In this edition of ‘C-suite Conversations,’ Drew Smith, Chief Data and Analytics Officer at OhioHealth, offers a perspective shaped by nearly two decades in the retail sector. Drawing on his experience as a data and analytics leader at IKEA and Little Caesars, Smith and OhioHealth are demonstrating how healthcare can rethink technology, data, and the patient experience through a consumer-centric lens. Drew takes a candid look at what happens when you bring outside thinking into a traditional industry, and why such a shift may be essential for the future of healthcare.

OhioHealth is a large, nonprofit healthcare system based in Columbus, Ohio. It operates 16 hospitals and over 200 urgent, primary, and specialty care sites spanning 55 Ohio counties.

Q & A With Drew

Judy Kirby: Drew, I’m really interested in your background in retail and how you’re able to look at healthcare differently from people who’ve grown up in healthcare.  

Judy: So, let’s start with the first one, technology.

Drew: It’s clear that in healthcare, technology is treated as a cost, and in other industries it’s more often seen as an investment. That mindset means systems that require more manual entry. They’re a lot less flexible and less responsive to changes. So, the technical landscape in healthcare is, to my mind, undernourished compared to other consumer-facing industries.

Judy: Okay, and the second challenge is data.

Drew: Right, and as I said, they are very interrelated. The reason you build a technical infrastructure is for the aggregation, harmonization, and normalization of data. You have all these different signals you’re getting – a loyalty club, data from Nielsen, stimulus from a website interaction, and many others. Those files typically come in different formats, and you have to spend a lot of time. If you don’t have the right system, smashing those together so we can get a complete picture of a customer. That’s why loyalty programs are so big in retail. They are basically just a data collection engine.

Judy: So, when you left retail to go into healthcare, how were you able to apply the loyalty programs? Because frankly, I don’t want to have a loyalty program where I show up at doctors’ offices more often.

Drew: Exactly. We certainly don’t want you to get sicker, so we can see you more often. We want to see you less, so there’s obviously an inversion there. This is when I first understood more clearly the term meaningful use. The way to think about it is if we’ve given you care for something, and later on you need care for another thing, how do we know that you’ll choose us?

So, meaningful use doesn’t mean that you can’t collect the data. It means that you can’t use the data to target people. You can know more about people when they interact with your system if you’ve caught that longitudinal record. So, you aren’t doing loyalty, as you say, but you log into MyChart before and after every appointment. You interact with our revenue cycle, and we know who you are. The principle still applies in healthcare, even if we don’t create a program to increase what we know about you.

Judy: And point three.

Drew: Point three is that mindset shift. So, now you’ve got this technical infrastructure, and you’re beginning to collect data. You don’t always have to know exactly why you’re collecting it. The big retailers did not inherently know everything they were going to do when they were building this longitudinal record. But as you’re collecting data, you can start to think a little differently about the people who interact with your system. Many people call this consumerism.

I remember very clearly the look on the face of the first doctor I said the word ‘consumerism’ to. I didn’t know I’d said a dirty word, but he valued patient-provider intimacy and thought consumerism cheapened it, and I understood that conceptually. But so much of how you interact with the healthcare system has less to do with the great appointment you had with your APP or your doc and more to do with whether you could get an appointment when you wanted it, whether the bill was clear and correct, and how easy it was to schedule a follow-up. That’s where the friction points lie, and that is the kind of thing you want to address with a consumerism mindset.

Judy: We hear that today’s consumers of healthcare want that Amazon experience. How are you able to give that to them with the data you have? How do you take what you did with your 17 years at IKEA and Little Caesars, and put that together to give patients that great digital experience?

Drew: Even if you are a data leader, an analytics leader within healthcare, you need partners in your digital product areas. You need to understand, for example, how your MyChart and your organization are set up. Your obligation as a data leader is to make the environment where you have positioned the data rightfully governed, so it’s safe and very flexible. So, if it is used for digital outreach or to close care gaps, it’s the same data used differently.

So, it goes back to having the right technical infrastructure, which these days is not as expensive as it once was. You have to build it with the notion that the data around patients and their interactions with your system will be useful for more than just one thing. Have you augmented that data? That’s another opportunity. For example, there are systems in most states that share where people are getting care. You can augment that data so you can figure out where you are most likely to need additional booking slots or understand where people are skipping medicine because they don’t have the financial means. So, you’re essentially creating a repository of data and a set of capabilities in your analytics that enable people to understand problems they’ve long been trying to solve. 

Judy: You were doing things at IKEA and Little Caesars before meaningful use and data really became important to healthcare. How are you able to apply that in healthcare?

Drew: EHR started to come on board in 2010, 2012, and 2014. In retail, point of sale (POS) systems and inventory systems have existed since the ’80s. So, they had a 20-year-plus head start building up this loyalty data, and they built up enough of it to understand how it worked together. So, healthcare doesn’t have to take as much time because the technology that was created to power up Amazon and Procter & Gamble’s massive machine is available in most cloud infrastructures, like Amazon, Google, or Azure, at a relatively low cost.

Retailers have been collecting data and putting different types of data together to get a clearer picture of why people choose their brand and why they drop out. That’s always possible within healthcare as well. Why do people go to your care site to have a baby, arguably one of the most important moments in their lives, but not for primary care? That’s a question you should ask.

Judy: So, how do you use your data to identify reasons and solve for that – people going elsewhere?

Drew: We call that leakage. I don’t hear that term used in healthcare because it feels cheap or tawdry. But I can give you an example where we were able to use data to understand the behavioral drivers of women who would not get a mammogram. We had a ton of data on women who were eligible for mammography, on those who got mammograms, and those who were eligible but didn’t get one. But we were able to get additional data from open sources, including an economic index in Ohio, which will identify the probability that you are under-resourced based on zip code. There are ones that do other health equity measures and demographics.

So, we profiled the data first and found there were about eight variables that were meaningful in moving people to get a mammogram, while two variables were meaningful in stopping them from getting one. If you had a primary care doc, you were going to get a mammogram. If someone in your family had breast cancer, you would get a mammogram. Some of these are not surprising, but others were. For example, we found that if you had a job, you were less likely to get a mammogram, which sounds counterintuitive. And this relates to another counterintuitive finding – distance to a mammography center didn’t seem matter.

Then we determined that most of the people who had a job but didn’t get a mammogram were low on the economic index. So, they most likely had jobs that were not flexible, like hotels and restaurants. So, that changed the way we would tackle that problem. So, instead of building a fixed asset, a new $10 million mammography center, we decided to send a mobile mammography downtown so that you can actually do scans near where people are working. And see if you can get large employers, like hotel chains, to give their employees an extra half-hour break to go get a mammogram. If we hadn’t done that analysis, what OhioHealth was going to do was build mammography centers.

So, when you get the data rich enough, and you augment it, and apply statistical principles to it, and then have a robust discussion between clinicians, operators, nurses, data scientists, and analysts, you can get sharper about the conclusion.

Judy: Healthcare has so much data. How do you work with data coming from many different sources, and tracked differently? For example, there are many different meanings for the length of stay. How do you make sense of it all and help a healthcare system bring consumerism and better health to its patients?

Drew: I think I’m better at doing the first and need to partner with people who know what good health looks like on the second. But you’ve brought us into something that is really important when you talk about length of stay, and that is data governance. Across all industries, I can say that everybody’s bad at this, but you have to get it right. We had a situation at OhioHealth where we were implementing a new predictive algorithm and found that we had four different definitions of length of stay. We simply cannot put into an AI or machine learning model four different definitions and expect a reliable output. It doesn’t work.

So, you have to get around the table and say, “Okay, why do we sometimes use length of stay starting at 12:01 on the day of admission, and then another, which is based on the exact time you’re admitted?” You find that there might be reasons. One might be a rev cycle calculation.

We had something similar at Little Caesars, too. We had four different definitions of on-time store opening. At Ikea, I had 14 different definitions of a sale. So, this is just the hard work of getting good AI, starting with getting aligned metrics, measures, terms, and definitions within your system.

Judy: When you moved into healthcare in 2023 as a chief of data and analytics officer, how far behind retail did you think the industry was? How did you help them scale and build, especially not understanding some of the nuances of healthcare that others did?

Drew: I think that the infrastructure was what was the furthest behind. It wasn’t data. Data was number three. So, infrastructure was number one, and data number three. Number two was the awareness that this longitudinal, non-episodic view, a complete picture of your patient population at an individual level has value. That has stood so long in retail that it’s intuitive. No one doesn’t believe in a customer 360 in retail, but in healthcare, they still need convincing.

People started seeing the value of having a complete picture of a patient over time through sharing some interesting stories. That created the need to make the data more complete, which we did, and to stand up more cloud platforms, which we did. So, we start with a little bit of inspiration around the types of things you can do if you have a complete picture of the patient, and then you go backwards down the pipeline from there.

Judy: Let’s talk about AI. Everybody wants big shiny things that are going to make life easier. People are afraid they’re losing their jobs. What are you seeing and what are your thoughts?

Drew: This was actually a big topic at Becker’s Annual Meeting. This notion that I can talk to a machine or train a couple of machines to talk to each other and do some of my work, that is still very much in the promise stage. People are finding simple generative AI tools like Microsoft Copilot useful, but all the studies say that it’s not returning cost savings. It is not necessarily reducing headcount. It might be reducing administrative load, making it possible for you to spend more time thinking strategically, but there’s no proof of that yet.

When you talk about agentic AI, there are a lot of people saying they’re building AI agents, but they are building them in really tiny environments. There’s a big difference between a Geek Squad doing a few agents for themselves and scaling that out to a healthcare system of 35,000 people. That still hasn’t happened. We’re still a ways away from how that will derive value. But there are many conceptual use cases in healthcare where agents would be extremely helpful. Think of anything at a healthcare provider where someone does the same thing over and over. You can train an agent to do that: coding, some admission stuff.

There are many useful applications of AI that aren’t a big shiny thing. Machine learning, which is classic AI that can be trained to anticipate patients’ decline as Epic has in their deterioration index, is a valuable model. That is also AI. The prediction for sepsis is also AI. The need for Palliative care can be anticipated with AI. I think healthcare just hasn’t necessarily seen that as an economic gain. That can be really frustrating because you have to spend money to get it to work, and you might have a CFO’s office saying, “If you’re going to spend a million dollars to build this thing, you’ve got to give me five 5 million in return.” The reality is that’s a very hard promise to deliver on right now.

Judy: Have you seen any agents or any type of AI deliver ROI at OhioHealth?

Drew: I try to avoid the term ROI because in healthcare, I think we need to measure value more than ROI. The classic example I would give is ambient listening. Almost everybody has ambient listening these days, and there are systems claiming they’re booking more appointments as a result. When we started that, our ambition was threefold. One: make it easier to be a provider because the competition for doctors, APPs, and nurses is fierce. We have great ones, and we want to keep them, so it became sort of a defensive play. Two is that patients appreciate being able to look at their doctor. From our consumerism mindset, we know you are more likely to stay with OhioHealth because you have eye contact with your doctor. Ipso facto, this is worthwhile. And third, it’s harder and harder to be a primary care doctor because your patient population is getting older and has more comorbidities. It’s hard to keep track of all that. But ambient listening separates things a little better than people can, and can listen to a patient talk about their condition of diabetes, and then say something which might be relevant to their heart condition. Doctors have said anecdotally, “It caught things I probably would have missed.” So, happier docs, happier patients, and a reduction of problems for people who have comorbidities. These don’t necessarily return cash into the system, but they are valuable things.

Judy: What about other AI that people are putting in while trying to generate returns to satisfy their CFO and CEO?

Drew: Especially in large health systems, you have an enormous supply chain. You have an enormous HR function. You have an enormous FP&A function, rev cycle function. So, if you want to get ROI out of AI, you can do that in those back office functions, which is what retailers have been doing for an eternity.

Judy: Coming from outside of healthcare, did you find resistance to your ideas and strategies?

Drew: No, I didn’t, but I thought I would. The exception would be the reaction from that physician when I use the term ‘consumerism’. I didn’t get resistance, but I did have to work at translating why I thought this could be done in healthcare and how it had been done in other industries. I certainly didn’t get a red carpet rolled out because I had saved $3 billion in supply chain at IKEA, and increased conversion at Little Caesar. There was negotiation, and there was engagement. There was discussion.

Judy: OhioHealth is a fairly large system comparatively. What can smaller systems with financial limitations extrapolate from what you’ve been able to accomplish?

Drew: If I were at a cost-constrained system, I would develop cross-functional capability to maximize the proven AI models coming out of Epic. I’m not talking about their latest gen AI model. I’m talking about using the models that they already have for claims denial, for sepsis. You can reduce the burden on your staff with those models. You can capture a lot of the value in the rev cycle. So, take a fast follower approach, get people in the room, your Epic specialist, your rev cycle experts, your nursing experts, and make them a pod to launch these models quickly and widely in your system so that you get the scale.