Workday Podcast: Why Future CFOs Must Embrace Generative AI in Finance

Bruno Navarro: As we’ve seen in Workday’s recent Global CFO Indicator Report, one-third of business leaders have embarked on their AI journey to boost their organization's efficiency and supercharge their ability to create value. The report called them “AI pioneers” due to their level of AI investment and adoption maturity. 

Among those AI pioneers, more than half of them call the technology a game-changer for the finance industry, and more than two in five respondents expect AI to drive increased revenue and profits. While people are firmly in the driver’s seat of this exciting technology, finance leaders are eager to understand how they can get started. 

Today on the Workday Podcast, we’ll explore some approaches to elevating the finance function with AI by speaking with Riyaz Kasmani, director of data science at Workday and head of the company’s Finance AI Innovation Center. Riyaz, welcome. Thank you for joining us.

Riyaz Kasmani: Thanks for having me, Bruno. It’s great to be here.

Navarro: Could you tell us a little bit about your role at Workday and your role heading up Workday's Finance AI Innovation Center?

Kasmani: Sure. As director of data science for the finance organization, my team and I leverage automation, advanced analytics, and artificial intelligence to identify areas for process improvement, risk mitigation, and streamlining within our business functions. Within Workday’s Finance AI Innovation Center, we aim to establish a centralized pipeline of AI use cases that are specifically designed to support the CFO organization. And the center operates through three core pillars. We have the first one, which is our transform pillar, which prioritizes business process innovation. So we achieve this by collaborating with the finance business to champion AI use cases that drive digital transformation.

Our second pillar, which is the influence pillar, focuses on collaboration with Workday product teams. Here, we work to encourage the adoption of Workday AI functionality and also co-develop reusable assets with the product team. So this co-development effort with the product team fosters the creation of AI features that will ultimately benefit our customers. 

And then the third pillar, I must admit it’s my favorite pillar. Here, we cultivate citizen data scientists within finance. So by empowering finance professionals to explore and experiment with AI and machine learning, we want to foster a culture of rapid ideation and experimentation.

Navarro: That’s great. You participated in the prompt-a-thon led by MIT and Workday today. As a data scientist steeped in AI and ML, what value did you find in participating in the session?

Kasmani: The session provided a unique opportunity to gain insights into the evolving landscape of large language models, especially as it relates to the finance space. So participating in the prompt-a-thon, observing how industry leaders are formulating prompts for their specific tasks gives you inside knowledge about LLM adoption within their fields of work. And also, these sessions, I always find them to be a goldmine for unconventional approaches, diverse perspectives, and usually, some outlier use cases that you may not find easily in mainstream discussions. So for me, it’s all about the creative and the innovative approaches to improve business processes. So participating in the session helped me think through strategic utilization of LLMs within the Finance AI Innovation Center.

Navarro: You lead a team tasked with identifying finance use cases for AI and ML at Workday and then vetting those use cases to find the best pilots to roll out. It’s my understanding that, in your initial assessment, you identified 67 use cases for AI/ML and narrowed that down to just 19 with high potential for implementation. What kind of value realization framework do you use to refine the list of potential use cases to adopt?

Kasmani: That’s a really good question. Our value realization framework for evaluating AI use cases considers three key criteria. The first is a combination of accuracy, quality, and efficiency. So accuracy speaks to how much will we rely on the accuracy of results produced by the AI solution, the quality aspect is about how well does an AI solution meet the business requirements, and the efficiency is about how much time and effort does the AI solution save.

The second criteria that we have in our value realization framework is about scalability. So when we look at these use cases, we think about how well can the AI solution be scaled to meet the growing business needs or even just scaled across multiple business functions.

When you think about finance, there’s a lot of different sub-teams within finance that are using or approaching the same kind of use case, but they have their own unique perspective or lens from which they look at that information, so scalability really is important. And then the final criteria in our value realization framework is about the complexity of the AI or machine learning solution, so how difficult is it to develop and maintain the AI solution? Does it use a combination of foundational large language models, or do we have to build a bespoke machine-learning model that also requires some end-to-end automation? So the higher the complexity, I would say, the lower the cost-benefit in building such a solution. So we have to look at it from that perspective as well.

Navarro: That’s so interesting. One of the use cases you identified as having a huge benefit to Workday is around contract generation, which the Financials Product Advisory Council also worked on. You found that you could save approximately $1.4 million based on the cost of 50 full-time employees for seven and a half weeks. What kind of prompts did you work on around contract generation, and how did those help you think about the value you’ve projected for this case, this use case?

Kasmani: As part of the FPAC group, we used a framework called the COSTAR framework, which is a helpful structure for creating effective prompts for large language models. So in the FPAC session, I was part of a group exercise. So we went with the majority vote in each group. 

Our group actually drafted a prompt for generating board books using both current and historical financial information. We got creative asking the LLM to generate a response in the style of Shakespeare with a lot of emojis. Needless to say, it was a noble effort, even if it was fated to fail. But coming back to your question about contract generation, so after this session was over, just as an exercise, I did use the COSTAR framework to draft prompts for generating a new procurement contract, which is aligned with the use case that we have on contract generation. And the generic results were amazing in terms of the efficiency in creating the contracts. I could just envision the reduced effort it would have of the procurement teams and the ability to flag high-risk terms in the contracts.

I say generic because, of course, in our real-life scenario, you would be using what’s called a retrieval augmented generation approach, where the LLM that we use in our production environment will have access to our organization’s contract database and standard templates. So it will be less prone to things like hallucinations and biases. In terms of value, the increased efficiency gained particularly excited me. It frees up so much time from procurement teams to focus on higher-value tasks that they do not have right now. It’s just amazing the value we would get from implementing such a use case.

Navarro: It sounds like it really frees up staff time to elevate the positions that are already there. Is that right?

Kasmani: Exactly. Yeah. That is correct. So you would want to cut down as much manual-labor-intensive tasks—and also just from a job satisfaction perspective, you don’t want to do too much manual labor. You want to be focused on things like, if I take procurement as an example, strategic sourcing, managing supplier relationships. You want to focus on risk management, spend analysis. I mean, in Workday, you want to be focused on sustainable sourcing, right? So those are things you want to spend more of your time on versus generating contracts.

Navarro: It sounds like that can also be part of a recruitment and retention talent strategy as well.

Kasmani: Oh, absolutely. Absolutely. So not in the contract management space, but we’ve had other spaces where some of the use cases that we’ve implemented, the feedback that we’ve received is it has helped reduce turnover, attrition in the teams. Similar scenario where the teams or the finance folks were doing a lot of manual, repeated grunt work. So machine learning and AI really helps a lot with the quality of jobs and job satisfaction.

Navarro: You also shared with the group today your plan to train Workday’s finance teams on AI and ML as part of the Finance AI Innovation Center charter. What kinds of initiatives are you planning to generate excitement and adoption?

Kasmani: What we plan to do is empower our finance teams as citizen data scientists because we believe that will fuel innovation and future-proof us, give us a competitive edge. We’ll be doing this through a variety of initiatives, starting with hosting workshops and meetings to share knowledge. The workshops will cover topics such as AI basics, responsible AI, and for folks in finance that are more technically aligned, I would say we’ll have technical topics like statistical analysis and machine learning. We are also discussing a progressive skills-based certification program that we will make available to our finance teams for folks that might be interested.

But when you think about these things, it’s not enough to just know the theory. You need the hands-on experience to develop the skills. And to enable skills development, we have a few different initiatives lined up. We’ll be hosting cross-functional gigs using Workday functionality where finance professionals can participate in real-life AI projects with my team. We are also planning a finance AI hackathon event where we’ll be curating balanced teams with a mix of finance professionals, data scientists, and programmers. And we’re also building a toolkit of sorts packed with resources for data exploration, data visualization, and automated machine learning to help cultivate and build our citizen data scientists. And finally, the last one that we have that we are thinking of is doing, similar to what they do in startup culture, there’s this thing called demo days. So we are planning to do finance AI demo days where these citizen data scientists can showcase their AI progress and learnings. That’s the plan.

Navarro: That certainly sounds exciting. Earlier, you mentioned responsible AI. How does responsible AI play a role in these initiatives that you just talked about?

Kasmani: As part of the single pipeline of use cases, one of the first things we do as part of our intake process when we prioritize use cases, we have a few questions specifically looking at the ethical considerations of the use cases. We want to make sure that these are not used in decisions that might be impacting any specific individuals. That’s a very, I would say, specific approach when we look at these particular use cases.

Navarro: Given how finance leaders are focused on risk and governance and compliance, how does AI play a role in that area?

Kasmani: Well, it will depend on the specific implementation of these use cases and how you do that. Let’s take an example of a use case where you’re looking for high-risk terms, going back to that contract generation thing. If you’re looking for high-risk terms in contracts, AI really makes it very easy to be able to do that kind of analysis of large documents that would be difficult for a finance professional to do manually—or, as they would do it in the current [way] of doing keyword searches. So AI really helps you with those kind of scenarios.

Navarro: That’s great. You mentioned having humans sort of in the driver’s seat here. There seems to be checkpoints with people having oversight of the work that AI generates, the end product. How human-centric will AI’s implementations be in the future?

Kasmani: It’ll be a learning process. I would say, at least for the foreseeable future, from my perspective, you will want to have a human in the loop until we are comfortable that the results of any given machine learning model or AI is reliable, accurate, and predictable. But as we get better and better with these AI models over time and the business gets more comfort on those, you will start seeing some of those checkpoints, the human checkpoints, come off, at least in some common scenarios. Of course, there will be always these high-risk scenarios where the decisions impact certain individuals or have very, very high risk. Those, you will still have a human in the loop. But as we get more comfortable with some of the non-riskier scenarios, I think you will start seeing a lot more adoption of what’s called LLM-based workflows or LLM agents that not only help you do content generation but, based on that, you can take certain actions beyond that automatically.

Navarro: Is the idea of a zero-day close within sight? Is that within the realm of possibility?

Kasmani: I would say yes, for sure. I think there’s a lot of interesting use cases that you can apply to some of the processes that happen right before a close. So I would say it’s a very real possibility that you would have a zero-day close because there’s a lot of current processes that I see being in finance. I see the accountants, the finance professionals do right before close. Some of it is a lot of manual work, and that is exactly the area where an AI can fit in and help remove some of those things. But then there will be another interesting phenomena, though, when it comes to zero-day close. What happens is there are so many manual processes that happen to make that happen, it does not give enough time for finance professionals to do what I call the real value-add work, like looking for those outlier transactions, being able to do that qualitative analysis of financial information. They are really crunched for time right before the close, and I think freeing up the bandwidth will allow for better risk management and a much more detailed qualitative analysis of the financial information.

Navarro: Great. Riyaz, thank you so much for taking time to speak to us today. It’s been a pleasure speaking with you.

Kasmani: Same here. Thank you.

Navarro: We’ve been speaking with Riyaz Kasmani, director of data science at Workday and head of the company’s Finance AI Innovation Center. Remember to follow us wherever you listen to your favorite podcasts. And remember, you can find our entire catalog at workday.com/podcasts. I’m your host, Bruno Navarro, and I hope you have a great workday.