Episode 28: Breaking First-Party Barriers with Privacy-Compliant Data Modeling 

Join Mike Julian, Principal Consultant for Population Intelligence and Activation at Definitive Healthcare, for an in-depth conversation about the transformation of healthcare data strategy in this episode of the Digital Clinic. The discussion reveals how healthcare organizations can overcome the constraints of first-party data through tokenization technology, addresses the future of healthcare marketing in an AI-driven landscape, and the critical importance of measurement frameworks that extend beyond basic performance metrics. Mike emphasizes why agencies must transition from tactical execution to strategic data consulting partnerships while delivering clear, expert insights on building privacy-compliant data strategies that enhance patient outcomes.

Listen & Subscribe:

Meet Mike Julian

Aaron Burnett: You’ve been in the same space for a long time, but you’ve also been in these very interesting, fast moving niches, and it looks to me like you’ve always been on the leading edge of what should be done, what needs to be done in terms of marketing for healthcare and med tech. Tell me about your background, your history, and how you came to be where you are now. 

Mike Julian: In healthcare, initially I was at Evariant, which was a CRM company built on Salesforce using claims and consumer data. I was VP of Healthcare Marketing when I joined that company after working in customer experience outside of healthcare. And it was like drinking from the fire hose when you first jumped into healthcare. And this was a different time than it is now. But essentially, a lot of these organizations, hospital and healthcare systems, were renting their infrastructure. They didn’t have big data warehouses. They needed all-in-one solution vendors. So at Evariant, we provided campaign management targeting, campaign support. And that company was ultimately acquired by Healthgrades shortly after in 2019, which then was sold off. Part of it became Mercury Health, which is now WebMD. And along the way during COVID, the pandemic, a lot of things changed for hospital healthcare systems. You have margin pressures. You have just everything that came along with COVID. 

Around right before the WebMD acquisition had happened, that’s when I joined Populi as an early employee. They had just come out of stealth. It was a bunch of my former colleagues from Evariant. And this was a different take on the market. So these previous solutions, these all-in-one CRM vendors, put the onus on the customer to have to hand over their first-party EHR data into a platform in a certain format. That means marketing has to go to IT to beg for resources. They have to wait six months to deploy, and they’re handing over their first-party data in order for an application to do something like predictive targeting or campaign measurement. Ultimately, the results were mixed because there’s a lot of effort that goes into this, and the market was starting to turn post COVID where rented infrastructures of buying these all-in-one package solutions were no longer the way. Marketing leadership coming from outside of healthcare, where there aren’t the rules and regulations, and they’re looking at bringing a more sophisticated way of targeting and marketing and understanding their consumers as a whole. Health hospital healthcare systems investing into data warehouses and analytics departments. So starting to do more of these functions that you would traditionally see outside of healthcare and building that functionality versus renting or buying. 

Overcoming First-Party Data Limitations

For Populi, what we do, we took a different approach. We bought all the claims data out there and tokenized it. And our whole thing was we did not want to build another application that was difficult to deploy. We wanted to provide direct access to the data and the insights and the tools that customers have known and love. So their existing data warehouse, their existing CRM, working with their existing agencies of record, and lowering the total cost of ownership and the barrier to entry with these types of solutions, but also going beyond first-party data. Traditionally, the places where I had previously worked were building predictive models for consumers using first-party data only, which is less relevant. It’s statistically limited. You only have part of the market. You can’t get the tertiary and secondary care types. If you’re using first-party data only, you’re limited to what you have. And what’s on the claim is a limited view of who that consumer is. So when we bought all the claims data and tokenized it, we started building volumes of strategic planning based analysis type products. 

But then we ultimately ended up getting the consumer data as well and tokenizing this. And this allowed us to build high performing, highly customizable models, predictive models that are based on third-party de-identified claims. So no PHI ever used in the modeling process. When you start with no PHI, it’s a fundamentally different risk profile. Security and IP organizations at these hospital health systems are happy because their dataset is not being put into a master dataset. And there’s the ease of deployment, but also the risk protection that their data’s not leaving their four walls and being co-mingled or a risk for violating HIPAA. When you use national claims data that’s de-identified where we’re a Medicare qualified entity as well, so we get the full parts A, B, and D of the Medicare file that gives us a much larger claims universe to start a modeling process with. 

What’s on the claim is limited compared to who the consumer is. I like to joke, people lie to their doctors about how much they drink and smoke all the time. And what we collect on the EHR care scenario setting and what’s in campaign response is really a limited view. There’s 8 billion people in the world roughly. There’s 5 billion smartphones, and every one of the applications on your smartphone is taking data and selling it to data brokers, and they have a more robust profile of who you are as a consumer. If you’ve ever purchased a house, TransUnion, Equifax, all these types of companies that do the check on you, they are taking your data and it’s going to the major data brokers out there. What we are able to do is take elements from the consumer universe de-identified to inform the modeling process. Build the model. Then we go back to our identifying consumer universe, which is PII only, and we’re enriching a lookalike audience score. So none of Aaron or Mike’s, no one’s procedure or diagnosis. None of our healthcare information or our browsing history is used in this modeling process, but we’re using all the digital breadcrumbs out there as well as massive scale of claims data. 

To assign a predictive risk score based on how you look against everyone that had knee surgery or had a screening or any type of specific procedure or diagnosis. This separation of duties in terms of third-party versus first-party really took off quickly. We grew rapidly, ultimately got acquired by Definitive in August 2023 after being in the market about two years. So very quick growth trajectory. Post acquisition, I’ve served as a principal consultant advising our partners and potential partners on how to use this type of data in their marketing strategy, in their segmentation strategy. And more recently, I’ve taken on a role building out alliances and channels, so putting this data in the hands of agency partners, as well as embedding it into CRMs, ad tech, DSPs and the like. I think the early experiences with these black box vendors led me to believe that the approach of direct access to data was going to be something that took off in healthcare. 

Tokenization and Privacy in Healthcare Data Modeling

Aaron Burnett: I want to ask a couple of fundamental questions because I know that what you’re doing certainly is cutting edge. It’s also something that is a frequent topic of conversation. It’s aspirational for most healthcare organizations. It’s aspirational for most agencies as well. It’s also, I think, not as well understood as people might pretend in conversation. 

I’m curious. So in this process, you are taking claims data, PII only, and you’re also taking data from the third-party consumer universe. And you’re using that for modeling purposes in order to create a model that makes best use, gets best value from that data. Don’t you need some key that creates an intersection between those two things? And how do you go through that process while still protecting privacy? 

Mike Julian: In 2018 and 19, this was out of reach for organizations to be able to do this. You have to buy all the claims, which is a massive investment. You have to hire data scientists. You need compute, and you need storage. So oftentimes when I talk to customers that are trying to embark upon this mission on their own and do it in house, they fail. But there has been a massive amount of innovation in this space. Tokenization vendors like DataVant allow us to tokenize everything on a claim or tokenize consumer data elements and create that intersection or link where we can then have everything available on the claim without exposing any PHI, as well as bring in what’s beyond the claim. Things like demo, advanced demographics, life stage groups, segments, social determinants of health (SDOH) behavior that traditionally wouldn’t be available in the modeling process. Having the scale of claims is going to make a more performant model. You can’t make a bone marrow cancer model or a living liver transplant model if you only have a handful of encounters from your EHR. But if you’re using national data and have access to all the procedure and diagnosis codes related to those types of risk-based procedures, you have a great starting point. 

So then ultimately, it’s the tokenization layer that allows these things to happen, to be able to create a link from a de-identified claim with no PHI to a de-identified consumer to hydrate what you’re modeling off of and to look for things that could be interesting to include in the modeling process. An example of this would be we built an anorexia model. Aside from having the claims data and encounter codes, when you start to look at the de-identified consumer activity related to the claim, you could see patterns like 80% of these people had a high smoking frequency, which makes sense. It curbs your appetite or had poor sleep quality factors that we might not even think about. Ultimately, that makes the model more performant. When we go back to the consumer data, which is PII, we’re assigning a score, and this really goes back to what you’re saying. Agencies may not have the ability to buy all the claims or the consumer data, but they’re really great at coming up with media planning strategies with acquisition strategies, with channel mix, with content. 

Aligning Marketing Strategy with Patient Care

Third-party consumer models that are HIPAA-compliant allow us to better understand who we’re targeting while staying in the bounds of HIPAA and ultimately getting the best eyeballs on that content and leading to no waste in the media dollars that we’re spending. Aligning risk type, “Give me everyone who’s most at risk for a certain cancer type, who also is a high responder on a channel like Facebook, LinkedIn, Twitter,” as well as any of the other elements. How do they consume care today? Ultimately, we want to get everyone the care that we need, but it’s all about taking, upleveling the marketing skill to be able to position marketing as a revenue generation hub, but ultimately put the consumer at the center of this experience. So often our care experiences are disjointed. 

There’s this idea “to know is to serve.” The more you know about the consumer as a whole, you can create a great care experience, but you can also anticipate what they’re going to need from a care scenario perspective. This can help influence the way that consumers are getting their care. Are they heavily utilizing urgent care and emergency versus having a dedicated primary care physician? Maybe there’s messaging and targeting parameters we want to use within the modeling process to be able to get a message to say, “Hey, we have availability and capacity for net new primary care.” Once they become primary care, understanding based on their risk profile, like I’m approaching 40, my risk for likelihood needing a colonoscopy might change. We’ve had our second child. There’s different care scenario settings that come with the growing family. Being able to understand beyond the claim and all of what’s going on in my life can help us personalize the approach that a health system has in terms of their content and in terms of actually getting those care needs met. This has traditionally been a heavy burden on health systems to try to do it alone due to you need people, process, and technology. You need claims. You need tokenization, STAT certification. You need the ability to do this in a way that’s compliant. You need to hire data scientists. 

Agents and AI are starting to be able to process data more quickly. But ultimately, through some of these innovations like tokenization, like AI, we’re able to now get a better understanding in a HIPAA-compliant way of who the consumer is, model off that, and apply it to net new third-party audiences, which can impact how agencies and end user health systems are reaching out top of funnel. So for me, the answer to cookies going away, which have been relied on, or the answer to more stringent HIPAA guidelines, or even states removing or broadening classification of what’s considered healthcare data is having a robust understanding from a third-party perspective of who that consumer is upon first touch. 

Aaron Burnett: Third-party platforms made it very easy to access and use their data and to feel like you were getting insight and value from that data. Their perspective was almost always attenuated, skewed to their own advantage. And it gave you ease of use, but not always clarity or true insight. But if you aggregate your own data, if you build a first-party data strategy, if you do the sorts of things that you’re talking about, you have control of your own destiny. You have much greater clarity, and you’re not subject to the whims of state by state privacy regulations or even variations in HIPAA regulations. 

First-Party vs. Third-Party Data Integration

Mike Julian: Yeah, I think it’s an excellent point. When you think about ultimate ad destinations, whether it’s Facebook, Meta, whatever you want to call it, or DSPs, they’re not going to expose the who. They want you to spend dollars on their platform to get clicks and impressions. So initially, third-party brings helps understand the exact who and what they’re at risk for, which leads to the better advertising effectiveness top of funnel. But then when you think about what I was fascinated about when I saw you present at SWAAY, which was really around agencies that are building HIPAA compliant data warehouses by signing BAAs with their partner. Now you’re hosting that first-party data, and HIPAA’s not about what you store, it’s about what you share and how you share it, which I think from an agency perspective, you’re one of the only agencies that are doing what you do. 

But there becomes an intersection. I have all this massive third-party data, and it’s great for top of funnel, but there are certain types of things that I don’t have. I don’t know which of these patients are your patients already. I don’t know which ones have been seen in the last three or six months, because that is first-party data that becomes in from the EHR. So being able to build a high performing segment that’s highly customized, that drives effectiveness is awesome, but then being able to join it in a safe and compliant way with first-party data. “Give me everyone who’s top risk factors for total joint surgery who’s commercially insured and is a Facebook or email responder who is a very active life stage group and segment who’s in a certain age and demographic already.” That’s going to be more performant. 

But then when we start to say, “How do we join this with our own first-party data,” which is of these people, remove the people who are patients or have been seen in the last three to six months. That’s where you need to be able to blend first and third-party together, either in a data warehouse or CDP, which you’re seeing the rise of these platforms in order to do more targeted top of funnel, but also that retention. A lot of agencies with good reason are focusing on net new acquisition because it’s easy to measure, but the bulk of the lion’s share of the dollars is that lifetime contribution margin that comes from the ongoing relationship that is created post first appointment, post that MRN being created. 

The best of both worlds is being able to work with tools, people, with this type of data that I have, and also be able to seamlessly join it with your own, within the context of the data warehouse to do those first and third-party types of segments, but also an even bigger opportunity. If you have a client partner that has a million patients in their EHR, again, what they know about them is a limited view of me and you because there’s all these other things out there in our lives that are on the claim. With my data set, I have an identified record with 600 attributes. The most important ones are the ones that we’ve created from a modeling process, but then there’s all the ones that come from data brokers. There’s an opportunity then to take what I have and what I call enrich or hydrate what we know about those first-party folks in the data warehouse or in a CRM through this enrichment process, which is essentially pushing a file of my data one way, taking nothing back, and for being able to match Aaron Burnett the patient to Aaron Burnett the consumer. And now we’re hydrating instead of the 25 things we know about him from EHR. I know 600 things about you. I know the additional risk factors that I couldn’t create unless I had claims in data science. I know things that are collected on the claim. This is good for reengaging and retention, but it’s even better for developing a segmentation strategy by service line. If I look at my patient mix and now I hydrate what I know about those existing patients, I can create a codex to say, “What are the top attributes that I am seeing by service line within this patient set?” I may have never known. These samples are not collected, but then these attributes that we hone in on, “Hey there, there’s a high frequency of this attribute or that attribute,” can now be used in the top of funnel segmentation process. 

So “to know is to serve.” To go beyond first-party data and hydrate and enrich creates segmentation and strategies that will ultimately propel marketing forward. But to do this, you need safeguards in place. You either need a CRM or a data warehouse that allows you to actually push and match. So either enterprise data warehouses like Snowflake, Amazon, or CRMs like Salesforce Health Cloud or anywhere that can accept a file and do the match. You need a partner that understands data strategy. So not all agency partners are going to sign BAAs. Wheelhouse does, because the way in which you approach is very different. You are thinking about the longer term strategy of going beyond performance metrics and really understanding the consumer as a whole. That’s just a bit about how I think first-party and third-party data intersect. It’s not one way or the other. Both are equally important. But how can we do better at top of funnel? Because these places where we spend our advertising dollars, they’re not beholden for us to spend less money, and therefore they keep these visitors anonymous. And marketing budgets are being continually shrunk, and there’s margin pressure. So we need to be able to smartly move and be effective with the data that we have at our disposal, which includes first-party data as well as third-party data sources. 

Organizational Readiness for Advanced Data Strategy

Aaron Burnett: From your perspective, what should healthcare organizations, what should med tech organizations be doing now to be prepared for ingesting, working with the sort of data that you can provide, being able to continue to be effective in terms of their marketing now and in the future as regulatory constraints increase, which I think will continue to be the case, and margin pressure and budget pressure continue to increase as well? 

Mike Julian: The answer really depends on where the organization is at from a maturity perspective. Are they starting from a blank slate, or are they already doing things with CRM and campaign measurement? And depending on where they are, one of the biggest things I think is you oftentimes see the various business functions within a health system or within different industries really working on different data sets. You have marketing having one view of the consumer, some of that’s informed by the EHR. You have pop intel using their own population health management solution. You have strategic planners using a different claims data set. You have community needs types of people working with another vendor. And all this creates an issue where there’s not a common data model between the various business functions. Even internally, this causes, I think, stress on IT organizations because they’re supporting feeds to different types of products. You have swell of costs because everyone’s using a different type of vendor. So I think the first thing is there needs to be an alignment outside of the business functions. Like I typically talk to marketers. Marketing has to align with care. It has to align with strategic planning. It has to align with their IT partners to understand, “How are we going to get to a common data model, and how are we going to do this in a safe, secure, and compliant way? And what are the major systems of record and reference that we’re going to go deep with?” 

The idea of buying canned applications that are black boxes that just sit on a shelf and go underutilized is a thing of the past. Big bets on “What is our common data warehouse? What is our common CRM? How are we going to action this data? How is strategic planning going to set goals based off the same data set that marketing can achieve? How is care going to triage the leads and patients that come in through marketing efforts?” All of that, I think, is organizational planning and expectation setting. 

I always say expectation is the mother of misery. I think in general, people do a bad job at that because they’re chasing a number. Getting the house in order takes cross-functional alignment internally and then moving towards a common data model with first-party and third-party data with the consumer, with a common claims data vendor as well. And then with the partners that are who are going to be strategic partners, whether it’s from a consultant perspective or from agency. A lot of times different, and as well as technology vendors and data providers like us. A lot of times systems that they have a creative partner, they have a performance partner, they have four different agency partners. 

You can be effective. I’m not saying you can’t, but if you’re trying to future proof your business and ultimately grow market share by getting a better understanding of the consumer, getting on a common data model, getting on systems of record like cloud storage and CRM, and then aligning to partners that can make the best use of this data, whether it’s on the claims consulting side, you see these like management consulting type vendors, or agency partners that can go all the way beyond collecting impression. They are willing to sign BAAs, or they partner with you in a way that goes beyond just performance metrics and they’re aligning content strategy to the segments that exist from first-party data and third-party data. Also, aligning on the measurement part as well. 

You can have the best data strategy, and you can have the best content, but if there’s not a framework to measure, then you’re going to be in difficulty showing that, “Hey, this is a journey we’re going on. Here are the results of these efforts. Here’s how it’s working,” or we’re able to adjust. To do that, again, it takes people, process, and technology. For organizations that are on the early stage, I think it’s developing cross-functional understanding, taking stock of what are the key entry points, systems and data we have in place? How are we going to align to a common data model with our IT partners? What partners are we going to allow to safely access this data? How are we going to align to our organizational goals with our strategic partners, our key bets on systems of record and reference, and ultimately to reach these acquisition marketing goals or market share goals? 

This is a multi-year journey for some organizations, and it’s not about going from step one to step four overnight. It’s about how do we uplevel the maturity model of marketing and other business lines, align the data model, then maybe taking stock. Then how do we set goals going beyond the clicks and impressions measurement to actually being able to say, “Here’s how marketing is without a doubt driving revenue,” which then opens up the door to do things that are maybe a little bit more outside the box. Some of these campaigns that you think that are more social determinant of health focused, access to transportation or healthy foods initiatives that you see that are still very important to the community. Ultimately, we want to get everyone the care we need, and we want healthy communities. To do that, you need to be having a marketing function that is running a healthy business first in terms of how they’re spending their organization’s dollars. 

Aaron Burnett: You’re absolutely right. Our experience is that this is a multi-year process, and it’s not that the technology issues or even data availability that are the hold up. It’s much more strategic alignment and change management that takes a lot of time. 

Mike Julian: I joke the problem is between the keyboard and the back of the chair. But all joking aside, we are seeing the technology innovations like with tokenization, like with the cost of compute and storage, but then it’s how do you get everyone rowing in the same boat from an organizational perspective? You told a story when I saw you speak about starting with one partner, starting small, getting buy-in, showing a win. You don’t have to do everything at once, but with small wins, that becomes a snowball to go even further and create deeper types of projects where you are doing something like standing up a data warehouse that’s HIPAA-compliant. 

Being able to access first-party data, that just alone, being able to do that from an agency perspective is a huge hurdle across. And once you’re able to get the trust of a client to build that in a way that’s HIPAA-compliant, and as an agency department be able to access this data, you’re going beyond market performance metrics and your goal and onto being able to help them with that data strategy and alignment, goal setting, as well as being able to do the measurement piece, which is the hardest. 

You also spoke about media mix modeling, which I think is fascinating. Being able to almost gamify scenarios about where we should spend our media dollars in order to get the best results, not only like results from clicks, clickthroughs, impressions, and form fills, but if you think about it in the care of how can we take a lens of empathy, understanding what our consumers are, and model our media mix to be able to get the patients or prospective patients the care that they need. I think that’s huge because it goes beyond performance metrics, and then not only modeling what the media placement should be, but ultimately when we are serving media, how are we able to account for contribution, margin, and revenue from these efforts in a way that is clear to the organization and defensible? 

Going beyond the static offline ROI reporting that I saw when I first came into healthcare every three or six months. “Here’s a report. We can’t drill into it, but we’re showing you that all this revenue came from these efforts,” and it’s very hard for your CMO stakeholder and partner to take that to their leadership team and stand behind it a hundred percent. Because as soon as it starts getting questioned, and you see there are leadership and clinician leaders that they are data driven, they are more savvy, they’re going to drill into it. So if it’s not explainable, defensible, and you can’t be able to analyze in real time, I think that it’s a detriment to the overall mission of going from stage one to the ultimate maturity model that we’re discussing here. 

Strategic Advice for Healthcare Marketers

Aaron Burnett: You know quite a lot about our approach. You understand the technology investments that we’ve made and the markets in which we focus. I’ll ask for some free advice. You also have seen a broad swath of the agency landscape, and you’ve been involved in healthcare and MedTech marketing for a long time. If you were in my seat, what would be your strategy? What additional investments would you be making, and what would you be focused on? 

Mike Julian: I think, not to be a shameless plug, there’s an actual overlap with being able to do the hydration of first-party data, like I spoke about before. Once you have your client’s first-party data stored securely in a data warehouse, the next thing to do is to fully understand those patients, and that comes through partnerships with third-party data vendors like myself. 

You could then start to model out scenarios with claims, but you’re limited to the claims data that your clients have. So that’s, again, a limited view of the market. They don’t have a hundred percent market share. Maybe you’re missing different socioeconomic groups. Maybe you’re missing a whole swath of the population. So being able to leverage national claims data, your options are invest in the data itself, which can be very expensive. Hire the data scientists, which you probably have great folks that are great with data on hand. It’s about speed, the value, and being able to have your best resources being freed up for their most meaningful work because there’s been innovations out there. There are third-party data vendors like mine who have a file of all the consumers with propensity models that are HIPAA-compliant, that can be used to then hydrate what you know from a first-party data perspective, that makes your strategy even more valuable. 

Ultimately, I think when we go to target new people, top of funnel outside of existing patients, using this new enriched data set as a key to understand, “How do we do top of funnel better?” in context to even the media mixed modeling. Now you have more robust data, so rather than hiring people and buying data to build the model, you’re freeing them up to interpret the data of, “Hey, this is what we’re seeing now. We have this enriched data set, and here’s how we should model our media mix, and here’s how we should plan our acquisition marketing to effectively target new people to grow market share.” I think this has been unobtainable for agencies, especially if they don’t have the data warehouse. For you, I think looking at partners like us, you do a lot of the compliance tracking, so ensuring that because top of funnel you can serve an ad, but then what happens when they start, and it’s this very state by state, but when people start to actually click or fill forms and show intent, now we’re starting to blur the lines of HIPAA. 

We’re going beyond third-party. So keeping that whole conversion journey safe is huge. You can build this framework out yourselves, which I think Wheelhouse has a framework already from what I understand. But there are also very interesting vendors like Ours Privacy and FreshPaint that do HIPAA-compliant journeys, which is more server side tracking and hashing type of stuff. But I think it’s interesting because being able to optimize the campaign spend based beyond the click on the full funnel of performance, it’s another differentiator that your agency and agencies that are actually able to track end-to-end in a compliant way consumer’s journey. It’s a big differentiator, and not many can achieve it. 

Continuing to lean into this approach that, “Hey, we do things compliantly. We can understand, help you understand your clients better than you do. To know is to serve. We can help you target the next best consumer on the channels of choice, but then understand their full conversion journey,” because so often we can market, and if the intake’s experience is horrible, if the call center is overwhelmed, we’re driving to a call tree, and you’re losing, you’re having drop offs. So the best marketing in the world, it’s only as good as the patient experience. And that includes the digital journey as well as to actually becoming a patient. So having a way to measure that piece is huge, and I think that’s where Wheelhouse excels based on when I saw you present and the maturity model of your customer. The free advice would be, let’s lean into looking at how we can partner together on the providing these types of propensity models to what you’re already doing. And I think starting small and testing and seeing if it can help the smart people at Wheelhouse do your jobs in a way that’s meaningful to your clients, whether it’s med tech, provider, or even pharma marketing, which is a whole other topic in itself. 

Comparing Data Providers

Aaron Burnett: There are other providers who also supply audience targeting data, audience enrichment data. How do you differ from, let’s say, a Swoop or another of those providers? 

Mike Julian: Yeah, so it’s a great question, and you’re starting to see all these vendors put out new types of marketing and product marketing collateral. When you talk about WebMD, where I worked at a company that’s now part of WebMD, they’re basing a lot of their models on first-party data only. And so again, there’s a statistical relevance piece. When you talk about Swoop, works, doing some great things. They do highly customized segments like we do highly customized segments. But in the past, when you talk about IQVIA, Crossix, these vendors, they’re starting with known procedure and diagnosis for the most part. 

So I’m trying to target, I’m trying to target 170 people who have peripheral artery disease, and they’re going to pull from the claim who those 70 people are, and they’re going to dilute that list of people with another 30 or 40 people to have a 30/70 dilution. And that’s when, like, looking at audience quality becomes important. How good is this list? Because they have to randomize and anonymize the list of what they’re starting with in order to make it usable to stay in compliance. But that’s because they’re starting with the actual known procedure and diagnosis of a cohort. And that’s worked. But there’s the problem is there’s a lot of waste from the initial list, and as an agency partner, if we’re being beholden to how we spend these dollars, starting with a list that is, as a whole cohort of people that aren’t even going to be interested in what we’re marketing for is, I think, a challenge. 

Because we’re being beholden to how we spend these dollars, is starting with the known procedure and diagnosis and having it diluted a better approach than using all of the claims data de-identified, and then pulling in consumer elements and then hydrating PII with a risk score based on the lookalike? Which one of these is going to perform better? 

I can’t say definitively because we haven’t run a study yet. Now hopefully I’ll have some agency partners that have been working with these data sets that can come up with publishable case studies to say, “Here was the uplift.” We see anywhere from three to five times uplift in performance metrics with the approach that we’re taking. I’ve had recent campaigns where we’ve seen 70% improvement in the top of funnel leading indicators. I’ve had healthcare systems speak about getting 12.6 or 13X ROI within orthopedics. 

It’s not about just the revenue. It’s the fact that they’re able to fully understand those consumers, reach and engage with them, and get them the care that they need. And that type of 13X ROI is going to get the attention of the organization. One of my client partners spoke about this at SHSMD, and then they’re able to actually then go and roll this type of approach out to other service lines or go with a bigger scale. So ultimately, the biggest difference is highly customized segmentation where we’re exposing the logic to agency partners and to end users like hospital healthcare systems. Our data lives in Google BigQuery. You have access to raw data to do hydration in your data warehouse, but we also expose the front end logic without being able to code in Tableau dashboards so we can build these highly customized segments. And do the consumer analysis as we’re applying things like clinical models or channel models or SDOH elements, we’re able to, in real time, understand what the audience size is, what the demographics look like, as opposed to having to go to a vendor and saying, “I’m looking for this,” and wait a week to get a list back. This is like a static audience or a canned segment. The performance metrics might still be there. But good is the enemy of great. And so when you actually know the who, and you can take the known procedure diagnosis out of the equation, I think it’s a great start. 

Now Swoop has this zero knowledge type of approach. And so they’re using claims data, but they’re saying, “We’re not using any of the consumer data elements.” And I think that approach is interesting, and they build highly customized segments. But I’ll say, think back to the anorexia example that I told you. When you’re able to token link de-identified claims to de-identified consumers in the initial model build process, and we score every consumer on every model, it opens the door to say, “Are there elements that aren’t on the claim that are important to identifying the cohort that we’re ultimately trying to reach net new,” such as sleep quality, smoking frequency, and any of the other 500 attributes? Then it’s combining seamlessly risk type with channel response, commercial insurance if you need it, care scenario setting. And then from there, it’s a consumption based CPM model when you’re digitally activating. So that last mile execution of digital activation typically happens in, you’re uploading a list to Meta or a DSP, or you’re choosing broad based demographic parameters and other things from the DSP. 

But with our dataset, what we’re able to do is build a segment and then seamlessly onboard it. We use LiveRamp. We’ve done a lot of work with the matching schema, and we are matching 90% plus of consumers from our universe into LiveRamp, where we get the identity resolution, and then we’re able to push to downstream destinations like any DSP that connects to LiveRamp, new advertising channels like Reddit and TikTok that might have been unreachable for your advertising partners, as well as many other destinations. So then in that case, it’s an efficiency game because not only do you have to worry about uploading, onboarding, and actioning, it’s showing up as a custom audience and allow you to do your best work. These organizations you mentioned, they’re super strong in pharma marketing, and I think that we’re going to see changes in pharma marketing over time. 

When you talk about pharma marketing, what is the call to action for the consumer? It’s “Talk to your doctor. Call your doctor.” So that’s why having the claims intelligence, understanding who’s doing what and who’s writing certain claims becomes important. Every one of these vendors targets physicians using NPI onboarding services. We have HCP targeting as well. We use that. We match 99% of our physicians, and we have great claims data to understand who’s doing what, where they’re doing it, as well as referral patterns. Referral pattern analysis is something that there, it takes some nuance. You need to understand claims. So in that case, building highly customized models for pharma and understanding what doctors are writing prescriptions, that’s how it’s always been, especially when you talk about the measurement component that comes with it. But oftentimes you need to target not only consumers, but the physicians before a patient becomes a claim. 

Think about bariatric surgery. Bariatric encounters have fallen off a cliff. Meanwhile, mental health, behavioral health is the fastest growing service line in America. Why are bariatric encounters being less and less? Because you have the rise of GLP-1s. So I can look and target physicians and say, “Who’s writing GLP-1s?” Now, if I am trying to market Mounjaro versus Wegovy versus Ozempic, getting a physician to start writing one versus the other, it’s very difficult because they’re already wed to a provider. The claim has already happened. So when life sciences, having the claims and the consumer intelligence can anticipate before a claim occurs, rather than just take claims volume alone as a way to target physicians, and there’s a lot of noise. Everyone’s using the same NPIs and targeting the same way. 

I think the two biggest advancements that I’m thinking about in pharma marketing is going beyond the physician NPI and their professional profile. “Dr. Aaron Burnett” exists as Aaron Burnett, the consumer. So I have 600 attributes on the consumer. Everyone’s targeting the doctor based on NPI and claims volume. The moment I’m able to link those things, now I can say, “How should we best engage “Dr. Aaron Burnett” based on his channel and consumption versus what channel he’s most likely to respond to,” or like recruitment use cases. I think open the door for that instead of just using claims volumes of “Dr. Burnett” to say we should serve ads because this guy’s writing a lot of GLP-1 claims codes. 

How about in the same way in which we build models, if we token link de-identified claim to de-identified consumer and roll up an aggregate so there’s no risk of re-identification, we can say physicians have consumer, have patients de-identified that have panels or co-morbidities or demographics that fit the certain patient profile of what we’re trying to sell. “This doctor sees these types of patients that have these co-morbidities. Maybe it’s weight, maybe it’s diabetes,” et cetera, et cetera. That creates a better TAM to go after the physician and go beyond just what’s written on the claim code. Ultimately, I think I’ve seen cool companies like Ada Health, which has chat bots that do qualification and can do some triage. You think about rare and orphan disease. I’ve talked to really cool companies that do really well mid and bottom of funnel, but understanding top of funnel for rare and orphan is very hard. 

You have to be able to build highly customized models and doing so in a way that’s compliant. If you have to loop out the audience you’re going after and rare and orphan disease space, you have the metrics are very difficult to have it be viable to have performance metrics that actually work. And I talk to a very large life sciences management consultancy, and a lot of times their agencies will say, “We’re doing 300% performance.” And what they’re saying actually in this context is the rare and orphan disease that we’re targeting for affects one out of a hundred people. “We found three out of a hundred.” That means there’s still 97% of effort that’s wasted. And I think so finding ways to tackle that problem is best for the agency, best for the consultancy, best for the end user. And we’re seeing traction in using this approach in clinical trials using de-identified claims data like we discussed, and understanding what consumers are most likely at risk for a certain clinical trial that we’re trying to fill. 

This is PII only that we’ve enriched. So we have the data use rights to market, and now let’s stack on channel models and things like, “How likely are they to participate in clinical trials?” Can we reduce the number of touches instead of taking a hundred thousand touches to fill a panel of 300? Can we half that? And if we can, that’s great for the pharma company and their commercialization process. That’s great for the CRO who has contracts with the pharma company to try to find these people. And then ultimately it’s great for the end user. And I know I talked a lot here, but I get excited about this. Easily direct mail, while boring, you can see the impact of this modeling in direct mail. Full stop. I have clients that have sent 40% less mailers and got over 120% results. 

Meaning instead of just spraying out direct mail pieces that are costly, when we send out way fewer pieces, but use clinical risk in context of the direct mailer, then we’re able to get better results. And there was one system, very small, they’re not even using a CRM and doing much in digital, but they always had their quarterly mailer. And for lung cancer screening, they would get anywhere from 50 to a hundred appointment requests from this. So they sent fewer mailers, used our lung cancer screening model. The results were, they got over 420 appointment requests. A lot of them were even seen in this period. And in one case, they were able to have a screening and catch lung cancer stage one. So really early on for this consumer, this makes all the difference. They’re able to get triaged to get the care they need early versus late. And that could be the matter of life and death. And so while I talk about the performance metrics and being able to target consumers, because I’m a marketer at heart, ultimately this has real impact on real people’s lives, and it can be seen in something even as antiquated as direct mail. 

AI’s Impact on the Future of Healthcare Marketing

Aaron Burnett: That’s fantastic. What have we not talked about that we should? What haven’t I asked? 

Mike Julian: I talked a lot about the way I see the world. I would like to see to understand if this aligns with what you’re seeing and when we think about how trends are going to impact what we do as digital marketers. I think the biggest thing is the rise of zero-click in search, and you have people no longer clicking, which when they’re searching for things, which a lot of, there’s a lot of revenue in marketing, and search is, having search strategy is super important. But how do you prepare yourself for the future for when people are no longer clicking and they’re getting answers from AI? This is a big problem for not necessarily us as marketers, but more so for Google itself, who’s making a ton of revenue for this. So that’s one theme. I would love to get your take on that. And then as we then delve into things like conversational ambient listening and semantic conversational platforms. How are inventions in AI going to impact everything we just talked about? And I think, I don’t have a good viewpoint today, but it’s what I’m thinking about because everything we’ve known that has changed so much in the last five to 10 years, it’s going to be disruptive and change. Again, what I do from what we do from a modeling perspective, we’ll get disrupted, and there’ll be other things that happen. So that’s what I’m thinking about. We’d love to get your thoughts on there’s any things we haven’t discussed that you see as big items to be cognizant of. And if you see those areas as big areas for you as a CEO leader of an agency today. 

Aaron Burnett: Those are some big questions. I can answer both of them probably in part, but not in whole, because there are all sorts of things that we don’t know yet. Our search strategies for our clients, and in particular, our content strategies have shifted from, and we’re really focused on enterprise grade content strategy. So things at scale, large sites, ecosystems of sites, that sort of thing, shifted from what was a conventional approach. “We need to develop a content strategy and information architecture and presentation layer for content that lives on a site, and we want to attract people to that content as part of the user journey, the conversion journey,” to “We need a content strategy that certainly includes the site, but is much more expansive.” 

The content strategy extends to where audiences are searching for this type of information. And so for us, in many instances, it’s absolutely fine that our content strategy results in inclusion in an AI application or in an AI overview. We simply want to be a part of the answer set there. And we want to create a dynamic in which we can be focused more on almost brand recovery searches. So now we’ve created an impression. We’ve started to inform people before they’ve ever even visited a site or an ecosystem of sites. They now have questions that are specific to consideration. We want now to have a content strategy that is focused on answering those questions in all of the venues in which such questions might be answered. 

That might be on the site. That might also be in Reddit. That might also be in the context of a particular AI platform or application. So we’re just thinking much more expansively. We also are shifting very quickly the metrics that we care about. Most of the metrics that previously were motherhood and apple pie for search marketing are not really so relevant anymore. They’re illusory, so we spend a lot more time on AI monitoring. So monitoring for presence and sentiment and that sort of thing. And then also deriving insights from that monitoring, which is directional with regard to content strategy and search strategy. 

I think the other thing that we’re doing, and we really, we’ve talked a bit about the fact that we have our own HIPAA-compliant data warehouse, that creates some very interesting challenges and constraints in marketing for healthcare and med tech and to an extent for pharma. Really interesting constraints where AI is concerned and the sort of advantages that AI can offer for insights, data mining. You talked about conversational search. We’re very interested in conversational business intelligence, and so we are also building an ecosystem of services that are AI enabled on top of the data warehouse because we have a unique advantage. We can do that on behalf of our clients because we’re under BAA and we have all of their data in this data warehouse. 

So more and more we view our HIPAA-compliant data warehouse as the foundation of all of the things that we need to do to be successful and to serve our clients. And that’s really our thesis. The extension of that thesis is that I think that the kind of privacy regulations that are absolutely present for healthcare and med tech now are coming to some extent for all industries over a relatively short period of time. And by becoming so good at driving exceptional performance in a data constrained, a conventionally data constrained environment in which you need to treat data, protect data, use data in a different way, that just positions us super well for what will become the reality for most, if not all industries in the near future. 

Mike Julian: Yeah, I completely agree. There’s a lot of scrutiny in pharma when it talks about price, transparency, and everything that you see in the media. I do think consumers are willing to trade off security when it comes to convenience, and when we reach a tipping point where things happen more conveniently, I’m able to do everything from whatever interface, social media, et cetera. I think then we’ll start to see maybe relaxation in security constraints file. I think it’s going to get worse. So having a foundation like you outlined is key, which I think is to your question, the first step of what you need to do alignment and have a hub, a centralized place in a common data model. Doing things like semantic conversational layer on top of a data warehouse, I think is fascinating. 

This is really cutting edge. Being able to have people talk to an AI interface, have that AI interface converted into SQL and to do the processing, to give back the results allows, that means speed to insight without having the technical capabilities. I think that is a huge direction where organizations and partners that can enable this are going to be able to get to insights and glean insights, maybe where they weren’t able to before. Ultimately, I think whether it’s third-party and first-party data, being able to understand what types of content to show to what groups of people and utilizing the different types of things like predictive models, et cetera, out there to surface the right content regardless of what’s being searched, but based on other parameters is interesting. 

Aligning segmentation strategy and content strategy to the massive amounts of data out there. So we’re showing the right thing where, regardless of where they’re being surfaced is another interesting opportunity. You’re seeing a lot of people do AI generated content, which is fine, but I think the magic isn’t generating the AI generation content. It’s exposing the right content to that segment in a way where you can anticipate what they need regardless of what their search term is, but regardless by utilizing a whole perspective of who they are. And it’s not easy to do that. You need third-party data, but you also need to be able to take anonymous person and link that in a compliant way to their third-party, PII, and then expose the right content in a channel that’s going to get them to continue to engage a little further down the funnel. So these are things I think about. I think what you outlined is awesome. Really enjoyed the opportunity to do this, Aaron, and I’m looking forward to continuing the discussion whenever time makes sense. 

Aaron Burnett: Me too. It was a great conversation. I really enjoyed it, and I appreciate all your time and insights.