Workiva Inc. ($WK)

Welcome, everyone, to our webinar. We know AI adoption and realizing its true value isn't just a trending topic, but it's likely a top priority in your company's goals and OKRs for the year. Many of you are actively figuring out the best ways to leverage AI for your accounting and finance functions. So today, we're going to walk you through practical real-world examples of AI and show you how to move from initial adoption to measurable impact. But before we get started, I am going to spend some time walking through housekeeping items. This session is being recorded and will be available on demand. If you are looking for additional resources on this topic, please see the links in the related content section. If you have any questions during the webinar, you can submit them through the Q&A engagement tool. Please note we capture all questions. And if your question is not addressed during the webinar, it will be answered later via e-mail. For those looking to receive CPE credit today, note that we will be asking you to answer four multiple choice questions during today's event. Receiving your CPE credit is dependent on submitting an answer for at least 3 of the 4 polling questions and attending the full duration of the webinar. These polling questions will pop up in the middle of your screen, and there will be a limited amount of time for the answer to be submitted. So please make sure you are ready to answer when the question appears. Also, please do not make your media player full screen as this will cause you to miss the pop-up polling questions. If you are watching this as a group, please log in individually so you can make sure you receive your CPE credit. Once you have met the CPE requirements, your certificate will be available for download in the CPE certificate engagement tool on your screen, and it will also be available in a post-event e-mail. And this e-mail is sent about an hour after the webinar. Unfortunately, we are unable to provide CPE credit to those who have technology issues that prevent them from qualifying or to miss the polling questions. All right. Let's kick things off with some quick introductions and an overview of today's learning objectives. I am Chelsea Hall. I'm an industry principal with Workiva. As a former practitioner, I've been through my fair share of implementations, so I can definitely relate to the kinds of challenges we're going to discuss today. Joining me are Travis Dean, a Senior Manager at Deloitte and Jason McLean, a Managing Director at Deloitte. Travis and Jason work with finance organizations every day on these implementation challenges. So I'm especially thankful to have them here today to provide insights on this topic. And a quick reminder, why are we all here today? This webinar is designed to provide all of you with a better perspective on the current AI landscape for finance and accounting professionals. The critical role of quality data, defined processes and governance frameworks in achieving AI readiness and the high -- some high-impact use cases of AI within accounting and finance teams. So let's set the stage. I think we can all agree that the AI Pandora box is open. And from what I can tell, it's definitely not closing. So the question is no longer whether to adopt AI, it's whether you're going to be AI-enabled or actually AI successful. These are 2 very different themes in my mind, especially when we're talking about the sensitive nature of work within accounting teams and the office of the CFO. Based on Workiva's 2026 executive benchmark survey, 2/3 of executives agree that AI will dramatically boost productivity. That's a really strong consensus and a signal of optimism of AI in general. And we also know that AI adoption rates are high and continue to grow. But here's the reality. Although investments in AI are increasing and its use has grown significantly, most organizations haven't implemented the policies, procedures and controls needed to use AI safely and effectively within their organizations. In fact, only 36% of CFOs report having high-quality data to use with AI and a majority of executives believe that the company's current approach to adopting AI could introduce risk to their organization. But even though integrating AI into accounting and finance work streams carries inherent risk, companies are increasing technology spending and digital transformation projects that enable and embed AI into their existing workflows. Most leaders are prioritizing the automation of data collection and validation as well as strengthening governance.

So Travis, Jason, I've just given a brief background on what Workiva is seeing as far as the state of AI for accounting and finance teams, but I'd love to get your perspective from working with accounting and finance professionals, how are you characterizing the current state of AI adoption within these accounting and finance teams?

Yes, sure. Chelsea, I would be happy to take that. The top-down pressure is real. every day when I talk to my clients, they are getting pressure from someone more senior from them about what is their AI strategy. And what's interesting about it is it's almost taken this form of a hammer looking for a nail in some cases at the moment because everyone is trying to figure out what's their AI strategy, what's their agentic roadmap for their accounting and finance function. In fact, I was just talking to a client the other day, and he said, "Jason, I need an agent for my team." And I'm like, all right, why do you need an agent for your team? He goes, everyone else has one. And he goes, I need it to be called something cool and I need to do something cool. And while we joked about that, joking aside, it's really what the environment is these days. Everyone is trying to figure out how to use this in the most effective and efficient means for their organization. And sometimes that pressure is causing them to react a little bit faster than they probably should by going through a more considerate design process from that perspective. Travis, what are you seeing?

Yes. I definitely hear a lot of that, right? A lot of that top-down pressure. I think what we see as far as number of organizations that have gotten started, what they have started with, I can probably put some numbers behind that. If I think of just like generally the types of AI out there, the capability, like predictive AI and machine learning versus Gen AI versus agentic. I think what we see is maybe somewhere around like 1/4 of the organizations we work with have something out there in that predictive and machine learning bucket. So think of something that maybe touches a forecast or an accrual estimate, something like that. It's interesting to me that number is stagnating a little bit for us in what we see because I think a lot of orgs who are starting are leapfrogging that a little bit and going straight to Gen AI. We're seeing about maybe 1/3 of the organizations we work with who have deployed something that falls into the Gen AI capability. The agentic bucket, that's a little harder to measure. I know that's the buzzword that's out there today. So I think in some cases, people are saying they're doing something agentic, but maybe it's glorified automation behind the scenes. Is it truly agentic? Is it reasoning and proving processes acting on its own? There's far fewer of that, that we actually see out there in practice, very small percentage. But -- and definitely, the area people are asking about interested in because of how much impact it's going to have on finance orgs going forward. But yes, largely to throw maybe one more stat at it, 2/3 of the organizations we talk with put themselves in a bucket of I'm either conservative on AI and in wait-and-see mode or I'm a follower, right? Those 2 numbers come out to about 2/3 and then 1/3 who've actually started. And when I say started, I'm not talking the copilots like user productivity efficiency type tools. I'm talking AI deployed against finance and accounting workflows and about 1/3 of the organizations we work with have started there. And earlier, I was giving the breakdown of, okay, where did they start and they have gotten started. And later, we'll talk a little bit more about specific use cases.

Great. And I think that insight is really helpful, and I think probably speaks. So it sounds like 2/3 of accounting finance professionals that you're speaking with, they're kind of waiting -- they're kind of taking like this wait-and-see approach with AI. They might not be like the earliest of adopter. And I think given by our nature, we're a little bit risk averse and we'll kind of talk about that later in the webinar to like why the stakes for AI and these teams are a little bit higher than maybe some other teams within an organization. So with that, let's see how everybody in attendance is feeling about the maturity of their AI program. So it is time for our first polling question. Everybody, if you're not looking at your screen, now it's time to look at your screen and answer, how would you rate the maturity of your AI program? A. You're just getting started; B. We have some pilots underway. C. We're scaling select use cases; and D, AI is fully embedded in our workflows. And I think, Travis, based on what you're seeing, we would probably anticipate most people are going to be responding A and B. So that is your response, don't feel like you're falling behind. You're in the right place. Hopefully, we'll help you identify some use cases you can scale by the end of today's webinar. So make sure you take time to answer that question. Give you -- I'll leave it on the screen, but I'll go ahead and move to kind of our next topic on trusted AI. What we're hearing from our customers is that there's excitement from executive teams to implement AI. We just heard that top-down pressure is real. They're feeling the pressure to realize the benefits that as we've just alluded to or I did in the opening, there are gaps in execution, especially in relation to establishing the governance and oversight necessary for the secure deployment of AI within an organization. And what we were just talking about, Travis said so beautifully, maybe AI for finance and accounting teams were more like late adopters or wait to see. And I think what makes this particularly high state for finance and accounting professionals is just the nature of our work. Finance and accounting teams are frequently working with sensitive data and the output of our work is generally in the form of financial statements, reports that you file with regulators and you have earnings releases that are going to the street and to your investor community. And these reports have to be accurate. So in this world, there is no margin for error. And then there's also a lot of heightened scrutiny, whether it's from auditors, executives, investors or stakeholders. That scrutiny is incredibly high. So being AI successful in accounting means something different than it might mean if you're like in a marketing or a sales team. You're in marketing using your AI tool to help with messaging versus an accounting using AI to help you make sure your disclosures are accurate or your reports tied out correctly. So to the Deloitte team, how do you define trusted AI in the context of accounting and finance operations? And we're talking about that pressure from the executive team, the C-suite. How are you advising teams to balance the pressure to implement AI at this fast pace with the need to make sure AI is secure and controlled?

Yes, sure. Chelsea, the one thing I like to talk to finance and accounting teams about is really to make sure that they understand which of their AI use cases that they're thinking about have more risk than others. That's such a key component. When this whole AI or generative AI craze or phase started up a couple of years ago, we saw a lot of visioning sessions, and we saw a lot of use case development coming out of there. And your more risk-averse accountants quickly looked at that list and is like some things are a little bit higher risk. And the reality is you can probably do it with AI, you probably get to a result.The question is, can you trust it? And that's really the challenge. I was recently talking with a Chief Tax Officer and he shared something very similar. He shared how his team had presented a list of their prioritized use cases, and he had to detail a few of them. He had to detail because he said, "Hey, I love the use case, but if I can't get comfortable with it, how are we going to do it?" -- we aren't going to do it. And that was his point is he needed to make sure there wasn't a black box. He needed to make sure he understood the inputs, the human in the loop, the human over the loop, the outputs and how he could trust it in order for him to give it a green light. And in several cases, he couldn't do that. What's interesting in using that tax function as an example is I've watched them over time. And what they really started to do is really mature their use cases, mature it to start to think about how they get -- how they embed trust in it, how they are going to get comfortable with it, how they're going to explain that level of comfort to their own management to a third party, maybe their auditors. And it's really kind of helped them kind of really stay focused on things that are practical, use cases that are practical and use cases that are really going to deliver value for them. Travis, what are you seeing?

Yes. Well, I love the stories from the ground there, Jason, and the quotes you're hearing. I think I hear a lot of that as well. When this comes up because we do get asked a lot about this, I think it's interesting, the framing of the question comes to me is like how do I get my auditors comfortable, like they're going to come ask me if we deploy AI. And the way I answer that is almost turning the question back on the person asking it because I think the auditor is not going to necessarily come with a long checklist of things, did you do all these things? Like the first question is going to be, how did you get comfortable with the AI and the results it's providing? You need to have a pretty robust answer to that. Like when we're talking trusted AI, I wish it were like a very clean, simple answer of one thing. But the fact is it's many different things around that are wrapped around any AI deployment, right? So I think you touched on a lot of these, Jason, but even from the beginning of picking your initial use cases, maybe there's a materiality threshold, right, and the data that it's touching and maybe there's a bottom half of the materiality and you start there. While you're building the AI, of course, you're going to want to back test it on historical data. You're going to run it in parallel to your existing process, see how the results are the same or different, like a lot of the standard things you do in sort of a modeling type deployment. And then, of course, once it's live, the work doesn't stop. We all know that AI and models drift over time and it could get worse and responses can be unexpected. So what do you have in place to continue to monitor this, where are the humans in the loop. All of these things are part of that like self-created checklist, I think organizations need to think about to have a good answer to that question. Like how did you get comfortable with this? One thing that's a bit difficult in this, of course, I don't think that the regulatory bodies out there have given very clear directives and guidance yet and policies like you must do these things. But if you take the approach that, Jason, you're talking about that I'm talking about here and have a pretty comprehensive view over the full life cycle of AI deployment, whenever those regulations and policies come out, you're likely going to already be conforming to them. When your auditors or anyone else comes asking questions, you're going to have very solid answers to that. So it's a list of many things to do. But I think all like within the realm of what's feasible and what's worth doing to be able to go on that AI journey.

And I know from Workiva, we talk about like trusting your AI a lot. And I think for people listening, as you think about like how can I trust AI or what can I show my auditors to see how I got comfort, I think there's like a couple of things like can you verify how -- what AI has done to reach its conclusions, how can you trace how that output was formed a couple of ways like does your AI include citations when it gets you an answer? Is there lineage tracking? Is there an audit trail within AI? Is it traceable? Is it explainable? So I think those are just like a couple of things to think about. And obviously, depending on the use case will depend on how many of those features you would need in the response or having that tracking back to the source or that audit t that those are just features that your AI could have embedded in them to make it more trusted. I love what you guys sharing what you're hearing from your customers. And I think one key takeaway that we want to make sure everybody leaves with today is that data readiness and process readiness really comes before AI readiness. This isn't a technology problem first. It's a data and process problem first. And I don't know about everybody else in the call, but I've definitely worked for executives who you could go to them with a problem and they're like, well, let's implement technology to solve this problem. And then you're just like throwing technology on bad of a bad process. And I have one example from a previous work experience, like we had an ancient AP invoicing tool that was supposed to process our invoices. Over 60% of our invoices were selling and they were requiring manual intervention for the invoice to get processed correctly in our ERP systems. So vendors were getting upset because they weren't getting paid on time. We didn't have a very reliable AP aging report because invoices weren't getting processed. So we finally got the green light to implement a new tool -- and this tool even had AI embedded in it. So the solution would continually learn and better capture invoices over time. And so as you can imagine, the AP team was excited about the tool. The implementation moved fast. But after all the work of implementing the new tool, a majority of our invoices were still failing, payments were still being delayed, vendors were still mad at us. So even though the new AI tool was great at reading, capturing and coding invoices, it was a lot better than the legacy system we had been using, we quickly found out that the problem wasn't 100% related to our technology. It was definitely an underlying problem in our process. We found out that POs were being raised in a timely manner. So even though our new tool was scanning these AI invoices and processing them more quickly, they still aren't getting paid because there wasn't a corresponding PO. So although we would eventually come to benefit from the software, it wasn't until everyone had to undergo new PO training to make sure POs are being raised properly that we could actually take full advantage of the new technology. So I think that's just an example of when you throw technology or AI at a problem when that's actually the underlying process or even the data is broken. So to the Deloitte team, you're advising clients and you're walking into an organization like mine, and I'm like, oh, our vendors aren't getting paid on time, and we want to adopt AI in our function. How do you assess their data and process readiness? And what are some signs that an organization is maybe not ready for the latest tech and technology? And what do you advise them to do first?

Yes. Chelsea, the first thing I talk to controllers and other accounting professionals about is how hard is it to do what I would call relatively simple data tasks in their current environment. Is it easy? If you ask for -- if a controller asked for a custom report or some analysis and the team said they need 3 days to do it, there's probably something behind the scenes there that they need to explore. In my younger years with Deloitte, I spent a lot of time doing data wrangling type of efforts with finance and accounting type data. And often, the secret to my team's successes during that time was not the technology. It was not our mastery of the technology. It was that we understood the data. We understood the nuances. We understood all the loopholes the dirty secrets about it. We knew that column 9 or Field 9 in this chart of accounts meant something. And then this other one, it was column 15. And we knew how to translate that and so forth. We built our own little cheat sheet that would help us to do that. And that's the reality in corporate America right now is there's still companies out there, many companies that have those secrets or those dirty aspects of their data that is going to complicate things as they move to an AI-first mentality or trying to use AI in new ways. Two things that we're seeing our clients or at least I'm seeing my clients start to really think about in this is one is at least from an accounting and finance perspective, your finance data model, your common information model fits in ERP. We're seeing a lot of companies really rethink that -- really rethink it to avoid those nuances of why does this division have a different chart of accounts than this, right? They're trying to get to some alignment because they see the value in that alignment and they see the value and how it will make everything just easier from an AI perspective. And the second thing is we're seeing companies really start to invest more in what kind of a buzzword out there is a semantic layer. A semantic layer is nothing more than just logic. It's just logic. It's my cheat sheet that my team had years ago. Now it's just put into code. and allows the AI to properly have the right context to query the data to interact with the data in a consistent manner. And that's really what we want. We want to be able to interact with the data in a consistent manner similar to what we've been doing from a human perspective. Travis?

Yes. I think, Jason, what I'd add to that, and I have to say, like as we're going through this, Chelsea, I'm realizing the intentionality of the order here because typically, everyone just wants to talk use cases, use cases, use cases, let's jump to the fun stuff. I like that we're saying, well, let's talk about trusting AI first. Let's talk about data readiness first. So I think we're going in the right order here. Jason hit on a lot of really good points here. Maybe the only thing I'll supplement that with is we've been talking about data readiness with clients for years. Like separate from AI, like maybe we're talking about it for reporting and insights and for automation long ago, maybe a decade ago, when this would come up, I would hear a lot from organizations like our data is the worst you've ever seen. We can't do any of this reporting stuff with it. I think there's been a lot of progress in that front, a lot of transformation as people have kind of gone to more mature systems on the cloud, maybe consolidated some of that data. Still some pain around data cleanliness, but I think we're in a better spot there. But then how do you get AI to function effectively with your data? I think -- and I know you mentioned one buzzword, Jason, that's going to come up as people are exploring data readiness for AI is the semantic layer, which is really important. I think another thing that people start hearing and you're probably already starting to hear is this concept of MCP servers that stands for model context protocol, which is a weird naming for it. It doesn't tell us much. But essentially, it defines how AI is going to talk to that data, how AI asks for the data, how system responds, what actions the AI is permitted to take. I mentioned this because it's just another kind of part of the equation. Like historically, when we were talking about data readiness, it was very like infrastructure heavy, building data lakes, ETL, everything in one place, then people can kind of query it from there. There is still some value and effectiveness in that consolidation, but the current answers to that emerging are around meaning of the data and access to the data, the semantic layer, the MCP servers. And maybe in some cases, like MCP servers are even allowing direct connection into source systems where the data currently sits. So I think just more options are emerging for how AI can interact with data in the systems and in an environment where it already exists, which I think is a lot -- goes a long way in accelerating maybe some of our -- the ways we use and deploy AI against that data.

Yes. And I think we're definitely hearing from our customers that more and more want for this broader AI ecosystem of all these AI tools like talking to each other and having like an AI stack like we all just think about your tech stack and now it's like your AI stack that's becoming a bigger and bigger request and making sure all of them are using that same trusted underlying data foundation. So I think we talked about getting your data ready and about system implementations and AI implementations. And I think in the example I gave, when you see a problem like invoice is not getting paid on time, you might look to have a quick fix of implementing technologies. But these types of implementations generally fell because the technology is implemented before the underlying broken process is fixed. If you contrast this to when an implementation actually goes well or succeeds, you don't use the system or in this case, AI to Band-Aid a broken process. The key is to get the process right and then you implement the technology. And kind of what we've all been saying, the exact same principle applies to AI. You don't want to use the AI to clean up your messy data, but you do need to focus on having clean, connected and governed data first. And once that foundation is solid, then you can truly let your AI run on top of that trusted data. I think the uncomfortable truth, even though like Travis has said, I think organizations over the last -- even before AI were working on cleaning up their data, most organizations still might not be quite there yet, having this data layer that AI can trust. If you think back to the statistic I shared at the beginning of today's webinar, almost 2/3 of finance leaders are skeptical about the data they're using in their AI. So they're either deploying AI on a shaky data foundation or they're holding back because they know they're not quite yet ready. Neither of these options is a great position to be in. So before we transition to like the next section, I think we need to ask ourselves what does data readiness actually mean. And we like to break it down into 4 key components. Your data readiness means your data is clean, it's connected, it's governed and it's audit ready. So when it's clean, your data, we consider it clean if it's complete, it's accurate and it has been validated. Your data is connected when your data isn't living in silos across disparate spreadsheets or disconnected systems. Your data is governed when there's clear ownership, access controls in place and there's an audit trial to how your data has changed or who's changing the data or how it's getting manipulated. And it's audit-ready when you can trace it back to its original source. So knowing having those definitions of what we mean by data readiness sorry, we gotten switched to this quote slide, but we can move to our polling question. How would you describe your organization's data quality for AI use, either, A. Not ready; B. Partially ready, some clean data; C. Mostly ready, a few gaps and D. Fully ready, clean, connected, governed. I always question like the fully ready is anybody actually have 100% connected or clean data. I think there's always going to be some nuances, especially when Jason was talking about global organizations and having all these chart of accounts, even with your best efforts, if you have hundreds of subsidiaries, one region is going to use or interpret an account usage maybe a little bit differently than than your maybe organization in another place in the world. So something to keep in mind. I think it's always an ongoing process to get to fully ready clean data. So now let's talk about governance, and I want to make sure that all of you here don't see governance as like a speed bump or something that slows you down. Governance is actually what makes speed possible. When you have solid governance, your teams can actually move faster because they have more confidence because they know what's allowed, what's not and how to act when something is unclear. So Travis and Jason, you work with the organization building these governance frameworks. What does good AI governance look like in the office of the CFO? And what are the key components leaders need to have in place?

Yes. I think you mentioned one important piece, Chelsea, which is, yes, this shouldn't be a speed bump. To me, it's also not something new necessarily, right? There are new novel risks associated with AI, but largely AI governance in a finance organization isn't fundamentally different from a financial control mindset that already exists. Again, like we are focused on a lot of different areas and systems where we want to make sure the outputs are accurate, traceable, accountability is clear. So I think the instinct is already there. It just needs to be extended to a new class of tools, new capabilities. As far as what the components are of good governance, again, probably give a handful of things that come to mind here. Of course, the first starts with what we were just talking about, the data, data governance, I think, comes before AI governance. I'm not going to say garbage in, garbage out or few, I won't go there. But ultimately, you can't govern AI outputs if you haven't governed the input. So you need to know what data the AI is touching, how current it is, who owns it, right, those types of things. And then beyond that, this seems like a simple one, but as it becomes easier maybe to build like quick agents, do your own things in AI, it becomes harder to have an inventory of where all AI is maybe being used in the organization. So I think like that a lot of organizations are past that point of total centralization here, like people are using the Copilots, the ChatGPT, like vendor embedded AIs. Leadership in a governance kind of council needs visibility into what's actually in use, not just what's formally approved by maybe bigger bang use cases. I think another critical component is human review thresholds, right? Not every AI output is going to carry the same risk. So I'm thinking of things like first draft of a Board memo is much different from an AI-generated journal entry, just for instance. So it doesn't necessarily need to be super bureaucratic there on what those thresholds are, but still very intentional. I think another piece is something, Chelsea, you touched on when we were talking about trusted AI, which is auditability of AI-assisted outputs when whoever it is, when someone comes asking, how did you get this number that answer needs to be traceable, even if AI was somewhere in the workflow. So that comes down to logging, what the model used, what data access, what the prompt was exactly, what human reviewed it as part of that checklist. I think maybe the last one I'll mention here is probably like the concept of ongoing monitoring. I hit this before, it's worth kind of touching on again, but AI tools change, models get updated, capabilities expand, integrations deepen, like a onetime approval once this is getting rolled out is not sufficient. Governance kind of needs to embed that periodic review cadence. I think ultimately, the way I sum this up is probably the hardest part about it because, again, a lot of the pieces, I think, are familiar. The hardest part about it is probably the speed mismatch that's happening here, like AI adoption, the sophistication of AI capabilities are moving typically faster than governance processes are designed to move. Those of us in finance and accounting tend to have strong controls culture, but maybe slow policy cycles and approving stuff, getting things set up and put into place. I think the practical answer is just building frameworks that are lightweight principles based. It can flex as tools evolve rather than trying to come up with the policy out of the gate that covers every specific tool, every specific use case because that specificity is going to come obsolete concrete pretty quickly.

And I think you bring up a good point, like when I think of updating policies and procedures, like I think most of them, like you maybe look at them once a year or you even look at them maybe like once every 3 years depending on like the policy. But for like your AI policy, just because of the vast nature of AI, you might be wanting to be looking and having a committee overseeing that policy much more frequently than like once a year because of the rapid pace of change. So definitely agree.

Yes. I think you hit on like dedicated governance councils who are thinking a lot about this, owning that process, who can be very agile, and that's definitely another important component to all of this.

I think just like what I'd add from a practitioner standpoint when you're thinking about your governance framework regardless of it's AI related or related to something not in an AI process, there are 4 things that need to be covered. one, the data going in; two, the tools being used; three, the output being acted on; and four, the people doing the acting or using that output and how they're further manipulating it. So if you can't answer like who approved this, what data was used and who used the output, you need to improve your governance structure because that's a sign that you don't have governance in place. And there's a quote I love that I'm actually borrowing from a Workiva customer Cognizant who said, "If you can't trace it, you can't trust it. And if you can't trust it, don't feed it to AI." And that to me is really summarizes the standard you should hold yourself to when you're thinking of placing heavy reliance on output from AI to make sure you can trust the data that you're feeding it. And I think this is also a good place to transition to kind of a build versus buy decision. And I think that's becoming really important, especially as AI is making it easier for everybody to become software engineers and develop their own tools. Because not all tools that -- all AI tools give you the level of traceability and auditability that we've been talking about like taking you back to the resource system or the transparency over how AI has arrived at a conclusion. So the question on whether you should build your own AI tools or buy a purpose-built solution with AI embedded, it often gets framed as like a cost question. But for me, it's something that also needs to be looked at from a risk and controls perspective. If you build something internally, I think you need to remember that you own the governance of that tool entirely. You own the security, the auditability and the ongoing maintenance and updates. And I think, Jason, like you alluded to that earlier, like you own the ongoing monitoring of it, especially if you're the one that built that tool internally. So I'd love to get both of your perspectives, Jason and Travis, on building versus buying AI tools. What are some precautions? And what are you seeing from your customers? Are people building? Are they buying a mixture of both depending on use case? Would love to get your insight.

Yes, Chelsea, I think you hit on at least what my more mature clients are thinking about there. The more mature ones are really starting to think about the risk and controls angle here and the burden and the overhead that, that is going to create on something that might be built internally. This whole topic of build versus buy is actually something that I talk about probably weekly with various companies and so forth. And the reason being is what you spoke about, it's the development process is short. It's getting quicker. It's getting easier. In some cases, it's even avoiding the use of your company's technology practitioners. They're able to do it with [indiscernible] coding or something on their desktop, they're able to produce some pretty good processes and tools and solutions there. But the part that still hurts at least from a finance and accounting perspective or the part that still hampers those type of solutions is all about the controls. It's the security model. It's how do I access the data? How do I build the audit logs in there? How do I build that traceability? And when you start going through those type of processes and start thinking about those features and functions you need to add to it, it quickly becomes apparent why there are vendor solutions out there in the marketplace that companies are using, and they're using it because there's years of research and years of development refinement on some of these features. And that's where I think it's going to take a while for some of these AI solutions to overcome from that perspective. So I -- like I said, it's a hot topic. I'm not saying it's always a buy versus build. There's definitely cases where you can -- where it makes sense to quickly build something and build a solution and deploy it. But there's still a nice spot for vendor-supported solutions, especially in finance and accounting.

Yes. And what I'll add to that, Jason, is I agree, sometimes it's build or buy -- in my experience, sometimes it's also kind of both, right, working with clients who maybe want to build, just as an example, like an AI-enabled reporting solution in their finance org. So instead of people being bogged down in spreadsheets and digging through data, it's like, I want to go to some traditional reporting, but be able to chat with it, talk with it, right? And -- the way those come together typically behind the scenes is it's not fully custom. It's not built from the ground up. It's also not bought -- out of the box. It's like you're buying and choosing the different components that are behind the scenes and the build work is sort of stringing them all together into a unified solution. So maybe there's a layer that's supplementing the source systems and storing some data, then there's foundational LLMs that are tuned to your data. That's another piece of it. There's reporting on top, maybe some of which is more out of the box, some custom. And so the end solution is like we bought a few different pieces and did some build work to bring them together into our end solution. So I think there's a lot of those combos out there. I think the other thing that organizations are weighing that we talk to them about on the build versus buy front is effectiveness of the end solution, right? I don't think there's any perfect answer here, but I think of some deployments we've done, say, in like anomaly detection, looking at transactional data that's flowing into the ERP, for example, and flagging things that look off potentially. Sure, that's a capability that I think is emerging out of the box in a lot of cases, whether it's ERPs or purpose-built solutions or offering that. And that's a great way to turn something on early, maybe leverage tech you're already paying for. Maybe you can just turn it on without any incremental cost. And so you're deploying AI faster and in some cases, cheaper. I have heard the flip side of that is maybe the effectiveness of the results, some noise in the results, which makes sense. These are models that are trying to be effective for hundreds and in most cases, thousands of different organizations data versus maybe a custom version of that, that maybe took a little bit longer to build, maybe it was a little bit more expensive to stand up, but highly tuned and trained to your data and the issues that you've had in your historical data. And so when you right out of the gate, you're getting really effective accurate results out of the AI system. Again, not a clean answer one side or the other. I think building versus buying has pros and cons even within the same use cases, right?

I agree. I do think we're -- what I've even heard though, like with the availability of the off-the-shelf AI tools, you are seeing more and more cases of [indiscernible] coding for accounting and finance. And I think -- when I think about it, depending on the use case, it might make sense. But if it's something that's going to be audited, it really starts to get scary when you think of internal controls and ITGCs. Like I have a former colleague who's CFO tasked him with building their own account reconciliation tool like using Claude. And to his credit, my former colleague with no coding technology background was able to build like what you call the light reconciliation tool, which if you work for a private company, I think it was like really impressive. But I think, okay, this maybe works now. But as you grow, is this something you can scale? And like what are the controls you have in place and depending on who your auditors are and like how -- like are they going to trust your reconciliations and those that information that's flowing into your reconciliation tool. So maybe you save time upfront, but in the end, the more robust third-party tool might have been a better solution, especially when you think of the cost of maintaining and updating the tool. So moving to our third polling question, where does your organization currently stand on AI governance? A. No formal framework; B. Informal policies in place, C. Formal framework and development or D. Fully established framework with Board oversight. And we are starting to kind of get short on time. So I think we definitely want to get into the use cases because to me, that's the most exciting part and probably why most of you are here today is to hear these use cases because this is where the time savings and the quality improvements become real and tangible for the people doing the work. So based -- I'll start with what we're seeing at Workiva and from conversations I've had with our customers. There are several financial reporting use cases for AI that are really gaining a lot of momentum. The first one that we highlight the most and that our customers see an immediate benefit from AI is that they're feeding AI, their 10-Ks and their 10-Qs and their historical earnings call transcripts, maybe and then their dropped earnings scripts and press releases into AI and then they're using AI to digest all these reports and these narratives and then asking the AI to help them anticipate questions, they might get asked by their investors or analysts either during a live earnings call, Q&A or during investor callbacks. So that's why like this earnings release review is one use case we're seeing. We're also seeing a major focus on our teams using AI to enhance or improve disclosure research and also disclosure quality. So in this case, they're leveraging AI to research a data set of pure SEC filings. So these are filings directly from EDGAR and then using this data set, they're summarizing and drafting disclosures using verifiable sources like these SEC filings. And in this instance, it's really important to make sure you have like a pure data set because we found that if you just use an LLM, they'll start pulling statements from a company's website instead of from like a true SEC disclosure. Kind of using that same concept of a custom knowledge base or data set, Beyond disclosure drafting, AI is also being used for analysis and peer benchmarking. So accounting and finance reporting teams can benchmark either their risk factors, MD&A disclosures or a new disclosure for a new transaction against their peer filings and then use AI to identify the gaps and find opportunities to improve their own disclosure quality. So everything I just mentioned are use cases that are already happening within the financial reporting space. Beyond what's happening right now in financial reporting, we're also hearing from accounting and finance professionals about tasks that they really want AI to handle in the future. And so these couple of future use cases I'm going to talk about, they're more complex and they're not readily available yet, but they're definitely getting closer to becoming a reality. One of the most anticipated features or use cases as an AI agent for report tie-outs. This would automate actually like that manual tie-out, the AI agent would check for data consistencies across financial documents and schedules. And so these manual tie-out efforts that could take hours or days, AI could cut down into a matter of minutes and the AI tool would flag either mismatches or anomalies in your tie-out. So then you would know exactly the areas to focus on those areas that are not tying out to the source or tying out internally. And then another big request is for AI to help with the disclosure checklist. This would kind of be like that pure benchmarking tool I just talked about. But instead, you would -- instead of comparing your disclosures to peer filings, your disclosures would be compared to FASB standards and SEC rules and regulations, and it would be compared directly against the report you're drafting. So this would allow teams and AI to flag, identify potential disclosure gaps while you're drafting the report, so you could close those gaps in real time. So these are the use cases that are generating the most excitement in the financial reporting space. Travis and Jason, when your clients come to you and say we want to use AI in our controllership or for our accounting and finance functions, what opportunities are you seeing beyond those that I've just described for financial reporting?

Yes, I'll take this one, Chelsea. The one I'm most excited about and the one that I see my clients get most excited about, it's really around what I call intelligent automation. It's really taken RPA on steroids, if I could use that analogy. We all went through the RPA phase 10 years ago, where we strung together a series of tasks based off a set of rules, and then we quickly realized the limits of that. With these AI solutions now, we're able to take these processes -- these automated processes and really take them to the final steps. We can look at the exceptions. We can do the root cause analysis. The root cause analysis can be done by an agentic process and really start to understand from itself without having to deploy a bunch of rules how to figure out what the root cause is and even starts to suggest to the end user, the accountant what a possible fixes. And I think those type of use cases because it's complete, it's complete, it gets to the end is really what's going to excite the controllers for the next couple of years.

Yes. I think to pile on too much of the use cases, but this is always a popular topic. I'd just add a few more in there, at least what I'm seeing, hearing or somewhat kind of touched on, yes, like I think there's a lot of effectiveness in pointing AI at large volumes of data and transactions that are sort of happening behind the scenes and usually sneak up on the teams we work with at month end when we're trying to close the books, like I gave that GL transaction example earlier, that same type of capability looking for auditees in anything from accounts payable or intercompany transactions, even some of the data you're looking at it in month-end reporting, like that's very popular because that sort of pays off in crunch time when teams are trying to close the books. I think a natural entry point for AI is, of course, anything forecasting, predictive related. So we have some FP&A teams maybe building -- and these don't have to be like big bang deployments either, building maybe a challenger model, a quick forecasting model that they're comparing to their existing process. Are we more accurate, less accurate? I've worked directly with clients and built those type of models to come up with accrual estimates. I think that's very popular. To me, the one that's near and dear to my heart, the area that is anything internal management reporting like insights related. I've worked with clients historically when we're trying to automate some of this stuff, and you can just automate a lot of the upfront steps like wrangling the data, pulling it all together, pulling it up to the surface to give to teams, but then they're kind of complaining like, well, that was all good, but I still have to do the hunting and pecking. I have to write the commentary, explain what's going on. Using Gen AI in large language models and pointing it at large amounts of data, results of financial analysis and summarizing what's going on and having teams and people be able to ask questions of that back and forth, constantly working with talking to finance and accounting teams about that capability where we could stand it up. I think there's huge potential in it, not only to save time, but to make finance teams more strategic going forward as well.

[indiscernible]. I think we could have spent like a lot more time on use cases. So hopefully, we'll have a follow-up webinar just on use cases. until then, I'll move to our fourth and final polling question. This is me, [indiscernible],re you interested in a demo of Workiva AI? And given that we're running up against time, while you answered that, I will go to one final topic. I want to spend a few minutes on something that is becoming an increased focus for everybody, and that's measuring whether any of these AI tools are actually providing benefits to team or organizations. If I flip to the next slide, this is from our Workiva benchmark survey, and it really just summarizes the top ways executives are measuring the ROI of AI -- and what they're really looking at are workforce optimization and productivity, savings from error detection and reduced fines, revenue from improved products and improved customer satisfaction. The one takeaway here is that C-level executives are indexing higher on customer satisfaction and revenue impact, while senior directors and managers are most focused on error reduction and productivity gains on their teams. So if you have any final questions, please submit them now through our Q&A engagement tool. As mentioned at the beginning of the webinar, we capture all questions. And if your question is not addressed during the webinar, it will be addressed later via e-mail. Before we close, we have about a minute left. I want to ask each of you if someone in the audience today walks away with just one thing from this conversation, what would you want that to be?

Maybe I'll go first on this one. Mine would be probably a call to action. It's easy to sort of be handcuffed a bit by all the precursors you feel like you need in place to AI. Like how am I going to pay for this? Does it pay for itself? What are all the governance pieces I need to have in place? All very important pieces, but I think the -- or pioneering organizations are taking kind of this no regrets approach, like these capabilities are here. This is going to be the way finance and accounting transforms over time. We need to get started now. Otherwise, our competitors are going to get started. They're going to be the ones moving faster, getting better insights. And so I think there are very low barrier to entry areas. Hopefully, we talked about some to jump in and get started and start to shape this stuff up along the way. So I would instill maybe that no regrets action-oriented approach on the AI journey. But how about you, Jason?

Yes. I would say the theoretical philosophical ideas that we had 20 years ago about like a touchless close, continuous auditing, a continuous close, whatever the concept was, we're really darn close to it now. We're -- the solutions are getting there. We're able to -- we're seeing companies get closer and closer to those concepts. And I think that's exciting. So we should all be excited about where this is all going and the opportunity it provides for all of our organizations.

No, I think that's great. And I think like you said, Travis, like identifying those like kind of maybe low-risk use cases and just jumping in, especially just for experimental purposes, just because then it kind of can help you open up a whole other level of ideas to be considered. So One final slide before we officially close. As a reminder, we have some additional resources on this topic. Links to these resources are included in the related content section of the webinar console. And with that, thank you, Travis. Thank you, Jason. Thanks for hanging out. I know we went a little bit over, and thank you to all of you for attending today.