---
id: "workshop:data-quality-and-management-for-the-public-sector"
type: "InnovateUS workshop"
title: "Data Quality and Management for the Public Sector"
url: "https://innovate-us.org/data-quality-and-management-for-the-public-sector"
date: "2024-10-17"
duration: "90 minutes"
language: "English"
instructors: ["Rebecca Friedman","Tony Vachino","Carolyn Carson"]
recording: "https://www.youtube.com/watch?v=BHpRPluvmko"
transcript: "automatic, from the recording"
source: "InnovateUS (innovate-us.org)"
licence: "CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/), as stated on innovate-us.org"
---

# Data Quality and Management for the Public Sector

InnovateUS workshop · 17 October 2024 · 90 minutes

Join us for a 90-minute virtual training with Rebecca Friedman, Tony Vachino, and Carolyn Carson of IBM.  For the public sector to fully realize the potential of innovative digital solutions, including automation and AI, effective data quality is a crucial foundation. By prioritizing data accuracy, completeness, and consistency, government agencies can ensure the successful implementation and integration of these technologies. In this workshop, participants will learn about current challenges in data management in the public sector, creating a data strategy to inform data quality, defining and measuring data quality, and about data standardization and harmonization. Participants will also hear about trends in data management, GenAI, and automation in the public sector and beyond.

## Learning goals

- Understand how effective data quality is a crucial foundation for the public sector to fully realize the potential of innovative digital solutions, including automation and AI
- Learn how, by prioritizing data accuracy, completeness, and consistency, government agencies can ensure the successful implementation and integration of these technologies
- Learn about current challenges in data management in the public sector
- Learn how to create a data strategy to inform data quality
- Understand defining and measuring data quality, and about data standardization and harmonization
- Discuss trends in data management, GenAI, and automation in the public sector and beyond
- **Sign up [here](https://airtable.com/appdAyP4l2W9gxQdc/pagx9sXF2F53dI7kn/form)**

## Speakers

- **Rebecca Friedman**, Data Platform Services Practice Leader, IBM
- **Tony Vachino**, Chief Intelligent Automation and Data Strategist, IBM Federal
- **Carolyn Carson**, Managing Consultant, IBM

Recording: https://www.youtube.com/watch?v=BHpRPluvmko

## Transcript

Automatic transcript of the recording: names and terms can be misheard, and the times can be a few seconds off. Each time links to that moment of the video. The transcript alone: https://innovate-us.org/data-quality-and-management-for-the-public-sector.txt (plain text) and https://innovate-us.org/data-quality-and-management-for-the-public-sector.json (every segment with its start and end second).

**Unknown speaker**

[0:00](https://youtu.be/BHpRPluvmko?t=0) It's turned to you when you're ready.

**Rebecca Friedman**

[0:07](https://youtu.be/BHpRPluvmko?t=7) Perfect. Can you confirm that you can see my screen okay? Great. Thank you again for having us here today. I really appreciate it. Looking forward to the presentation and the conversation that we will have at the conclusion of the presentation today. For the public sector to fully realize the potential of innovative digital solutions, including such things like automation and AI, effective data quality is a crucial foundation. By prioritizing data accuracy, completeness, and consistency, government agencies can ensure the successful implementation and integration of these technologies. Over the next hour or so, we're going to discuss a number of different topics with you in detail, and we'll be sure to leave time at the end for questions and for further discussion, but please don't hesitate to put things in the chat. We'll either hit them as we're going through the presentation or at the end in that last half an hour. These are the main topics that we plan to go over with you all today. We're going to start talking about some current challenges in data quality in the public sector. We're then going to dive into discussing the definition and how we measure data quality and also how we can use data strategy to inform our data quality. We're going to spend some time talking about trends in data management and data quality, and then also at the end, we're going to bring it all together and talk about how we can actually implement all of these things and where we've done it before.

[1:40](https://youtu.be/BHpRPluvmko?t=100) As I mentioned, we're going to start with diving into some of the challenges we see that hinder our data quality. There are really five challenges that we typically see that are preventing us from achieving the data quality that we're looking to achieve. The first is around data silos. This is really the separation of data into isolated self-contained units, often within different departments or systems. The next challenge that we often see is a lack of data standardization. We're going to talk a lot more about this coming up, but this is where we need to establish a common structure and format for data across an organization to be able to mitigate this challenge. Often we'll have a lack of process data. This is the ability to consistently deliver high-quality results through well-designed, efficient, and effective processes.

[2:40](https://youtu.be/BHpRPluvmko?t=160) The last two are underutilized or often misused technology or the lack of a modernized data platform. For technology gaps, we often see that we need to implement strategies to maximize the technology value to help your organization see the true data quality. Additionally, a platform can help in managing the integrated data and helps a lot of the previous challenges I discussed. The data silos and the lack of process data. This slide here takes those five different challenges that I mentioned and shows where we typically see these challenges manifest. The first error is introduced when the system is providing the data on the top left corner. The next, the source system often can provide data that is of high quality, but an error can occur that causes an error with the extraction of the data or when it's transformed or moving or being published for your use. The third data, the data in process on the bottom left here, is when the source system or the target system processes the process correctly, but the business process caused some type of error with inputting the data or the error was created, an error on the back end that prevented the source system from being able to achieve it properly. The fourth error is when the information perceives a problem, sorry, when the user perceives a problem with the data and there's a gap in knowledge regarding some aspect of the information. So, user errors,

[4:40](https://youtu.be/BHpRPluvmko?t=280) all this. And then the fifth is the data just doesn't integrate. There are two similar business processes and maybe there's some confusion in the business rules and there are variances that reflect, that are reflected in the data content. Throughout this presentation, we're going to talk about a few different use cases and I wanted to start one, start with one here that really speaks to how we can resolve data quality at its highest level. So, with a large UK government department that our team worked with, we created a center of excellence to work through a lot of the challenges the organization was experiencing. We, for our approach with this, we made sure that we engaged all of the senior leaders to set up this data quality center of excellence to really enable the business to improve the data quality. We found that it was important to start with coming to an agreement on the scope of this, but also the key performance indicators and then road tested these objectives across the business to make sure we all had the same vision of what this is going to look like moving forward. We worked with a range of stakeholders that both from the business side and the technical side to create a guide on the steps of what was going to be needed to manage the data quality and we co-created this. We're going to talk about this on the back end when it comes to implementation, but it's ensure that there is, as we're working across the board, that the entire organization is in agreement so there's really some teeth involved across the business that we're all in agreement with. So, with that, you need defined key roles and have responsibilities that enable us to implement a robust system and process for managing the incoming data moving forward. In terms of key outcomes, the COE was operationalized

[6:43](https://youtu.be/BHpRPluvmko?t=403) with not just business stakeholders but technical stakeholders to provide full support across the organization and we built in a roadmap to enable this vision longer term. I'm now going to transition to my colleague Carolyn to go in depth on what we really mean when we say data quality.

**Carolyn Carson**

[7:06](https://youtu.be/BHpRPluvmko?t=426) Awesome. Thank you so much, Rebecca. Thank you everybody for having us here this afternoon. I'm really excited to be spending the next couple of minutes talking about how we define and measure data quality and then some strategic considerations and then some themes that we're seeing in data strategy writ large. We can go ahead and move to the next slide. I want to start with a data quality definition. Data is of high quality to the degree that it meets expectations and needs of data consumers, i.e. it is fit for purpose. I want to emphasize that the data users or the data consumers are the ones who get to decide whether or not data is of high quality. This definition is great in the sense that it's pretty easy to understand. It is less good about being measurable, which is something that data strategists and data scientists have strong feelings about. In order to make this more measurable, the data quality definition is further decomposed into dimensions that can be measured. There isn't a single agreed upon set of dimensions that everybody uses. There are different frameworks that people use. Oftentimes our clients will come up with their own that best represents their clients' needs. What you see on the screen here are eight that we think are often there, some of the most common eight. The data is accurate.

[8:33](https://youtu.be/BHpRPluvmko?t=513) It's complete. It's consistent. For example, if distance is measured in miles in one data set, it isn't measured in kilometers in another data set. Referential integrity. Let's say we're tracking customers and purchases. We don't have a purchase floating around in our database without the customer who purchased it. Reasonability, the data meets the expected patterns. Timeliness, how recent the data values are, how frequently it's updated. Uniqueness, and then validity. The data values fall within the possible solution set. Let's say you have a yes, no field. You wouldn't want maybe to show up in there if the values are just yes and no. Going to the next slide, I want to provide an example of how one of IBM's clients is defining quality data. This is the Department of Defense. It's from their DoD data analytics and AI adoption strategy, which you can find on their website. The data quality dimensions that they've chosen to use are visible, accessible, understandable, linked, trustworthy, interoperable, and secure. These are the seven dimensions that they think meet their data consumer's needs most accurately, and it also spells a fun acronym, which is very important in the public space. I also really like this example because it really shows how they are planning to leverage data quality as the foundation for their future data strategy writ large around analytics and metrics and then responsible AI.

[10:07](https://youtu.be/BHpRPluvmko?t=607) Next slide. So, that was some of the basics. Now we're going to pivot into talking about some strategy considerations around data quality. So, any time an organization or a client is considering embarking on a course of action to improve their data quality holistically, we recommend starting with a data quality maturity assessment. This is really useful to help clients or organizations benchmark their capabilities to see whether they have the underlying capabilities that they need to support high-quality data throughout the organization. Multiple frameworks exist. What you see on the screen here is one of the ones that IBM uses. Across the top, you can see that we are rating data maturity from level one, which would be initial or low, to level five, which would be optimizing or the highest. I also want to point out that these data quality maturity assessments need to go beyond just rating the data itself, and they also need to look at the people, processes, and technology that the organization has to support quality data. So, for example, here, that top row is talking about the technological capabilities that the organization has. The next row is talking about the processes that the organization has to support high-quality data. Then the following row talks about the roles and responsibilities and the people that are in place to support high-quality data. Then the last two rows are talking about the data itself, whether people have trust or confidence in the data, and whether the data is appropriately defined. Moving to the next slide, I also wanted to show a data quality framework, so kind of at its highest level of abstraction, how we recommend that clients or organizations set up a data quality program. This follows the same people, process, and technology framework that I discussed on the previous slide. In the enablement area, you can see

[12:08](https://youtu.be/BHpRPluvmko?t=728) data quality organization and governance, which are all about the roles and responsibilities and people who are in play to manage data quality and how they relate to each other. Those darker blue rectangles are about processes, ensuring there are processes in place to define, assess, implement, and monitor data quality, and also to identify data quality issues before they become bigger issues. Then finally, that last row, measurement and reporting, has to do with how the folks within an organization are enabled to leverage the technical capabilities within the organization. Next slide. We're going from broad to specific. I want to spend a little bit of time on some basic principles. Many of these will probably be familiar to folks. The first one is that data has value, and it should be managed that way. People will often say that data should be treated as an asset, like, you know, say, customer names. A newer approach is to treat it like a product, so like iPhones. If you can put a dollar value on your data, that's going to be your best bet in terms of getting people to invest in it and be able to articulate return on investment for it.

[13:21](https://youtu.be/BHpRPluvmko?t=801) The next is to automate data entry and transformation processes. A huge number of data quality errors come from people just typing things in wrong, and so automating those processes can get rid of an entire category of data quality issues in one fell swoop. The next is to clearly define who's responsible for data management. There are many different ways that organizations can assign responsibility for data quality and data management. The important piece here is that the people who are responsible for it need to know that they're responsible for it, and they need to be able to enforce good behavior. Along with that, it's helpful to establish incentives for managing and using data appropriately. The most common one here is to include data quality on performance evaluations, to clearly articulate how data impacts the organization's KPIs, tell people, you know, how would the organization be affected if we misquoted our budget for, you know, this fiscal year, and then to embed data quality in operational and mission critical processes.

[14:29](https://youtu.be/BHpRPluvmko?t=869) Everybody has probably experienced this if you've ever logged into a banking app and it's made you verify your email address or your phone number before checking your balance or your statement. That's an example of a system enforcing a data quality check before allowing the end user to do the transaction that they're actually interested in. Next slide. And then going into the tactics, so the tools that we recommend that all data quality professionals or folks who are managing data quality The first step of this is to figure out which data elements are the most important in your organization so that you're not boiling the ocean, and then defining them so that everybody knows what they mean and what you're talking about. Then you should define rules that help you to measure data quality. These are usually based on those different data quality dimensions, accuracy, completeness, those sorts of things that we talked about earlier. They're often tripartite. They have a principle, a metric, and a target.

[15:32](https://youtu.be/BHpRPluvmko?t=932) For example, let's say we were hosting an event. We want to capture attendee data, and our principle is that all attendee names must be captured. The metric is we want to quote this as the number of names captured over the number of tickets sold as a percent, and our target is 97%. Data profiling is a computational technique that allows these data fields, data values, to be checked against the data quality rules in a more automated fashion. Most data catalog, semantic layer type technologies, so Colibra, or Informatica, or Watson Knowledge Catalog, any of those are going to have data profiling tools built in, and those can be ... If folks could go on mute, that would be really helpful, thank you. That is a useful way to speed up the process, and then ensuring that data quality assessment plans are built into the organization, are proceduralized, so that once a year, or more frequently, folks are going through and checking their data quality, and then ensuring that root causes are identified and validated.

[16:50](https://youtu.be/BHpRPluvmko?t=1010) Next slide. I want to take a minute to talk about what I call the data quality conundrum, because we frequently see clients who are doing absolutely everything right. They are doing all of these things, and their data consumers are still complaining about low-quality data, and this is because of this fundamental disconnect where data producers are the ones who control data quality, but the data consumers define quality. There are two different strategies that organizations can take to try to bridge that gap between data producers and data consumers. The first one is to make them both report up to the same individual. If your organization has a chief data officer, or a chief data office, that's the solution that your organization is taking. This requires that the person who is in charge have enforcement authority, and be close enough to the data that they can actually influence the behavior of the data producers and data consumers, actually change the way that they are either defining quality or creating quality.

[17:58](https://youtu.be/BHpRPluvmko?t=1078) This can be difficult to pull off, because oftentimes decision-makers are too distant from the data to be appropriately informed. The other option is to use data contracts, so this would be a formal or an informal agreement between data producers and data consumers that they're going to work together, which can be tricky to achieve. If you are using a data mesh, or you have a decentralized data governance organization, this is what's required for you. Two potential implementation risks is that it requires a higher degree of data literacy across the organization to ensure that the producers and consumers can understand each other, and it also requires that there be technological tools available to everybody throughout the organization, so that the producers can be producing high-quality data, and the consumers can be evaluating it. Next slide. I want to take an example of how one of our clients is trying to solve this data quality conundrum.

[18:59](https://youtu.be/BHpRPluvmko?t=1139) This is from the Army's Unified Data Reference Architecture, or UDRA, which is available on army.gov if anybody is interested in it. They are trying to implement a data mesh type of data architecture. You can see here how they're trying to bring the consumer on the left and the producer on the right closer together. They do still have, you know, a technological piece in between them, but the lines going from the consumer to the producer are supposed to show the constant feedback that the consumers are providing, and then how the producers are using that feedback to improve the quality of the data that they are publishing. Next slide. Cool. So that was a really, really quick, you know, kind of bare-bones data quality strategy, data quality definition. We want to pivot a little bit and talk a little bit about some trends in data management, gen AI, and automation that we're seeing in the public sector today. Next slide. The first one is intelligent automation.

[20:01](https://youtu.be/BHpRPluvmko?t=1201) So this is a set of technologies that a lot of clients are using to speed up highly manual processes or very complex processes, to free up staff time for creative work rather than doing administrative tasks. It is built upon a number of different technologies, three of which are artificial intelligence to make decisions, answer questions, business process management to orchestrate workflows, and then integration using something like RPA or APIs. Underpinning all three of these, there is a requirement that the data be of high quality. If not, investments in automation are likely to not be returned nearly as highly as otherwise, because they are all, you know, the artificial intelligence is predicated on, you know, making Next slide. decisions on quality data, et cetera. We want to talk about artificial intelligence and generative AI in particular a little bit more.

[21:05](https://youtu.be/BHpRPluvmko?t=1265) IBM has a point of view on how generative AI can be used responsibly in enterprises across the board, particularly in government, which is that the AI has to be open, targeted, trusted, And again, quality data is kind of the prerequisite or the cornerstone for each of those. and empowering. Quality data needs to be available to everyone so that they can openly innovate. It needs to be vetted so that use cases can be evaluated. It needs to be trusted to ensure that there are no security breaches, et cetera. Next slide. I also want to talk a little bit, just a little bit about two different data architecture frameworks or approaches that are getting kind of a lot of news these days, and that's Data Fabric and Data Mesh. They're often conflated, or it can be difficult to tell the difference between the two. The simplest way to think about it, I think, is that Data Fabric is a technical architecture. In the past, the way that we tried to solve getting our end users access to data was to take all of our data from across the organization and dump it into a data lake or an enterprise data warehouse.

[22:16](https://youtu.be/BHpRPluvmko?t=1336) Now today, we're beginning to pivot to a Data Fabric approach, or many organizations are, which is saying, let's leave the data where it is and just make it discoverable so that end users can find it where it currently lives. This can be a pretty complex technology problem. Each of these rectangles show some of the different technological components that need to be in play in order to be able to serve up data in real time to end users from where It also requires a really relentless commitment to data quality because the data has to be it currently lives. really, really well-defined for people to be able to find what they're looking for. Next slide. And then Data Mesh is a socio-technical approach, which is... Those are fancy words, nickel-sized words. Basically what this means is that the Data Mesh concept is focusing more on the roles and responsibilities than on the technology that's being used.

[23:23](https://youtu.be/BHpRPluvmko?t=1403) It's technology agnostic. On the top, you can see some representations of the state of how people work together or don't work together when big data platforms are used. The failure mode that we were primarily seeing is that having a big data platform separated the data producers from the data consumers pretty significantly. Often there were a lot of different technical teams doing database administration or platform development that stood between the two of them. It made it really difficult for them to communicate, which made data quality really hard to manage. This Data Mesh paradigm shift is essentially saying that the people who produce the data are now responsible for an end-to-end product that is going directly to the consumers. And we are trying to make that through-line between the two as obvious and as embedded as possible so that folks can really be dedicated to data quality in terms of meeting the end-users' needs.

[24:25](https://youtu.be/BHpRPluvmko?t=1465) It's more complicated than that, but kind of at the highest level. And I would be happy to talk about this a little bit more if folks have questions about it. I want to end my section, turn it over to my colleague, Tony, who is going to bring it all together and talk about a real-life use case that we have at IBM.

**Tony Vachino**

[24:50](https://youtu.be/BHpRPluvmko?t=1490) Great. Thanks, Carolyn. Appreciate it, everyone. I'm looking forward to being here today and talking to you. So far we've talked about what data quality is, what are the types of measurements that we look for. We talked about frameworks. How do we organize around governing data? We also talked about where it sits underlying with some technology. And then a few architectures that can kind of help bring it together along with some of the tools that would go into a data quality or data management program. And specifically what I'd like to do is try to wrap those things together a little bit. So come at it from a few different angles here and walk away with kind of a use case of how we do this today. So just to kind of lay the groundwork for that, to the left here, it's sort of talking a little bit about from a technology and a piloting perspective. To the left, it's talking a little bit about AI.

[25:53](https://youtu.be/BHpRPluvmko?t=1553) But the reality is, let's think about this of any technology. So automation is imperative across the federal government today. Data stored on various clouds, various systems, everything that might have an API. Maybe we have to integrate with RPA to get to it. So really picking and applying the right kind of technology solutions and really picking use cases from a standpoint that we understand what that ROI is going to be. So how do we measure that? So business context is extremely important. But that also tells us where does that data live today and in what systems. So we'll talk about how we actually build some models to support that. And finally, how do we create trust? So one, it's bringing technology to bear, having the data quality in place that we can actually trust that automations are able to scale.

[26:54](https://youtu.be/BHpRPluvmko?t=1614) And when I say automation, it can be everything from simple things of updating records with basic business rules. But sometimes those things go off the rails as well. So we need to make sure that the data inputs to those are correct and that we're locking things down as we enter it into another system. So not just a user that can read something on a screen and understand that you can't put in an incorrect number or verify something, but also a system to system integration and really the governance around it. So as we start to train data sets for things like artificial intelligence, making sure that we have trusted data that we're building models for, and also that we're monitoring those. So the other flip side of this is, are we getting the right quality business analytics around these systems and these programs, and can we trust that data to ensure that whatever we put in place is not off the rails? Next slide. There we go.

[27:55](https://youtu.be/BHpRPluvmko?t=1675) So one of the things that we really focus on kind of coming back to all of this is, and we talked about data cataloging, and that's kind of one set of tools that you could use to kind of develop what we, and what industry is really relying on today is a semantic layer. It's really a place where we can kind of connect the data to business users. And often that has to go through process. And process and policy is a lot of the places where when we talked about those frameworks of what does data need to look like, what shape, what are our values, timeliness, all of those things that comes from our deep understanding of process. So we can use some tools today to help us mine those things where we actually have it, whether it's existing in documents. So various companies now have kind of LLM-based technologies which can read through policy and help you start to create an ontology or business understanding of your data.

[28:56](https://youtu.be/BHpRPluvmko?t=1736) But also looking at it in terms of where it's coming from, what your authoritative sources are, what the inputs are to a system, what the outputs, if you're automating processes, what are the rules and how are people evaluating that data and making decisions today? And all of that comes from process and analyzing process data. And done correctly, one of the things that we work towards is building a semantic layer where we actually use that process metadata along with those business definitions. And that helps in a couple ways, right? So it can help our builders. So if you're implementing, say, a data mesh, how do engineers that are building interfaces know to use semantically correct business terminology or definitions to create the right products that people can use and reuse? So kind of starting up front and now pushing that towards your future engineering activities. But also mapping historical data, which is the use case I'll be talking about, or legacy systems.

[30:00](https://youtu.be/BHpRPluvmko?t=1800) into a new framework that's going to allow you to trust your data and ultimately build things kind of faster and with more fidelity. So just on a quick note, some of the things that happen as you go through this process that are very important, understanding where to deduplicate data. So in many cases, again, those authoritative sources are very important. I do want to just touch on that, that sometimes you might have IDs or people's information that are stored across the enterprise and which one should be sort of the gold standard. And that's a really important thing to understand as folks are coming in to kind of help build and automate business processes. Standardization, it's often not going to exist. You're not going to have standard data formats across systems that have been built over the past 20 years. You're not going to have standard schemas of how those things come together. So building out that catalog or that semantic layer, which allows you to look at those products and then map them to something common, but have lineage to know where they're coming from, what they mean, how they've been built, having a schema that allows you to understand how to integrate with them is a really powerful capability for consistency and interoperability. And finally, how do we talk about things? Harmonizing, kind of having common semantics. If you're going to have a chief data officer, it's important for them to understand how the business is actually using the data, because they may not be aligned to a certain business unit and now they're making decisions for it. So kind of that top-down, bottom-up model really puts reality in place around data and helps you start to scale it. And finally, that's going to give you a lot of that normalization that we're looking for to be consistent and to be accurate as we start to leverage data,

[32:02](https://youtu.be/BHpRPluvmko?t=1922) especially around future technologies. All right. So the use case I'll talk about today is actually one that we have over at the Department of Veteran Affairs. They have been kind of undertaking an initiative since around 2019 to automate various aspects of their claims process, but with a vision of kind of going end-to-end and providing humans as much benefit as possible. And what that means is right, both from a veteran standpoint, helping them kind of get their benefits faster. So moving through all of the legacy systems and processes, but also employees. It might sort of seem simple on the surface when you think about kind of a medical claim or any insurance claim, right? But those things have thousands of potential data elements that are examined at any point in order to make a certain decision. And there can be hundreds, if not thousands of claim types, right? Things that are financial for pensions, things that are compensations that are based on very complicated medical diagnosis that have to be related to a person and their service at a certain point in time. So this is kind of outlining a little bit of kind of the challenge they have in terms of data about a person. And ultimately, this information exists in upwards of 50 various systems and different access points. So it's a challenge to understand, but if we can bring it all together in a model that allows us to understand how data is being used, right, from a process perspective across all of these claims, what are all of those inputs? What are the decisions that are being made? We can tie together some very powerful capabilities and also inform more of the future in terms of how they

[34:08](https://youtu.be/BHpRPluvmko?t=2048) can automate and improve their processes. So one of the programs that we have is specifically around building what I'll actually call more of an actionable semantic layer. So what we've done is, and this is just a very small snippet, it doesn't get into sort of all of the properties around, say, a person, a veteran, their experiences, but it kind of touches on the surface of when I was talking about those relationships and interdependencies. Although we see sort of these four main categories of, you know, a person in the center, which you kind of see in purple, different nodes of maybe locations or common things, their medical diagnosis, along with all of the claim processing off to the right side of this graph and their experiences. So what the graph does is it allows us to, and underlying this, we have a multitude of data sources, right? So things that we're actively, you know, Carolyn talked about profiling, right? We profile those sources and we start to understand how they fit in this common ontology, right? So one, we can understand it, we can understand how the data relates, but if something was an apple 10 years ago and now we need to call it an orange, it helps us start to fix and resolve those issues and then analyze the data in context of how it's being, of how it was being used or how it would be used today. So it's a powerful capability for driving direction of where you want to head. And part of that, I want to kind of zone in on, we talk a lot about process and everything to the right side is really about those phases. So we kind of call them phases within this and there's a number of types that sort of roll out. So we actually get all of that context

[36:09](https://youtu.be/BHpRPluvmko?t=2169) and it all relates back to a person and under every one of those activity types, we can see how their information was used and all of that to the top comes out towards a profile and a decision type. So it lets us see kind of end to end the experience and how data from the enterprise perspective is being used to create and manage their core business. So I guess a couple things here from, you know, to wrap us up in terms of tips for implementation. You know, starting everything with empathy, recognizing what's happening in the organization, who you're working with, they're going to be a lot of data owners in different places and understanding where everyone's coming from and being able to bring that together. couple highlights for me, I also would say are number two, being obsessive about, you know, A couple highlights for me, I also would say are number two, being able to have a data driven insights. So that profiling aspect is very important. What do we have? Is it a quality today? But ultimately, how does it map into our mission, right? What we're doing, what we're delivering and of what value is it? And then as you start to build capabilities around quality data sets, I think it's important to consistently measure those outcomes and trust those as well, right? Are we actually delivering value with the data that we've integrated?

[37:37](https://youtu.be/BHpRPluvmko?t=2257) Are we building, you know, selected subset data sets, which is part of that program at VA that we can trust to help train an AI model, ensuring that the organization's aligning data correctly. I think executives at the table are very important. It's one thing to go and say we want to pilot technology, but it's kind of another when folks are raising their hands talking about data quality, right? But if you want to move forward with things like automation and driving business value, we need to make sure that executives are on board and understand the importance of preparing data for those initiatives. And I think having a large strategy of where you want to head is going to help with that and ultimately planning for scale. So, you know, that's exactly what VA is looking at today is how do we really understand our data? How do we know that and trust that we're building data sets that can be used to govern an AI, to train a data set? Things that we, you know, things that happened say 10 years ago, they wouldn't want to train on that data. The policies have changed, there's discrepancies, there's new capabilities that have come to bear within their processes. So I think it's really important to also look at your data holistically, what your goals and what your vision happens to be and then build it out in a manner that allows you to kind of work through it.

[39:17](https://youtu.be/BHpRPluvmko?t=2357) I'm going to stop sharing my screen so I can see the questions unless you want to go through the

**Unknown speaker**

[39:25](https://youtu.be/BHpRPluvmko?t=2365) questions. No? Okay. Are there any questions? Okay. Now I'm saying, now please feel free to drop the questions on the chat or use, I mean, raising hands using the reaction tool. And we are starting now the Q&A. So I will stop the recording first. Okay. I have one question. I am muting you. Okay.

**Audience**

[40:03](https://youtu.be/BHpRPluvmko?t=2403) Hello, everyone. Thank you so much for the presentation. There was so much there that was really rich and I'm going to look forward to reviewing the slides when you all provide them. I'm a

* * *

Original page: https://innovate-us.org/data-quality-and-management-for-the-public-sector · Licence: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/), as stated on innovate-us.org
