From Data to the Front Lines of Care: The Recipe for AI in Production with Domo + AWS

Laura: Welcome, everyone. Thanks for joining us today. We're so excited to have you with us for From Data to the Frontlines of Care: The Recipe for AI in Production. Healthcare organizations have invested significantly in data and AI in the last few years. But the real challenge isn't just building AI, it's turning that investment into measurable outcomes and getting AI into the hands of people who can actually use it. Today, we're going to show you what that looks like in practice.

Laura: We'll hear from Regional One Health Solutions about how they built agentic solutions on Domo, powered by AWS, to address real healthcare challenges, and how one of those agents identified more than $5 million in revenue opportunity in just three months. We'll walk through the recipe behind that success. What does it actually look like to architect these solutions on Domo and AWS? And ultimately, how do you put those insights directly into the workflows of clinical, financial, and operational teams?

Laura: Now, before we jump in, I want to take a moment to introduce our incredible speakers for today's session. So first up, we're going to hear from Jannie Rad. She's the CEO of Regional One Health Solutions. And then we'll also hear from Aman Tiwari, who is a senior solutions architect at AWS. We have such a wealth of knowledge between these two experts. We're lucky to have them joining us for today's session. Also, feel free, as we're going along, to drop questions into the chat. We've got some Domo folks and Regional One folks on the call today, and AWS, who can answer those in real time, and we can also cover them during a live Q&A at the end of the session.

Laura: So Jannie, if you want to join. And as she gets set up, I want to direct you to the chat section of the screen. There's going to be a poll that pops up, and we're just curious who is in the room with us today. Let's see. Yeah, where's everyone from? Okay, upstate New York. Hi, Alexis. And let's see. Okay, looks like we got quite a mix between healthcare providers and those in clinical, maybe some administration. Well, we're happy to have you all here. So Jannie, I'll let you take it away when you're ready.

Laura: Jannie, looks like we're having a little bit of an issue with your audio. Let's just... One second. Bear with us. Thank you. Let's see. Atlanta, Georgia, Memphis, Tennessee, Des Moines, Huntsville. Thanks for joining us.

Laura: Okay. Well, while we're working that out, I'll introduce myself. I'm the host of today's session. My name is Laura Quealy. I've actually been with Domo about 11 years, which is a long time. So I've been kind of on the front lines of watching us roll out these solutions with AWS and enhance our AI features and start to target more vertical-focused solutions, which we're going to talk through today.

Laura: Regional One Health Solutions is a fantastic partner of ours that has been building and rolling these solutions out into the wild and seeing really impactful returns. And we'll be so thrilled to have Jannie walk us through those solutions today. Let's see. Aman, do you want to introduce yourself as well?

Aman: Sure. Nice to meet you, everyone. My name is Aman, and I'm a senior solutions architect at AWS. I've been with AWS for six years now. So Domo is one of my customers that I work with, and I'm very excited for this session today to talk about how Domo and AWS, along with Regional One Health as one of the customers, have been doing some great work in the field of AI. It's nice to meet you, everyone. I'm very excited.

Jannie: Am I unmuted?

Laura: Yes, you are.

Jannie: Amazing. Thank you. Sorry, everyone. It's funny when you're doing a presentation on technology and then you have issues with technology. Well, we're going to try sharing the screen one more time. Thanks, everyone, for your patience.

Laura: Sounds great. And again, a reminder, feel free to put questions in the chat. This can be really interactive. We can answer those as we go along. And we're going to hop off stage and let Jannie take it away.

Jannie: All right. Let's see if this works this time. How are we doing?

Laura: Looks pretty good this time.

Jannie: Awesome. Well, I was listening to all of the introductions, and it's always great to see who's in the room. So I'll give everybody a little bit of background to who I am. I actually currently live in Asheville, North Carolina, but I still work for my wonderful hospital that is in Memphis, Tennessee. In terms of my role, I started off as an analyst for our Center for Population Health at Regional One Health, then I moved into leading our internal analytics department, and now I lead our analytics consulting arm.

Jannie: But at the end of the day, I am part of this group of people that we like to say help the people who help the people. And it looks like there are some of you in the room today that also do that. But before I dive in, I want to set the stage and the context to give you some background on what is this environment that we were implementing data analytics solutions at.

Jannie: Regional One Health is a level one trauma center, and it's also a safety net hospital in Memphis, and we serve patients across six states. And nine years ago, we entered our data journey. And at that time, probably like many of you, either in the past or now, our data lived in silos. Nobody really fully trusted it, and our frontline staff barely saw it. Or if they did, it's data that they would kind of have to pull themselves, or if someone did pull it, they were seeing it months late, so really too late for it to even be actionable.

Jannie: And oftentimes, that was the same for our leadership, too. The data would take so long to pull together reports that you would see it one month, and you would go, "This looks great. How soon can I see this?" And it's like, "Well, it took me a month to put it together, so you'll see it again next month." And that was not helpful. It was not actionable. And that is the starting point for everything that you're going to see today.

Jannie: So I'm going to start off by speaking about one of our earliest wins, and then we're going to pivot to where we're at today in AI solutions. So this very early use case for us was around encouraging our patients to fill prescriptions in our own system. Because as it was, people were receiving their care at Regional One, but oftentimes getting their prescriptions filled at competitors like CVS and Walgreens. And so we were leaving a lot of revenue on the table.

Jannie: We had the data, but it was kind of locked up in all of our systems. We had goals, but there was not really a clear ownership, and a lot of times the goals were set, but nobody could truly reach them. They weren't realistic or attainable. So we decided to bring in Domo and lean on the platform, not necessarily to build fancier reporting, but we wanted to unify all that data and align teams around one path forward and have everybody look at one thing.

Jannie: This project, if successful, we knew would have a big ROI, which is one of the reasons we wanted to focus a lot of effort onto it. But not just financially. This was going to have, if it worked, financial benefits and also our patient experience would improve and our provider experience would improve.

Jannie: So here is what we built. As you can see, it's not really anything fancy, right? There's no AI, there's no predictions. Honestly, it's really just bar graphs, right? But what made this special and unique at this time was that we were pulling in all kinds of data from all kinds of systems. We had our prescriptions, our prescribers, our patients' feedback all in one place, and it was refreshed daily.

Jannie: Then we went a step further. We operationalized it. We added stair-step monthly goals specific to each clinic, and then there were regular accountability check-ins that were tied to this data. So we used this data, but more importantly, we leveraged it to make space to ask why. What's working? What's not working? Where are the gaps?

Jannie: And different people engage differently. If you are a provider or a clinic leader, you might have engaged with it maybe mostly looking at the dashboards, clicking in, looking and seeing what prescribers have what patterns. But then if you are on the leadership administration level, you're probably looking more at email summaries, just kind of getting high-level updates. So we were able to meet people where they were at with this early use case.

Jannie: And as simple as this was, in that first year, we more than doubled our capture rate. And over a few years, that was equivalent to adding $14 million to our bottom line. Plus, what we foresaw happened. There was better provider-patient communication, better follow-up, and we actually then added new pharmacy services. This tool really showed us an opportunity. What happens when you have champions across departments that are able to turn the data into actual results?

Jannie: And this project fast-forwarded our track. It proved really what was possible when you have data and people and process that lines up. It also shifted how our teams internally related to data. But also a fun benefit, it shifted our reputation externally. Because from just being kind of known as this safety net trauma center in Memphis, Tennessee, we became known as a place that leaned into data-driven innovation.

Jannie: So much so that we took one of our leadership executive applications, the video that you're seeing here, and it was our first app that we officially built and put on the Domo marketplace, that put all of the clinical, financial, and operational metrics in one view, and it was kind of like this executive healthcare starter app. And our leaders started to share about this work nationally.

Jannie: So where are we today? Now we have gone past fancy bar graphs, and we are now leaning into AI solutions, all built on Domo. So this is the very first one we built. It's called PX Engine. And essentially, our patient surveys were being administered by a vendor, and that data was on that vendor's dashboard. Now, we did have access to some reports, and some of that data was coming into Domo, but it still just wasn't detailed enough. Teams were still having to dig through mountains of data to find exactly where that real opportunity was.

Jannie: So we built this, which does two things. There's a sentiment analysis on what patients are actually saying, and then a statistical analysis of what is actually moving the needle on the key question, rate your hospital from one to 10. For those of you who don't know, it's a core question that's related to HCAHPS and essentially CMS funding. So it's really important that people rate your hospital a nine to 10. And that combination gives every unit their top two personalized, statistically driven drivers. Plus, AI-generated unit-level interventions that the nurses can actually act upon.

Jannie: So right now, this tool is live in six of our inpatient units. We have over 20 nurses using it, and organizationally wide, we've identified that our top two drivers are around pain management and discharge communication. But again, every unit gets what's specific to them. The nurses, they can select and adapt from these AI-generated insights, and they can document their action plans directly in this tool. So they don't have to go to the vendor dashboard. They don't have to pull up multiple things. Everything is right here in this one place.

Jannie: One example that's really cool is one of these AI-generated ideas for one of our units was so good that the nurses actually decided to roll it out everywhere. And in this one KPI metric around call light responsiveness, we are now in the top percentile, starting from the lowest. So a huge improvement just right there. And in terms of users, just last week, I got a request to expand to the rest of our inpatient units and expanding this to our outpatient centers. So people are using it, and they are loving it.

Jannie: So after PX Engine, we just decided, let's keep building on our success. And I actually showcased this tool at our board meeting, and our CFO was in the room and asked, "Can we do something for my team? Can we do something for finance?" And so the next day, we met with the finance teams, and what I'm about to describe, the ideation to go live was in 30 business days.

Jannie: Not only that, but it was during November, December, and January. So I don't know about you all, but usually nothing really happens during those months. So the fact that we were able to do this in 30 business days during those months spoke volumes.

Jannie: So what did we do? We started by asking the teams that were actually experiencing the problems: what is broken and what do you need fixed? Here's what we heard. Deadlines were being missed, not because people weren't doing the work, but there was just so much detail to have to sift through that accounts were kind of falling through the cracks and deadlines were falling through the cracks.

Jannie: There was also a blind spot on small variances tied to denials. Essentially, if a variance was so small, it was not passing the threshold that the vendor had marked to pass that information along to us. But what our teams knew is that, well, if you have 100 of those variances that are all equal, it's probably a system-level pattern that needs to be addressed. So none of that was happening before.

Jannie: In addition, there was not really a huge workflow, no prioritization, not a lot of consistency. There was tracking happening, but it was kind of spread out over different people's emails. Someone was tracking on a spreadsheet, someone had a Word doc, and it really just varied person to person.

Jannie: So they defined a scope. They wanted to make data visible, make patterns obvious, and help teams prioritize. So the goal of Smart Rev, defined on day one, was they wanted to reduce avoidable denials, preventable claims issues like documentation, coding, and eligibility errors, because they felt that faster resolution meant faster payment. So that was the scope and vision statement that we started with.

Jannie: And here is what we delivered. Within just days, we had a working prototype. It starts with an executive view where you can see those statistics, like how much revenue in real time are you bringing in. Our CFO loved that. And also what's being flagged as potential because there's an appeal in process.

Jannie: Now, where that old approach was the vendor handing over that list of the flagged accounts by their prioritization rules, now we're able to define those priorities. There's not a threshold. They're able to actually just dig into the data themselves, but in a very organized way. What does that mean? It gives more raw-level, row-level evidence to our teams so they can see the errors, and it gives them more that they can actually negotiate with payers and also point out all of these systematic errors. And we're bringing in rules from external data sources.

Jannie: So you're getting AI-generated recommendations on how to support your appeals process, how to correct accounts, and how to improve documentation to prevent future denials. So not only are we tackling the appeals process, but we're just trying to prevent it more upstream. The notes live right alongside the data. There's no more jumping between emails and spreadsheets and Word documents to track conversations. Everyone's able to follow the workflow in one place.

Jannie: And again, I want to emphasize this. This was built from the ground up from what the end users actually said they needed. They wanted unified claim visibility. They wanted automated trend analysis, denial intelligence with next steps, appeals tracking to see financial impact, and they wanted to be able to prioritize, and we were able to deliver. It's not a reporting tool. This whole thing recovers revenue and prevents future loss.

Jannie: And of course, the outcomes, right? Well, our recovery rate went from 19% to 53%. This is equivalent to, as of today, $1.6 million has been recovered at the patient level. Now, speaking of the potential with appeals in process and negotiations and conversations in process with payers, we've identified another $6 million at that patient level in opportunity and—

Laura: Okay. Looks like we're having a little bit of technical difficulties with Jannie. We'll invite her back as soon as she's ready. So cool to see the evolution of those solutions into production and the actual impact they're having with real healthcare networks.

Laura: Let me see where we were at on our agenda here. So I think while she's working on that, I'll skip ahead to my section, and we can go back and wrap up Jannie's last thoughts on their solution and kind of the impact that it's had for Regional One Health Solutions across the board.

Laura: So let's jump in and talk about the recipe. And this is basically at its core, Domo helps you go from raw data and AI to actual agentic outcomes, and that's all running on your cloud using your inference. So you can see, kind of regardless of where your data strategy is at the moment. Let's say you're centralized on Snowflake. That's fantastic. We can layer on top of that and help you get your AI-ready data foundation built, activate that, and distribute it out to those who actually need it.

Laura: There's one piece that I want to layer in as its own component here, though, and that is governance. And we all know in healthcare, governance is so incredibly important, especially when you're working with sensitive data around patients or any part of the hospital process. We know how important that is to have.

Laura: So an AI-ready data foundation is what takes data from being raw and actually being useful. Activation is how AI interacts with governed data. Distribution is how the value actually reaches people. And of course, I mentioned before, we do all three on top of your cloud and using your preferred inference. Now, none of those solutions, like I mentioned, can be productionized if they aren't built with the proper governance and security.

Laura: So Domo hosts a plethora of enterprise-grade governance features, which allow you to see what is actually going on, who is building what and connecting to what, and have a kind of a comprehensive view of governance across the platform.

Laura: Now, I'd love to hear you all chime in in the chat, too. As you look at this visual, where are you and your company at? Have you brought an agentic solution end to end using each of these steps? Are you maybe stuck in one of them? If we look left to right here, your AI-ready data foundation is going to be the first and most important step. We all know our AI is only as good as the data it has access to.

Laura: So maybe you've moved past that. You've centralized and cleaned your data, and you're in the building phase, but not sure how to actually get your solution out to the front lines. And that's where the activation and driving of business outcomes really comes into play. And in Domo, that can be delivered in many ways, which I'll walk through here in a second. I think it's always helpful to see a sample architecture. What does it actually look like to build something like this?

Laura: So this is a sample architecture, and I want to point out, too, the recipe for building these AI solutions end to end is really the same regardless of industry or use case or vertical. Having an AI-ready data foundation is fundamental regardless of what you're building towards. So let's talk through this. This is an actual example of something that we've built and are using live for Domo internally right now, the sales operations agent.

Laura: On the left side, you can see we've got our data sources. Domo has a fantastic integration layer. So we have connectors, a thousand plus, on-prem, basically the way to connect to any data source and push that directly into your cloud. So like I mentioned, if you centralize your data strategy on a cloud, AWS or Snowflake or Databricks, we can actually push that data. And this is a no-copy data model. So the data actually doesn't go to Domo. We make a copy and then push it into your environment. It's actually going directly into your warehouse.

Laura: From there, as Jannie will show in her solution as well, we have a fantastic ETL layer. So we have a drag and drop tool called Magic ETL. It's kind of a fan favorite of a lot of our users, and it's a way for you to clean your data, combine multiple data sources together in a really friendly environment where you don't have to know a ton of code. You just have to know a bit about your data to get started.

Laura: From there, I also want to call out, Domo is built on AWS technology. So our native LLM is Amazon Bedrock. So if you're using Domo GPT or any of our Domo AI features, that is AWS technology under the hood, and that makes all of our features so much more powerful.

Laura: So from there, let's say we centralized our data in our cloud warehouse, and we're running an LLM Amazon Bedrock across it. Agent Catalyst is sort of the umbrella term for all of our AI features, and that can include automated workflows, an integrated chat feature, and so much more: Code Engine, our MCP, a document library for your agents to access. It's really comprehensive.

Laura: And then from there, you can see in this example, so we run our data through Bedrock, and that's multiple systems, of course, to surface milestone tracking, surface risks before you even see them on opportunities or accounts or projects. And then even the Deal Coach Assistant, where you can interact live with an agent to help you prepare for your next call or next meeting.

Laura: And then we talk about that last mile piece. So that is getting actually into the hands of the people that need to use it, and that could be via low-code, no-code, pro-code apps. It could be a mobile app. We can even embed a solution where your users are at, so they don't have to exit any of their core systems to use these insights and make their business outcomes more positive. So that is the gist when it comes to building on Domo. I'm going to stop share and just invite Jannie back to wrap up her section.

Laura: Hello, welcome back.

Jannie: Thank you.

Jannie: Well, thanks again for everyone's flexibility and patience here. Luckily, I was almost done. But I will say going back to Smart Rev, I did talk about the system-level opportunities. But the other thing I did want to speak about was we're already expanding just within our own hospital, right?

Jannie: We started with the team that was focusing on claims and denials, but right after that, we expanded to our pharmacy team, our patient financial services team. We have three other departments that have requested this, and the reason I like to speak on this is that what this has uncovered is that these types of problems are not in a silo. Everyone has recognized their piece of the puzzle, and so this has actually broken down silos to the point where we're trying to bring in people now all the way from the registration teams, the minute somebody walks into the hospital.

Jannie: So another outcome that I love to share is the workflow and some of those soft outcomes that you can't really measure numerically, but I think are just as important.

Jannie: I know Laura started to speak a little bit about behind the hood, but I wanted to show you all a screenshot from what my analysts are actually seeing when they build these tools. So Domo does have a low-code, no-code, drag-and-drop SQL called Magic ETL. It truly is magical. Now, if your teams do know coding, there is that side of Domo as well. But our first foray into making these tools was to rely on Magic ETL.

Jannie: You can do all of the cleaning of the data, you can add your formulas. So if they have spreadsheets and they already have all their formulas locked in, you can just put that in here. So anytime that data updates, all of those calculations will be automated. But this text generation tile directly within the Magic ETL, that's where you can add your prompts. So all that sentiment analysis that I spoke about, that was done right here. So, the team and colleagues will go into more of that behind the scenes, but just wanted to show you what our team actually used to build this.

Jannie: And again, just to speak about my favorite outcomes, right? These are not what you see on dashboards, but these are actually photos from our actual staff. Top left is our hospital CEO. Down below, we're actually celebrating the fact that we've achieved our best Leapfrog score. So that's a hospital rating. We actually improved that two whole letter grades.

Jannie: And also, since using Domo, we have had the lowest acute care length of stay in our hospital's history. And we are the oldest hospital in Tennessee, so it's kind of a big deal. So, these are outcomes that we are celebrating. We have broken down silos. You can see in these photos, we have a mix of people in scrubs, people in suits. It truly has become so collaborative.

Jannie: So it has done a 180 from that culture that I talked about at the beginning, where there's not a lot of trust and no one really knows what's going on. Now, it's so collaborative. People are asking us constantly to build more projects, more dashboards, more apps, more AI-based apps. We are getting requests every single day.

Jannie: One thing that I did want to speak on as well, which we can do more of a Q&A in the end, but like I mentioned, we've been doing this for a while. We've got hundreds of use cases, and I think what we have decided and realized that our bread and butter is, is not necessarily the fact that we can build hundreds of solutions on Domo, but we know what works. We know out of those hundreds of use cases, what people will just keep on their computer and never look at again, versus what is actually being used and driving towards results.

Jannie: So that is truly our expertise. Again, we can speak healthcare because we are healthcare. My role now is more shifted towards helping others in healthcare and taking everything that we've done and packaging them together and bringing them to you, but still customizing them to how your providers and your nurses and your doctors and your administration work. So we can very much bridge that gap, and happy to answer more during Q&A, but I wanted to just pop in and say that. So thank you, and Laura, I'll hand it back over to you.

Laura: Yeah. Quick question before you drop. So one, I think it was so interesting to see the evolution and even seeing impact on as basic of a dashboard as you first showed, right? Just cards, being able to bring those data sources into one place and actually see your data and make decisions on it. And it had an impact even though it was in its simplest form.

Laura: Quick question. So we may have some folks here who are just starting to build an agentic solution for the very first time. What are some things that they should keep in mind as they start to scope their use case?

Jannie: Yeah, absolutely. I think hindsight is 2020. I think lesson learned is really focus on governance and data definition. I think if we had done some of that in the beginning, we probably could have taken even shorter to do PX Engine. I mentioned Smart Rev took us about 30 days. PX Engine took us about five months to get live, and that's because there were still a lot of questions that we were having to answer.

Jannie: So if you go in right with a strategy of "this is the data we trust, these are the metrics we look at, this is what each of these things mean, and this is where the data source comes from." Having all that defined up front, and I think what they call making your data AI ready, and I know Domo actually has some great checklists that you can follow. If you can do that on day one, it'll make your life so much easier.

Laura: Fantastic. Great insight. Okay. Thank you, Jannie. I'd like to invite Aman up to join us. All right. Aman's going to talk to us about governance and architecting on AWS, and I'll drop and let you take it away.

Aman: Yeah, that was a great story, Jannie, and so excited to meet everyone and now talk about the infrastructure that powers Domo, and it also powers the stories like Jannie had. So what I'm going to talk about is a few things here. So we will set the stage with our managed service. We call it as Amazon Bedrock.

Aman: Then we will talk about the security and privacy model of this particular service and why Domo has chosen Amazon Bedrock as its inference backend, and how making that decision has resulted in better customer stories and better customer outcomes. So we will talk about what is Amazon Bedrock, the value proposition of Amazon Bedrock, and the security, privacy, and governance model of Amazon Bedrock. So let me move to this slide.

Aman: Amazon Bedrock is our fully managed service that provides customers access to leading foundation models from different model providers. As of today, if you sign up for a Domo account, and if you want to choose their AI functionalities, Domo has an AI service layer. And if you want to start using the functionalities like Domo AI Chat or their other AI products to get insights on top of your data, there are two options that Domo provides.

Aman: The first option is that they manage the inference layer for you, and they call it as Domo GPT, which is the Anthropic models hosted via Amazon Bedrock. But at the same time, Domo also provides customers the choice to bring their own models. So you can connect your models with the data that is sitting inside Domo.

Aman: The main value proposition here is that Amazon Bedrock and Domo provide you model choice so that you don't get locked in with a single model provider. So via the single API, you can access models from different model providers because, based on the use case, the decision to choose the model may vary. And we believe at AWS that no single model is going to rule the entire world. There will be different models used by different organizations for different use cases.

Aman: So Amazon Bedrock is a fully managed service that provides you access to these foundation models from different model providers. And some of the famous model providers, as you can see, are Anthropic, OpenAI, but this list is not exhaustive. And as of today, we have more than 100 models spanning from commercial models to the open-source models as well.

Aman: Now I would like to talk about the security and compliance of Amazon Bedrock. So when you use Amazon Bedrock as an inference layer, we don't use your data to train our base models. In fact, we don't even have access to your data because how this framework works is that we take a deep copy of a model, and we host that model on your behalf in our secure infrastructure. And these model deployment accounts are only governed and accessed by AWS. So not even the model provider has access to your data and the API calls that you're making for the inference.

Aman: Now, when it comes to security, there are two layers of security. So one is the data in transit, which is encrypted by AWS, and the minimum standard is TLS 1.2, but we recommend TLS 1.3. And then is the data at rest encryption. Now, we encrypt the data using our key management service, but customers can also bring their own encryption keys, and you can govern the keys on top of your data.

Aman: Now, when it comes to compliance, Amazon Bedrock is a HIPAA-eligible service, and there are other compliance certifications that Amazon Bedrock has gone through. For example, SOC 1, SOC 2, FedRAMP authorized type.

Aman: Now, this is where I would love to call Jannie to get her point of view. But our service is really built for the regulated industries, as Bedrock is in scope for common compliance standards, as you can see, but at the same time, it's also HIPAA eligible. Like I said, your content is not used to improve the base models, so your content, the inputs and outputs, are never shared by the model provider.

Aman: When it comes to the data residency requirement, I will go in depth of how Amazon Bedrock has a global view of the world. But your customer data, unless and until you have configured explicitly, remains in that particular region, helping you to meet the data residency requirements. So Jannie, I would love to get your point of view of how Regional One Health has been thinking about the regulatory and compliance requirements and the HIPAA eligibility when thinking about using AI within your own organization.

Jannie: Yeah. Great question. And I'll be honest, we're kind of in it right now. We just started an organizational-wide security governance committee. We are constantly looking at different regulatory requirements that are, I think, also being created at the same time. And then for anything that doesn't already have a national standard, we are creating what works in our own hospital right now.

Jannie: So I'll be honest, for some things, there probably aren't defined guardrails, and we won't really know until we start playing and giving access to more of our users. Self-service is where the game is going with all of the data now getting AI ready and everyone now being able to access, for example, AI Chat. You want people to be able to use it.

Jannie: And so I think what we're probably in the process of doing is slowly rolling out and giving little bits of access to different end users based on their roles and responsibility, letting people play a little, knowing that we're taking maybe a little bit of a risk and some things might break here and there. But based on some early users and pilot users, we'll be able to figure out how to put more guardrails so we can fully expand it to our teams. But that's work that we're doing right now.

Aman: Yeah. And what I would like everyone to think about is how the model choice really also affects the other big concerns that organizations today have. And one of the biggest concerns is token economics. If you're thinking to use the AI models, I would definitely encourage you to think about: is this model the right model for my use case? How do the input and output tokens of this model work? Is this a big model or a small model?

Aman: So definitely start with the use case and then ask yourself which model might be the most appropriate model. Because with AWS and Domo, you get the model choice, so you are not locked into a single model provider with big commitments.

Aman: Now, this is where I would love to talk about the global presence of Amazon Bedrock. So AWS has a global infrastructure. So based on your customer base and based on where your organization is, Amazon Bedrock is available in those regions. Some of those regions are commercial regions, other regions are air-gapped, like AWS GovCloud regions.

Aman: So definitely, go to our AWS public documentation for where Amazon Bedrock is available, where those particular models are available in those regions. And based on your use case and your customer base, you can choose to deploy those models in those regions.

Aman: Now, here I would love to go a little bit deeper into, once you start using your AI functionalities with Domo, what actually happens under the hood. So I'm not going to go at a very deeper level into the architecture, but at a high level so you can understand how your data has been used and what actually happens. So we call this architecture as a compliance and data residency architecture of Domo on AWS.

Aman: Domo is a software that you use—a business intelligence software—and Domo runs on AWS. What it means is that their application servers, and in general, the application plane, is running on AWS, and your data is in Domo. Now, since Domo is running on AWS and Amazon Bedrock, the models that we host are also hosted in AWS. What this architecture is talking about is your data and the model are co-located.

Aman: And this is where you really want to be from a technology perspective—your data and the model should be as close as possible. And because of this, you can see how it helps your organizations to meet their data residency requirement, which ties back to HIPAA eligibility as well.

Aman: So you can see that with Bedrock, Domo has access to different model providers. They are able to pass this benefit to you, Mr. Customers. And at the same time, you can see how the data never leaves the AWS perimeters. Domo is able to interact with the models on behalf of their customers privately using one of our services known as AWS PrivateLink that routes the traffic through our AWS backbone network rather than sending the traffic over the public internet.

Aman: So at this point, I would love to pass it now to Laura, and we can get into the Q&A if anyone has any questions. Happy to answer.

Laura: Fantastic. Thank you. Yes. If you have any questions, please drop those in the chat. Love to cover those. And in the meantime, we had quite a few submissions with our sign-up. So let's start with one of those. So first up, for organizations that are still experimenting with AI, what are the biggest steps to actually getting an AI solution into production? Maybe Jannie, do you want to start and answer that, and Aman, if you have anything to add.

Jannie: Yeah. I obviously talked about getting your AI ready. I do think that's a really big step. The other thing that we've been slowly navigating is realizing that you couldn't technically build an AI solution for everything. So you might need to build some type of funnel to help prioritize.

Jannie: That's kind of the process that we're in now of knowing, okay, well, now that everyone is seeing how exciting all this technology is, doesn't mean that we need to build out and churn out an AI-based solution for every single thing. So now we're kind of doing the work to say, "Okay, what's the biggest value add? What's tied to our strategic goals? Which is going to touch the most amount of people and actually make their workflows better?"

Jannie: So now after having those conversations, we're able to really focus and say, "Okay, these are the AI solutions that we're going to build out." So I would say those are my top two tips.

Laura: Yep. Anything to add, Aman?

Aman: Yeah. So I would definitely start defining: is this an internal use case or a customer-facing use case? Because that changes practically everything. I have experience working with customers where they have jumped to a customer-facing use case, and that really just gives me a little bit of a panic attack. So start with that internal use case and get your organization very comfortable with the adoption. Test heavily, I would say.

Aman: But definitely, it depends on what use case you have. Right now, I think one of the simple use cases that I've been seeing a lot of customers trying are internal generative AI applications that are using retrieval-augmented generation for search purposes over the knowledge base. So that is something that you can start with. And most of the use cases that I've seen are about how the AI is able to tap into the disparate data systems that a company has.

Aman: And if you ask a question to that internal application, the internal application has access to these data systems via governed means, and then the AI is able to synthesize a response that provides you insights. So this is one of the use cases I have seen—an internal document Q&A.

Aman: And when it comes to production, there are a lot of parameters that we need to take care of. For example, the model capacity, the security and the governance, the identity. And that's why it ties back to: if your use case is internal, the threat model is a little bit different as compared to a customer-facing use case. So definitely start there, and then work your way up once your organization is comfortable, I would say.

Laura: Fantastic. Thank you. Okay, we've got a great question for Jannie from Evan. Thanks for writing in. "Just curious if analysis has been done on the impact of cost to patients with your in-house prescription solution. Are patients mandated to use your Rx service versus others? Is there any benefit to patients beyond the improved ROI for the institution?"

Jannie: Yeah, those are great questions. So I'll say our pharmacy does have competitive pricing. Nothing is mandated; we really do want to ensure that patients do have choice at the end of the day. So I should clarify that, speaking to that next question in terms of what's the benefit, is that patients and providers are more aware of the services that the pharmacy is actually able to offer, to where they're actually then making the choice to fill their prescription. So I would say they are more empowered with information.

Jannie: And in terms of pricing, our pharmacy does offer competitive pricing, so oftentimes they will be getting lower-cost drug pricing from our own pharmacy. So not only would they be benefiting from that, but I mentioned that we actually expanded pharmacy services. So the pharmacy, from this project, opened up new locations, offers free delivery back to your home, and extended hours.

Jannie: And so those are definitely benefits in terms of just the patient experience of being able to obtain your prescriptions in a better way. So not only is the cost down, but just being able to receive your prescription and not have to leave your house—that's great.

Jannie: So I would say just as a whole, there's definitely been a lot of ROI outside of just cost. But definitely, yeah, just wanting the patient to feel like they feel confident in their choice to come to our pharmacy, and also giving the prescribers, the physicians, that confidence to say, "If I encourage my patient to fill their prescription at our pharmacy, I know that they will get a great experience." So being able to give that confidence to the prescribers has been a big win, too.

Laura: Fantastic. Thank you. Okay. One that was submitted with signups. "For an existing Domo customer, where would you recommend they start if they want to make better use of Domo AI?"

Laura: I can take this one. A great place to start and a great introduction is Domo GPT, so the Domo chat feature. And one fantastic component of having that work really well is our AI Readiness feature. So you can actually go into a dataset and, using natural language, basically create a dictionary of terms for Domo GPT to access. So if you use the term "revenue" or "dollars" interchangeably, you can create all of the definitions that are specific to your vocabulary around that data and prime Domo GPT to be able to interact, engage, and answer questions.

Laura: That is a great place to start. It will also get you thinking on the data foundation piece. So, do you have the right fields in your data? Is the data clean? Where do you start? Are you bringing in the right sources? So that's where I'd recommend that. Anything to add to that, Jannie, as a Domo power user?

Jannie: I would also selfishly plug that if you have questions on use cases, you can always come chat with me.

Laura: There you go. There you go. Okay. Next up. We've got just a few minutes. We'll hit a couple more. What does data need to look like before an organization is ready to start building agents? So maybe some thoughts on getting your AI data ready. What does that look like in practice? Aman, do you have some thoughts on that, and then Jannie can add on?

Aman: Yeah. So this is where the true value proposition of Domo comes into picture. Data ready is AI ready. So what has happened in the industry is that a lot of the data systems that were built were built for humans to interact with that data, right? And now what we are seeing is, increasingly, the AI agents will be interacting with these data systems.

Aman: And that is why we need to make sure that now the data is in a format that is easily consumable by the AI. So one of the formats that I've been seeing in the industry is the open source format of the data, like Apache Iceberg.

Aman: Now, this is where Domo really comes into the equation, because Domo handles those friction points for you. Using Magic ETL—the extract, transform, load—you can actually use Domo to transform your data in a manner where the Domo AI Chat or the Agent Catalyst, the AI products of Domo, can consume the data in a very clean manner. And Laura, correct me if I'm wrong, that's where the AI Readiness scorecard also comes into the picture. It rates how ready your data is to be consumable via AI.

Laura: Exactly.

Aman: So this is something that organizations are definitely thinking about. I can talk about some of our native services on AWS that our customers are now using to make sure that the data is in the format that can be used by AI. One of these services is S3 Tables. And Amazon S3 is our highly scalable object store, and S3 Tables are able to store this data in Apache Iceberg format.

Aman: So organizations are definitely rethinking how the data schema looks like and how this data is stored in these systems. So my recommendation is definitely talk with your Domo account team, and they will definitely help you with how you need to transform the data so that these AI agents can interact. And in a lot of scenarios, it also helps the AI agents to... it lowers the queries that you do, lowering the cost of the query that you perform on your data. Great question there.

Laura: Anything to add, Jannie?

Jannie: No, not necessarily for that, but actually, I want to piggyback off of what you just said around lowering your queries and your costs, because I think that might be related to what you're asking. And I actually, I'm really curious about your question. I would love to chat with you if you have a specific use case in mind, because, of course, I'm going to give you the best answer, which is: it depends, right? We all love that. But I would say, first of all, no, the majority of our AI that we have used in our use cases don't rely on that text generation tile.

Jannie: I will also say, there's a difference between building an AI agent that triggers automatically a workflow versus just using AI Chat versus asking an agent to do something specifically based on a trigger that you set. So all of those are different use cases, and they kind of get to the heart of what you're asking, which is that that way, that agent doesn't run every single time that a user is asking a question, because you obviously want to keep that cost and consumption down.

Jannie: I will say from just a chat perspective, and you can define these different parameters and governance and security settings, where per user they can access their history, so you don't necessarily have to ask that same question every time. Now, if it's that the data is updating, so you want to get refreshed answers, then you can either set up that automation to where that agent is running when there's a data refresh. You can actually even set up a button to where when you are ready for that agent to run, you can click that button, and it will refresh and give you a new response.

Jannie: But again, that's why I'm really curious if you have a use case in mind, because that can kind of differ. But the fun answer is that you can essentially design it for whatever you want. If you want it to be more automated, you can; if you want to be able to track history and responses, you can.

Jannie: So there's a lot of different ways you can customize it based on the exact use case you're doing. And just because you're doing it for one use case doesn't mean that that has to be the way you're doing it for another.

Laura: Perfect. Thanks for that question. Okay, I think we've got time for one more. So let's go to our list. Okay, so this is kind of an industry trend question. What are the other use cases we're seeing with the same approach, where that could be applied to that use case? So kind of beyond the revenue, beyond the patient experience, are there other kind of trends that are coming up with clinical care or risk management or patient frontline care? What are we seeing in the industry?

Jannie: Yeah, I can take that from my perspective, and then Aman, I'd love to get yours as well. So I would say from both being on the hospital side and the service side that helps others, definitely we see a lot of what we call patient access. So that's really looking at your volume and your movement of your people in and out of your system.

Jannie: A big thing that we see around AI use cases of that is optimizing schedules, figuring out how to reduce your no-show rate, scheduling in relation to your no-show rate, and figuring out the relationship between that—seasonality or trends of why there might be certain volumes that pick up here and there. Being able to identify a lot of that in terms of clinics and scheduling.

Jannie: In terms of, like you mentioned, risk management, obviously harm events and hospital-acquired infections are obviously something that's big, especially in an inpatient hospital setting. So anything that we can do to help prevent those falls and infections and being able to look at your data, look for any type of preexisting risk, so that way you're setting up your patient in a better situation to minimize that. So, yeah, I would say anything related to quality and patient movement volume are definitely big requests that we see a lot.

Laura: Great. Anything to add, Aman?

Aman: Yeah. Absolutely agree with Jannie. And what I've seen in this particular space are use cases like clinical documentation. Then we have patient summarization. Last year, I built a prototype for one of the organizations, and they were very much interested in prior authorization automation—like when a patient appointment is scheduled, the system can automatically determine the authorization requirements by analyzing the historical approval patterns for similar services. So these are some of the use cases that we are seeing in this space, at least based on my experience.

Laura: Fantastic. Okay, well, that wraps us up. I want to thank Aman and Jannie for joining us and sharing their wealth of knowledge. Thank you all for tuning in. If you have questions, please reach out. You can click "Talk to an AI Expert" if you're curious about Domo solutions. Jannie's link is in the chat. But thank you so much for your time. This was a wonderful session.

‍