AI without risking data leakage, then matured into a companywide portal where teams learn, experiment, and put AI to work on real tasks that tie back to Highmark’s Living Health strategy. The AI Center of Excellence positioned Sidekick as a people-first capability.

Two ingredients show up again and again in the Sidekick story: executive buy-in and responsible guardrails. The C-suite backed the rollout and set expectations for adoption, while a Responsible AI framework, modeled on NIST, defined policy, governance, and oversight. That allowed teams to move quickly without cutting corners on security, privacy, or compliance. The outcome is a safe, shared place to try AI, learn what works, and scale it when it does.

Sidekick’s success is rooted in blending top-down initiatives and bottom-up experimentation. There are training paths, “power hours,” and a super-user network to spread know-how. There is also an open intake process so frontline teams can surface pain points and test ideas. When a pattern proves useful, the COE can standardize it and make it available to others. That rhythm has produced measurable adoption and tangible time savings, from core operations to clinical workflows.

Looking ahead, Highmark is extending Sidekick beyond chat and workbooks to secure, monitored agents that can take action within a user’s permissions. The goal is federated access where employees build or reuse agents on approved platforms and the enterprise monitors how they operate with maximum security. In short, Sidekick is evolving from an AI tool into an enabling layer for how work gets done.

Highmark Health has been a long-standing client of IIA’s and we’ve had the pleasure of supporting their data and analytics team in a wide range of initiatives through the years. In this conversation, we sit down with Julia McDowell, VP of Highmark’s Artificial Intelligence Center of Excellence, to dig into how Sidekick has evolved, what was learned along the way, and where the team is taking it next.

Give me the 30,000-foot view of Highmark’s posture toward advanced analytics and AI for competitive advantage. What’s the organization’s history, culture, and ambition here?

Julia McDowell: Historically, we were a fast follower. We watched the market, saw what was coming, and waited for someone else to test it before we made a strategic investment. Over the past 12 to 18 months, we’ve moved to more of an applied-AI stance. Our Living Health Strategy serves as the guide with AI serving as a key enabler. We want to be out in front, testing, innovating, and showing the value of this technology to support health.

That helps set the stage. Now, let’s talk about Sidekick. What problem inside Highmark were you trying to solve with it?

Julia McDowell: Honestly, it started as risk mitigation. From an information security standpoint, we didn’t want employees using open public tools given the amount of sensitive data we handle. So we shut down access. People still wanted the capability because they were using it in their personal lives and saw how helpful it could be. Sidekick began as our “okay, here’s a safe, secure option” so employees could experiment, and our developers could stay focused on the healthcare use cases that needed deeper expertise. It started there, and it has since grown into something much more powerful.

Before we get into how Sidekick has grown, I’m curious about executive sponsorship. Highmark’s history has been as a fast-follower, and you’re pushing to lead. How did sponsorship show up in practice, and what early decisions or resources gave the initiative momentum?

Julia McDowell: As with most, when generative AI hit the scene the dynamic of data scientists pushing AI to business leaders shifted overnight to a sudden increase in demand. The business turned to the data scientists and said, “Hey, tell me how I can use generative AI to solve my problems.” The expectation has always been that we must prioritize use cases where generative AI can deliver clear value. At the same time our inboxes filled up with requests from employees who wanted to try these tools for basic productivity gains.

So we adjusted. Sidekick began as a simple wrapper. Again, the goal was to create a safe, secure place for people to experiment while our team stayed focused on higher-impact use cases. Once it was in people’s hands and we saw how they used it, the conversation with senior leaders changed. They could see the impact for themselves.

From there we formalized the change effort. The C-suite set up monthly forums to discuss progress and guardrails around Sidekick and our other AI use cases, and we set a company-wide goal from the VP level and up. In short, the bottom-up energy met a top-down push, and that combination gave Sidekick real momentum.

Given Highmark’s culture and leadership support, selling Sidekick may have been easier than at some firms. But even for you, were there moments when it was hard to get buy-in? If so, how did you get it over the goal line?

Julia McDowell: Two things. First, broad productivity is hard to quantify, just like it was when email first showed up. It’s tough to imagine doing our jobs without it now, and generative AI feels similar. Because of that, we made a strategic choice: give broad access to Sidekick and don’t try to measure ROI on the tool itself, while also investing in a small set of high-priority use cases with a clear ROI. The returns from those can be reinvested to keep making Sidekick better. That two-prong approach keeps us balanced.

Second, we made Sidekick the place to start experimenting. Any time a use case pops up—whether we identify it or it comes from the workforce—we ask two questions: What’s the near-term solution and the long-term solution? Have we tried it in Sidekick, and how close does it get us to the gold standard for that use case? Does it get us to 50 percent? 70 percent? 90 percent? If Sidekick gets us there and it’s something that can scale beyond one team, we build it into the product and ship it. We release features every two weeks.

The simple example of document translation comes to mind. Someone in the business came to us with this need. Sidekick got them part of the way. We then added a true translation feature, and now anyone can translate a document with about 99 percent accuracy. It started with a problem brought to us, we tried it in Sidekick, and then we made it available to everyone.

What I’m hearing is Sidekick is this amazing sandbox for Highmark. For things that roll out to the whole company, does the central dev team still take something from “pretty good” to production, or can teams build in Sidekick and push it out themselves?

Julia McDowell: Right now, the central dev team partners with the end user for testing, and anything that needs broad scale goes through that team. Sometimes the end user can handle the testing, but for an enterprise release it routes to us.

That said, over the coming weeks we’re adding features that let people share more on their own. I will be able to create a prompt and share it with my team or with all Highmark employees. Same with Workbooks, which are similar to NotebookLM. I can connect a workbook to a data set, share that data set with the people who should have access, and they can start working with it. I can also make a workbook that is public to the organization.

The idea is to empower the workforce to share, expand, and scale good ideas without always going through the central dev team. Whether something still needs us depends on the guardrails and the level of technical know-how required to make it work at scale.

Opening it up to the business is a big deal. So for now there is still a checkpoint with the central dev team. But soon this becomes something a team in the business can build on and roll out if there is a legitimate, company-wide use case?

Julia McDowell: Exactly. The vision is federated access. You can build and use AI in your own environment with the tools and data you already have. Our job is to put strong guardrails in the system so people can do that safely. We want the company confident there is zero data leakage, no harm, and secure systems, while employees feel empowered. That is the goal: empower people as much as we can and keep building the automation, guardrails, and architecture that make it safe.

How does Highmark’s Living Health strategy, the company’s north star, inform this initiative? I guess what I’m asking is how the enterprise strategy ties into to what AI literacy means for the organization? How are you defining it?

Julia McDowell: Good question. You know, people ask me all the time, “What’s your AI strategy?” and I say, “I don’t have one.” Our business strategy is the north star. Living Health is about simple, personalized, affordable health experiences so people can manage their health. It’s not just health care; it’s health and wellness, with each person an active participant.

To do that at scale—to know you and guide you at a human level—we need a lot of data. We have millions of members, so AI plays a key role. Every strategic investment, every niche use case we choose to build, has to tie back to the member, patient, and clinician experience that supports Living Health.

A big part of that is giving the workforce a safe, secure sandbox. They’re on the front lines and should be working at the top of their license. How do we remove the drudgery that slows them down and give them space to be creative? That’s where Sidekick comes in. Not only does it build AI literacy, but it fuels the workforce with the ability to learn more about the power of generative AI and new and innovative ways we can deliver on our Living Health promise.

It sounds like adoption is high. A lot of people are in Sidekick and using it. Do any super users come to mind? Who were the early adopters?

Julia McDowell: There isn’t one profile, and that’s what’s fascinating. We just crossed 20,000 active users, which is a big milestone. Highmark has about 50,000 employees, most of them clinicians who live in the electronic medical record, so they are not our target population. Among our target users, we are close to 100 percent active.

Interestingly, although they aren’t in our target population, early super users included some frontline clinicians. A few doctors asked, “How can I respond to this patient’s email in a more empathetic tone?” One physician who talks publicly about AI was one of our first Sidekick power users.

Most of our super users fall within the core business operations and support functions. We saw super users in marketing and customer experience. HR uses it a lot. Our developers use it to read and translate complex legacy code. We still have systems on COBOL, which newer engineers may not know, so they used Sidekick to understand it and make fixes before we invested in more code-specific tools. It is truly across the board, which is what makes the story so compelling.

Hitting essentially 100% of your target users is phenomenal. And the momentum that creates… you can see and feel it across the enterprise. We’ve talked about some wins. What were the biggest hurdles or trade-offs getting Sidekick off the ground, and how did you work through them?

Julia McDowell: We were paving a new path. Our standard tech-evaluation process covers security, access reviews, architecture, all of it…and we had to align on what “good” looked like for this kind of application. Sidekick was our first all-employee application on GCP with single sign-on, so it took time to figure out the approvals and documentation.

It was slower at first, but it set the path we now use. Honestly, that’s true across a lot of what we’re doing with generative AI and, now, agents. Things are moving so fast that you’re basically creating new processes as you go. The first pass takes longer. The next ones move faster.

So what’s next? You’ve talked about taking more of the blockers down and opening Sidekick up as a truly federated tool. What else is on the horizon? How are you thinking about the roadmap?

Julia McDowell: We launched Sidekick about 18 months ago to bring generative AI to the masses in a safe, zero-data-leakage environment. Then we let people bring their own data in through workbooks and other connections. Next up is building agents that can take actions on your behalf.

For instance, you go into Sidekick, stay within the environment and data you already have access to, and create agents that act for you. You can monitor what they do, test them, and when it makes sense, give others access so they can use what you built. That’s where Sidekick is headed. Over the next six to twelve months, that is the focus.

Are you thinking about agents as general-use tools across the enterprise, or as more vertical, use-case specific helpers?

Julia McDowell: Both. Just like our approach with Sidekick specifically, we want wide access to safe, secure environments where people can experiment. We expect to stand up core agent platforms where employees can browse and use approved agents, almost like an internal app store.

We also see a place for more tailored agents that a centralized team builds. Sometimes those will be single purpose to support a specific vertical or process. Some may be more complex and core to Highmark’s business and the Living Health strategy. Other times they may start in one environment and later be deployed in our general productivity tool so more people can use them. We are still working through the technical details, but those are the pillars.

What matters most is investing in a single pane of glass. No matter where an agent is built or deployed, we need to monitor what it is doing, how agents interact with one another, and how orchestration happens across the company. Alongside that, we are putting in the right security controls so we can see how external agents interact with us and keep the environment safe.