Sponsored by Fivetran | From Backlog to Breakthrough: Inova Scales Data and AI with Fivetran

Summary

Inova Health compressed a 4-year data modernization roadmap into just 6 months by standardizing on Fivetran, dbt, and the Databricks Data and AI platform as a unified, automated data movement foundation.
By replacing fragmented pipelines with a modern, automated data stack, Inova reduced data movement costs by up to 8x and cleared a backlog of more than 500 outstanding data requests.
The modernization created a near-real-time, trusted data foundation now powering analytics and AI applications across Inova's five hospitals and more than 30,000 employees, enabling faster insights and a scalable base for intelligent systems.

Chapters

FAQs

How did Inova Health compress a 4-year roadmap into 6 months?

Inova Health standardized on Fivetran, dbt, and Databricks as a unified data movement and transformation foundation, replacing fragmented point-to-point pipelines with a single automated layer. This approach eliminated manual pipeline work and enabled rapid delivery of trusted data across the organization.

What cost and efficiency improvements did Inova Health achieve?

By modernizing with Fivetran and the Databricks Data and AI platform, Inova reduced data movement costs by up to 8x and cleared a backlog of more than 500 data requests. Dashboard delivery time also dropped from approximately 3 months to 1 week after adopting the new foundation.

What governance and CI/CD practices did Inova implement for healthcare data?

Inova built CI/CD pipelines with GitHub integration and automated metadata management to govern healthcare data at scale. The architecture includes schema generation, SQL automation, and dbt-driven transformations designed to meet the reliability and compliance requirements of a healthcare organization with over 4 million patient visits annually.

How is Inova Health using AI to accelerate its data engineering workflows?

Inova is infusing AI into its data stack through automated schema generation, SQL assistance, and dbt automation to accelerate developer productivity. This video describes how AI is being used to further compress delivery timelines and build the scalable foundation needed for the next generation of intelligent healthcare systems.

Full transcript

[00:08] Welcome everyone. Uh, hi. I'm I'm Taylor, co-founder and chief operating officer at Fiverr. Um, Fiverr helps organizations across the world build scalable and reliable analytics and AI.
[00:21] Uh, I'm excited for this conversation. I appreciate you all coming out. I think right now in the world of AI, everyone's investing heavily in AI, but I think the challenge is actually making AI
[00:33] pragmatic, useful, and getting the return on investment for AI. So, I'm excited to have a conversation uh with John McManis here from Innova Health, who's been leading major transformations
[00:47] uh for one of the most complex data environments out there. Um before we get into chatting with John, wanted to just run through a little bit more about Fiverr. So Fiverr del delivers trusted
[00:59] data foundation for enterprises that need scale uh for AI and analytics. We have uh over 8,000 customers on the Fiverr side and we recently merged with DBT. So combined we have almost 15,000
[01:13] customers worldwide. uh and helping these customers like Autodesk or LVMH uh Fizer, JetBlue and Nova to build scalable and reliable infrastructure for
[01:26] analytics and AI. Um and uh just you know a quick second before we get into questions with John. Um you know we the the infrastructure
[01:38] that we are building towards what we're calling open data infrastructure we think of it as sort of modern data stack version two uh which is essentially you get all the data in we help you manage
[01:51] that data uh and then we help you transform that data uh and build context for it for analytics and AI um so enough about about us and where we're going
[02:03] let's talk for a second about anova health. So, Anova Health is uh one of the largest and leading nonprofit healthc care organizations or providers uh and has over 4 million patient visits
[02:17] annually uh over 30,000 employees, five hospitals uh and a lot of you know legacy data problems that they have worked through quite efficiently um to
[02:29] to build um a foundation and an actual AI practice. Um, so I guess to to start off, John, maybe you can introduce yourself, talk a little bit about, uh, what you've been leading in in the
[02:41] transformation for AI, uh, and go from there. >> Sure. Uh, Taylor, thanks for having me. Uh, it's great to be here. Again, I'm John McManis. I'm, uh, I have the privilege to serve as Anova's chief data and AI officer. Um, and so again, we're
[02:55] based in the Washington DC, Northern Virginia area. Um, I was, uh, tasked to come to Anova about a year ago. um to really position the organization's data
[03:07] and AI infrastructure to be successful going forward. Um and it's not uncommon uh to run into organizations like Anova was about a year ago where we still had a lot of onremise SQL server stored
[03:20] procedures, SSIS jobs, a lot of federated activity. Um things took a long time to get done, whether it was a complex reporter dashboard, ingesting a
[03:32] new data source. I mean these were things that were measured at months at best. Uh they were not things that were done timely. And so we came in really with a plan on how we wanted to pursue a modern data stack. Um the secret to
[03:46] being successful with AI we firmly believe is a high quality and easy to use modern data platform to really house and deliver those capabilities. Um and so we look for key partners to help us
[03:58] be successful in doing that. Um, and we've been hard at work for it for a year and we've had a lot of great outcomes. >> Yeah, I I would say that. So, let's let's get into it a little bit more. You know, I think obviously coming
[04:10] of AI, everyone's saying, "Hey, you got to go faster and build with AI," but like what was the what was the first step in doing that? Like what how how did you build this foundation? Um, and and what have been some of the results
[04:22] you've seen so far? >> Yeah, so I I get to wear two different hats. So, I'm the company's data officer and I'm the company's AI officer. So, it's really doing a couple things at once. Um, on the AI side, there's a lot
[04:35] of governance mechanics that have to be stood up. Uh, do we know all the AI going on? How do we identify it? How do we determine if it's safe and ethical? How do we have all the protocols necessary for that? And that's just what
[04:48] comes from the vendors we're partnering and buying, let alone when do we start building? how do we start differentiating the work that we do? In parallel to that, I have to look at the analytic uh infrastructure and the data
[05:01] infrastructure um and we really felt like we needed to modernize that very quickly. So the days of working on premise and in a federated capacity, you just can't be successful anymore. So we
[05:14] looked at how could we bring upwards of 70 different major data sources. And so when I say that, most health systems have about a thousand applications,
[05:26] unique vendors that we partner with. Yes, not all are the same size. Some at that number can be really small, but we're usually dealing with dozens of major vendors from the electronic health
[05:39] record to the consumer platform to the marketing tools to the supply chain capability. These are all different vendors. And so it's really important for us to be able to position that data
[05:52] >> in a uh flexible and cloud infrastructure so that we can develop upon it and really build repeatable patterns. And so one of the things that was important to us was figuring out who would be the right partner for that data
[06:05] platform, who would be the right partners to come with that platform to help make that a reality of how to bring those federated sources together. And then while we got our hands around all the AI stuff going
[06:17] on, how do we set ourselves up to really build and modernize these two different capabilities together when we choose to? And I think the way I like to explain to people is there's um things that we need
[06:30] to uh make, there's things we need to shape, and there's things we need to take. And what I mean by that is sometimes we're bringing in something that's like an FDA approved algorithm.
[06:42] We're not touching it. We're bringing it in, we're implementing, we're taking it. Other times, and to be honest, most of what we do is we shape. We bring in a partner's product and we're able to configure it in whatever low code
[06:54] interface or capability that provider gives us, but you also have to have the ability to make so that you can differentiate yourself in your capabilities. And so we really wanted to
[07:06] establish a balanced portfolio across those three concepts uh around all the different ideas we have for things we want to make the capabilities our vendors are bringing and the things we wanted to do with our data.
[07:19] >> It make makes total sense. C can you break down a little bit of the the actual stack of what you're using? What are some of the the different tools like obviously 5chain and dbt you know? Sure. >> Data bricks like maybe break down some of those pieces.
[07:31] >> Yeah. So um let's start with the the data stack. Um so uh we decided to uh have a large investment with Azure. So we're um Azure uh from a like large public cloud standpoint. Um our so I'm
[07:46] going to frame this really around the concept of a CI/CD pipeline because that's important to us. So we use Fiverr for ingestion. Uh we use datab bricks as our primary data platform. I know data
[07:59] bricks is kind of hard to explain. It's like a utility in a lot of ways, but that's that's our core data platform. We use DBT as the data transformation vehicle. We use Airflow. Um, and so
[08:13] we're using Astronomer uh for orchestration. Um, and I think when you when you start to package those major products together uh to eventually get like a PowerBI or various uh uh analytic
[08:25] deliverables on the back side, there's a lot of key points in that process that you connect other partners to. And so we think of DBT as a real like I'm going to say this, it's not going to hopefully doesn't sound bad. It's not a choke
[08:38] point, but it is and something we take full advantage of. So when we want to draft SQL in a nonproduction environment, so in a in a datab bricks context like you know our bronze tier
[08:50] before we go to silver when we're ready to move that SQL and that code through DBT, we're able to capture and execute a number of microservices. So we can write things down. So we can automatically
[09:02] populate GitHub repositories. We can write metadata down to data hub. We can do some AI scripting to build lineage in key areas that we need or data science products that are adjacent. And so as we
[09:14] bring data in with Fiverr, we land it in data bricks, we transform it with DBT, we orchestrate it with Airflow, we're also able to connect a whole family of
[09:26] products that allow us to get about 80% of the heavy lifting of data governance and metadata management done while we're just doing the development pipeline. And so that is really the core of our data
[09:40] stack. And then we have started to infuse AI all over the place. So we do a lot of work with Anthropic on top of that platform. Uh everything from how we're starting to truly like automate
[09:52] how we do BI and I could get into a whole discussion about what that would look like. um how we are creating specific apps within the data bricks platform taking leveraging genie spaces
[10:05] for example but the biggest thing that I would say is our ability to do this at scale with the right partners has been unbelievably successful and in a lot of ways and I know we talk about the use
[10:17] case we've essentially done about four years worth of traditional work to come to a new organization and get all this in place in about 6 months to a year um and That's been really profound for us
[10:29] to actually go from like onrem to this type of cloud infrastructure, this modern CI/CD pipeline and really feeling like we're AI ready within a year. I've only been at the organization a year and
[10:42] we're moving really quick. >> I mean,
[10:54] you know it took you about four years to go through that transformation. The last company when you moved to Anova and said hey I want to do this in six months he literally was like we got to do this in six months and I I I was like okay we are on board to make that happen but you
[11:07] know I think uh it's challenging more than just the technology. I mean, one part of it obviously is figuring out the right technology, how that plays together, and I think we did a really good job of, I think, helping you all with that, but there's also a political side to it. Like, how did you make this
[11:20] happen so quickly? I mean, this is fast for like any company of your size, but especially in healthcare, this is like extraordinarily fast. >> Yeah, it's it's it's a so the political thing is a little bit layered. Like you
[11:32] have to have a detailed plan. You have to be able to communicate what outcomes you're after. You have to explain to your finance part of the organization. and you need to double pay for a certain period of time to transition from an old
[11:44] stack to a new. I mean to put things in perspective within this year we now have 60 connections in place through our work with Fiverr supporting about 30
[11:56] different vendor products feeding information into our data brick stack so we can do development. We've brought in our major cost accounting finance system. We were able to bring in the
[12:08] Adobe Experience platform which was a net new install for us for consumer websites marketing in under a week. And a lot of that is because Fiverr has these off-the-shelf connectors that we
[12:20] can take advantage of. We've been able to really position a lot of the traditional ETL ELT code maintenance for our electronic healthcare partner uh
[12:32] healthcare record partner. um because they have connectors that we've worked on with them and we're able to position that off. So that is not something that's been able to bog us down. I've been able to position key developers in
[12:45] the actual act of how do you build a CI/CD pipeline? How do you train people how to use it? Uh we have a whole franchising system. So when we identify analytic teams in the organization that
[12:57] don't have to report to my central team and corporate, we franchise them. Like I'm Chick-fil-A and they're going to buy their own Chick-fil-A store regionally, but you can't buy your own kitchen equipment. You got to use ours. You have
[13:10] to be closed on Sundays. And so we're able to scale those kind of political elements to really make this successful quickly across the organization. But a lot of that's, you know, planning, knowing your outcomes, knowing how much
[13:23] this costs and executing on the deliverables that you commit to. >> Yeah. I mean, it's still again astounding. C can you talk just like for a second about how the organization is structured? I mean, is it all centralized or you said you obviously
[13:35] there's some hospitals or groups that have their own, you know, teams like h how do you work through that piece of it? >> Sure. So, I have about a team of 135 people. Again, we're close to a 30,000
[13:47] person organization. uh our IT division for perspective is about 600 people across our organization. So I have a good chunk of those folks. Now that's split between data and AI. You know it's
[14:00] not all the same jobs. uh I have groups that folk on finance BI develop clinical uh BI development but essentially we are doing this primary modern data stack
[14:12] building the workflows in the AI infusion with about eight what I'm starting to call like industrial engineers um to be honest when you get into these tools you actually want less
[14:25] hands touching them particularly when you're like orchestrating entire um farms of agents to kind of execute BI. And when I say that, like I genuinely mean that like a customer
[14:37] engages with a chat assistant that is capturing the requirements before it submits via API the ticket into our uh system. The YAML files created behind the scenes. We upload that into an AI
[14:51] inbox and we orchestrate the creation of schema, SQL, DBT runs, code merges, GitHub repo commits, all the way to creating the actual draft PowerBI
[15:05] dashboard. And then we go and present it to the customer the next day. So we've done that with about eight industrial engineers and then we have a data engineering team that's about I'd
[15:17] say 18 or so individuals that really help with a lot of the templated and playbook work that it takes to execute these workflows at scale. And then of course there's still good old legacy
[15:30] engineering tasks that need to be done like data exports or this vendor needs stuff done inside that application that's not going to be AI automatable anytime soon. So it's still a mix but we
[15:42] have not done this with that many people. We really essentially leveraged about eight engineer architect level individuals to do this work at scale. >> It's crazy impressive. I mean this would have taken an army of people in a
[15:55] previous life. they used to. >> So maybe we can shift for a second to some of the the like business outcomes that you're driving thus far and what what moving to this stack has unlocked for you. >> Yeah. So we are starting to commit to
[16:10] SLAs's where we develop analytic content that's measured in single week increments. And that's a huge commitment. That's a very different world where you know the typical complex
[16:23] dashboard for the emergency room might have taken easily three months or more with all the back and forth of requirements development actual development validation back and forth
[16:36] etc. And when you can have these type of platforms tee you up to be able to essentially get the requirements done without any intervention from your team with that particular customer. Get them
[16:49] to submit that YAML into our workflow. Have it run for about an hour at a very expensive price point. Produce the PowerBI dashboard as the output. The
[17:01] very next day someone submitted the idea for something. someone calls them and says, "Hey, I have the prototype ready to review with you. We got to validate it. We got to make adjustments." Now, we're still going to kind of get stuff
[17:14] done over weeks, but like a one or two week timetable >> versus a 3mon plus timetable. And if you just think about how much everyone makes an hour in that process, what I'd like
[17:26] to tell people is every report we've ever created, even the bad ones that like never get used, cost at least $8,000 if you add everything up. And even if some of this like high-end AI
[17:41] costs a couple hundred for a single report to run at scale, that is still potentially dropping the price point of shorter requirements gathering, shorter validation, and then of course shorter
[17:53] development. My reports might now cost $2,000. And so if you scale that out across building literally a thousand reports a year, these outcomes can really matter.
[18:06] So we've been able to connect data sources um so people can start combining concepts of I want to know how our performances in the clinical EHR. I also want to know some of the survey
[18:18] information that's usually housed in the HR land for experience and feedback and I want to tie that to cost which is usually yet another cost system. we can now have those conversations of
[18:30] integrated dashboards because all of those data sources now reside in data bricks and we're able to move that much quicker. So a lot of our outcomes right now because again we only just got this stood up are really about showing the
[18:43] improved productivity performance and ability to meet the organization's needs quicker. I think the outcomes that are coming is what if I didn't have a backlog of requests?
[18:55] I I've asked my team, what would you even do with yourselves if you didn't have a tsunami of requests you could never get to? What does it mean to actually look for new value work that the business isn't asking you to do?
[19:07] We're actually starting to have those questions as we move forward, which is really exciting. >> I mean, it's it's like one, I think you're you're saving cost for the organization. You're providing much better experience for doctors, which are
[19:19] ultimately providing better experience to the patients. And my guess is you're probably also changing the pace at which you're innovating, improving. Like a lot of times hospitals, it takes forever to get anything done. But it's like, oh, if if the if this the reporting layer, the
[19:32] information that we're getting is coming much faster, it enables the rest of the organization to also speed up and actually improve over time, right? >> Yeah. We um we are very much looking forward to firing on all cylinders as
[19:44] early as this fall to really try and drive to zero ticket Q depth across all of my analytic teams. It's amazing >> and what that would open up for my organization about net new value capture
[19:56] is exciting and in some ways new brown. So, it's something we're looking forward to. >> Well, awesome. Well, I know we're at time, so I really appreciate you today taking time and coming. Thank you, John. Thank you all for listening. Appreciate it. Thank you so much. Thank you.

Learn more about the Databricks Data and AI platform.

The information provided herein is for general informational purposes only and may not reflect the most current product capabilities or configurations.