- 10 minutes ago
Join Fortune as we explore how leading CFOs are rethinking their approach to the future of work in the age of Agentic AI. We’ll explore the journey finance teams are on to embrace AI and when and where it makes sense to use AI agents to accelerate automation, the ROI tradeoffs of deploying human vs. digital talent, and the upskilling strategies these CFOs are using to optimize their workforce for the future.
Category
🤖
TechTranscript
00:00Hello, and welcome to the latest iteration of our Emerging CFO series,
00:11Unlocking Injective AI to Drive Business Innovation. I'm Cheryl Estrada, Senior Writer
00:16at Fortune and author of the CFO Daily Newsletter. I'm joined today by my colleague, Jeff Colvin,
00:22Senior Editor-at-Large at Fortune. In our Emerging CFO series, we dive into topics
00:28most interest to CFOs. We explore their evolving roles within the organizations and the skills
00:34necessary to climb the finance ladder in any organization. I'd like to extend a tremendous
00:39thank you to our partner Workday for their support on this series. In just a moment, I will kick off
00:46today's conversation, but first, some brief housekeeping. Today's gathering is on the
00:51record. We will be taking audience questions that will be posed to our CFO discussion leaders later
00:57in the program. Please use the chat function to submit your questions for both sessions,
01:02and we will do our best to ask as many questions as possible. With that, I want to hand the program
01:08over to Jeff for our opening conversation with James Glover, Principal of Finance Transformation
01:15and AI and Innovation Leader at Deloitte. Over to you, Jeff. Thank you, Cheryl. And James,
01:21let's get started. You see a lot of CFOs across industries, quite a few of them. What are they
01:31wanting help with you from with regard to agentic AI? Yeah, Jeff, first of all, thanks for having me
01:41and getting into the question that you've asked. Really, when CFOs are coming to Deloitte and asking,
01:46they're asking largely three questions. First one is, what does the ROI profile look like for AI?
01:53We're at the beginning of a journey of adoption, right? Technology moving very quickly, but adoption
01:59moving maybe a little slower. So CFOs are very curious on sort of what are the ROI that we're seeing
02:04from clients that are early adopters? The second question is that I call it sort of the where question,
02:10like where should I be looking at deploying AI and applying AI within my finance function?
02:16And the technology being so powerful, I think the opportunities are fairly significant and there's
02:22a lot of breadth, but CFOs really want to know, where should I start? And then the final question,
02:26which is obviously something CFOs have to consult with CIOs and tech leaders, is what's the path to
02:33enablement? Should I be using vendors? Should I be building custom tools? Should I be using cloud tools?
02:38Like right now, that landscape is highly complex. So really three questions, sort of ROI,
02:44where do I think about AI within my finance organization and the things that we do? And
02:49then the final one is, what's the technology path to enable it, to bring it to life?
02:53Yeah. Well, initially with the regard to ROI, one could argue, I suppose, that at this stage
03:00is a little early to be looking for the return in the R in the ROI, or are some of them already
03:08seeing a return?
03:10Yeah. Jeff, I would describe it as sort of green shoots in that the early movers are starting to
03:15see some real benefit in terms of the deployment of AI. And again, the deployment of AI can take
03:20many, many forms. So we should probably discuss that, but they are seeing early green shoots and
03:26early benefits and also like where they think the technology is going to go next. But with a lot of
03:31the largest statements that are being made out there about like what's going to happen to finance and
03:36how big or small it's going to be in the future, you know, I would sort of put most of those still
03:41down to somewhat of a speculative position. Do we believe that technology is going to get us there?
03:47I think we do. Do we believe that it's going to be a long journey? Absolutely.
03:52Well, that makes a lot of sense. Now, the second thing you mentioned, where, where do we start?
03:57I bet that's a very important question and you must hear it from everybody.
04:02And we have two types of sort of asks around that question. There are clients that say,
04:09you tell me where to start. You've been doing this with other clients. You tell me where to start.
04:14And then there are clients that say, well, you know, I need to figure out where to start in my
04:19organization based on what's the right move for the objective I'm trying to achieve.
04:24And I prefer the latter question, because I really think that application of AI within your
04:29organization needs to align to what is your overall strategy for finance? Like, is this an efficiency
04:35play? Is it an effectiveness play? Is it a control play, an engagement play? Like, why are you doing
04:40it? Why are you making the change? And then look at your entire sort of finance services or capabilities
04:46and say, where do I have alignment between achieving my objective and the application of AI
04:51AI and go after those areas? I'd call those the most impactful areas, but in a more meaningful sort
04:57of ambitious way. We are finding companies that are going after things that are use case by use
05:02case way are struggling to see how it turns into sort of real benefit for the enterprise.
05:08So it's a larger picture than a lot of companies are imagining. How much time does it take to do this
05:17kind of larger thing so that then they can really see something that they get back?
05:24And Jeff, this is a great question, because now I'm going to talk about sort of the types of AI that
05:28we are seeing in the marketplace, because it'll give a sense to how quickly things can happen.
05:32You know, a lot of our clients, 80% plus have deployed what I call user productivity tools.
05:37Think Microsoft Copay, Claude AI, Gemini, you know, there's a number of them that can help the user.
05:42But the onus is on the user to sort of drive efficiency into their job. We have predictive AI
05:47models that help with forecasting. We have generative AI solutions that are more sort of
05:52singular in nature, looking at unstructured data. And then we have the more exciting side, which is
05:57largely where this conversation is going, where we think about enterprise solutions and agentic
06:01platforms. And how can I sort of stand those up and roll on sort of use cases or roll on
06:07capabilities to drive, you know, efficiency, effectiveness within a finance organization.
06:11When I think about user productivity, you can install that, right? It takes no time,
06:15but you actually have to train your people to use it. Otherwise, they're going to sort of treat it
06:19like Google search. And then on the other side of the equation, where you've got these agentic
06:23platforms, you know, they take real time to implement and to stand up something in a kind of
06:28a POC, you know, that's in the that's weeks, but to move a POC to production, and have your users
06:34ultimately start to kind of, you know, augment the tool and themselves in the workplace, that is a
06:41little bit of a longer journey. Like we're seeing those journeys taking sort of north of six months,
06:45maybe up to nine, maybe up to 12, depending on the complexity of and the ambition that the client's
06:50going after. The technologies, obviously, the timeframes in terms of getting to outcomes is going
06:57to come down over time, as people get more, you know, competency in deploying AI within finance.
07:04Yeah. And so you mentioned a number of different ways that people are using it.
07:09Is that the correct order in which to do it? Or can they go right? Can they start with
07:15the big picture, the real agentic big picture stuff?
07:19You know, my latest thinking around this is sort of come at it from both angles. So,
07:25you know, deploy user productivity tools to your finance organization. So the finance organization
07:30can start getting comfortable with, you know, what is good prompting versus bad prompting look like,
07:35you know, how can I create some efficiencies and some capacity in my day job? Because a lot of
07:38finance organizations, frankly, you know, have capacity challenges. That's kind of one of the
07:43biggest things that we see. And ultimately, these user productivity tools can create some capacity for
07:48the employee base. But also look at the bigger solutions that take a bit more thought, you know,
07:53you have to put a business case together, you have to think about how you're going to bring it to life
07:56for your organization. Those are going to be very, very beneficial to the enterprise at large.
08:02Whereas user productivity tools, whilst they're going to be very benefit to the employee,
08:05it's not necessarily something an enterprise is going to be able to harvest as a benefit.
08:09So I do think that the strategy where you take a view on both sides to get your
08:14employee base sort of more native to use AI tools, but at the same time, you're thinking about sort of
08:20bigger, bolder plays that can be really accretive and from a value perspective to the finance organization.
08:25Yeah, a related issue here that I'm sure is going through the minds of all CFOs is, look,
08:32this is the finance function. You can't have mistakes. You don't want mistakes anyplace. But in
08:40finance, obviously, it's really, really important that everything be done really exactly right. And
08:48it's got to be a little worrisome to some people who are just getting into this. What do you say to
08:53people? Right. And that's, you know, if we think about adoption and finance, accuracy, guardians,
09:00you know, conservatism, those are good behavior traits for finance organizations. So they need
09:06to take that to these AI tools. Look, there are, you know, when AI and general AI, you know, first came
09:13to everybody's attention, there was a lot of talk about hallucinations. And frankly, if you ask an AI tool
09:18sort of question 10 times, you may get some fluctuations in the types of answers you get,
09:23which is not good for an enterprise. Enterprise wants to see the same answer over and over again,
09:27if you ask the same question. But those problems are being solved with how we think about the
09:32solutions. When we use AI tools and LLMs to, you know, write SQL and retrieve data and analyze that
09:39specific set of data, like we tend to get a lot more accuracy, a lot higher degree of accuracy.
09:44The final point of this is that AI implementations are going to require a phase that we're not
09:49necessarily comfortable with in the technology world. And I call it the augmentation phase or
09:54the human in the loop phase. So there's always going to be a point where tools have to go into
09:57production for humans to use them for the AI to sort of get reinforcement learning and get better at
10:03what what you're asking it to do. And that's a kind of a new phase of a project or a delivery
10:10that, frankly, we haven't seen on traditional technologies, where once you test, you put into
10:15production, you're kind of you're kind of good to go. But this, this world is going to be slightly
10:18different. And I think, again, the pioneers in this space are starting to see, you know, how to
10:24implement these tools, but also, you know, where it's going to take them as an organization.
10:30Yeah, well, and this gets to another issue that's going to apply to everybody, which is,
10:33what new skills do people in the finance function have to learn? Obviously, they will be different,
10:41I assume, depending on where they are in the finance function. But what can you tell us about
10:47the skills that will have to be added? Yeah, I mean, great, great question. I mean,
10:53we do think that the kind of the make the skills makeup of finance over the next 10 years is going
10:57to change, you know, dramatically, we have been saying to CFOs, for the last 10 years that you need
11:02to start complementing traditional finance skills, like MBA finance and accounting skills with,
11:06you know, skills that are more tech focused SQL, Python. Now AI comes to the sort of market in the
11:14forefront. And, you know, the ability to sort of write Python and SQL scripts is obviously going to
11:20get democratized, because you can do it through these AI tools using natural language. But I go back to the
11:25technology that if you don't have a just as some foundational understanding of how these technologies
11:31really work, you're going to kind of like struggle to be very effective in how you work with them. So
11:36definitely going to see more education and skills around the kind of tech domains, maybe not as deep,
11:43more breadth, so that folks can understand how to work with AI tools in terms of accessing data and
11:49collecting data and analyzing and data, looking at data quality and all those type of things.
11:53And really, for the workforce to trust AI, which is going to be a big thing that has to happen for
11:59adoption to really progress, they really have to understand how the technology works. So, you know,
12:04we do think that more and more technology skills are going to be required within the finance
12:10organization to really leverage some of these very powerful technologies that are coming to the
12:15organization soon. Yeah, we see across industries and across functions, that AI, at least
12:23now, is taking over some entry level jobs. Are you seeing that in finance? And if so,
12:34I've heard people wondering, how will people learn the finance function, if they are not themselves
12:42doing some of these entry level jobs? It's a great question. One thing we have to always remind
12:48ourselves is that the change is going to happen slowly. So we're going to have to respond to,
12:53as the technology, as we adopt the technology, we're going to respond to sort of what does that
12:57mean for the workforce? What does that mean for how we structure finance? What does that mean for
13:00how we create career paths, you know, and all those elements, like, we're going to, we're going to have
13:06to work that out. Like, I don't know if anybody has the perfect answer to exactly where this is going
13:10to take us from a, from a talent and machine perspective. But we know that change is coming,
13:15and we're going to have to be responsive to change. I will say in other functions outside of finance
13:19and other enterprise functions, like IT, marketing, servicing, even HR are seeing a lot more adoption.
13:27And to your point, there's a lot of junior staff and IT organizations, junior software developers that,
13:32you know, companies feel they don't need anymore, because they've got AI tools that can increase the
13:36productivity of senior developers. So they've seen a lot of progress. And that's creating a lot of,
13:40like pressure on the CFO, because CEOs are going, well, we're getting benefits in other areas,
13:45why are we not getting it in finance? And I think largely the patterns and the type of work in
13:49finance, anything volume and transaction based can be automated using traditional systems. So then
13:55you're left with, well, what are the, how does the, what is the part of AI is going to help us kind
14:00of achieve within the finance organization around what humans do today? And again, that, as you
14:06mentioned, the focus on accuracy, somewhat of a conservative view, like we need to get things
14:10right first, you know, those are the things that are going to have to be navigated for finance to
14:15kind of get to sort of mass adoption. As well as thinking about user productivity tools, I go back
14:21to them, I feel like a lot of my clients have them. And they have, you know, been getting some benefit,
14:27but there's a lot of potential there for companies that are more progressive in training their
14:31organization on how to use these tools and connecting the tools to, to datasets within the
14:37organization. You know, let me see if I'm correct. I am. I can't believe it, but I think we're out of
14:45our time. I, there were lots of things to talk about here and we covered a lot of territory in a
14:51fairly small amount of time. So I just want to making, be making sure that I'm right, but I am. So
15:00James, thank you very much for the look into the agentic in agentic AI. Um, I'm thrilled now to
15:09invite back Cheryl, kick off this next part of our, uh, uh, program.
15:15Thank you so much, Jeff. That was a great discussion. And to discuss further about how to
15:20drive business innovation with agentic AI, please welcome Michelle Chung, chief financial officer and
15:26chief operating officer at green light, Greg Mestel, chief financial officer at webflow and dad
15:33stricker, chief financial officer at in bricks. Thank you so much for joining us today.
15:40Um, I'm going to kick it off with our first question. Um, how about I toss it to you that
15:46how is AI reshaping how your company or even how your teams view work and productivity?
15:52So, uh, two ways. Um, so Enric's mobility data, we have over 50 petabytes of data across 146
16:01countries. We collect about 45 billion data points a day. So a massive amount of data to set AI on top
16:08of, and we're seeing tremendous results and how cities are able to utilize that along with AI when
16:14you layer it on and pull the insights out of it from a finance perspective right now in the short
16:21term, it's been improving the accuracy of our reporting and our forecasting. So just an easy
16:28example of our ARR, we dropped a tool on top of it called cluster and it analyzes the last three years
16:36of our CRM data and is now forecasting our ARR of nine months. Our accuracy has gone up to about 95%
16:47utilizing the tool with just very limited input from us. But one of the things it provides is the
16:54transparency to how it gets there, the accuracy rate over time versus our traditional forecasting.
16:59And so it's easy for us to monitor it and see how well it's doing and what it's doing. Longer term,
17:06I think we'll transition those things to some tools in Workday that are building the same capabilities.
17:13How do you come up with the use cases for AI?
17:18So not to get into the ROI side too soon, like what are the things that we're having the biggest
17:23challenge with? And so that ARR forecasting was something we were struggling with and we just
17:27started looking for tools to go solve it. So what we generally do is where are the areas we have
17:32the biggest problem? And is there a tool today that we can leverage? Hopefully something that is not
17:37overly expensive that we can test out and validate the results on.
17:43How about you, Craig?
17:45I think, first of all, thanks for having me. And for those who don't know Webflow,
17:50we're a marketing platform that brings marketers, designers, and developers together
17:54in one place so that marketing teams can build, manage, and optimize their websites.
18:00So in my organization, I'll first talk about just some of the use cases that we're using
18:07from a finance perspective. One of the challenges that my teams have, we're a very
18:11under-resourced team, probably like most CFOs think, is they spend a lot of time asking generic
18:18questions that the organization has. So what, you know, can I do this within my travel policy?
18:25What's my procurement policy? So it's a lot of questions about the policies and stuff like that.
18:30So we've used a GBT to, and enabled that experience so that people can ask those questions and really
18:37unlock the time that my team has to spend time on real value-added activities versus doing that.
18:45The second thing of, and this is kind of where you kind of get started to alluding to that point,
18:50is we were struggling with kind of, you know, just insights and data. And, you know, the prior
18:56speaker talked about the fact that we've, we got our data model in a place where we could chat with
19:03our data. So non-technical analysts on my team can go use natural language and actually chat with the
19:10data to start to get insights out the other way. So those are two, two places where we've gotten,
19:17gotten started. Now, Craig, at your company, you pretty much oversee the strategy for, for technology.
19:26The, the IT organization reports to you, is that correct? That is correct. Yes. So your, your scope
19:31is, it's certainly finance, but it encompasses on how everyone is going to use AI. So, so there must be a lot
19:39of collaboration with that. Yeah, it, there is a lot of collaboration and we've, we've, we strongly
19:45believe that it's, you know, AI will make people, the team, the human capital we have at the company
19:52better. And it really will supercharge the, the amazing talent that, that we have. What we've done
19:58in the company is we've set it up in a way where we have both tops down and bottoms up ways in which
20:05we drive adoption. So first from the tops down perspective, our CEO has been very clear with
20:10people about what the expectations are. We've partnered with our HR team to actually create
20:15a framework that goes into people's performance reviews. So there actually is clear criteria,
20:22transparent criteria of like what's expected in terms of AI, just like, you know, every performance
20:27view has cross-functional collaboration as a, as a pillar. This is says AI as a skill. So we've been
20:33very clear about that and we're, and we're rolling that out. And then, and for the first time, we'll be
20:36putting that in the performance reviews at the, at the end of the year. That, and that's an example
20:40of how we're collaborating there. We also have a centralized governance committee that's tops down
20:45that unblocks people, make some tool selection. So we talk about, you know, agentic platform.
20:50We're developing like which agentic technology are we going to use and roll out to people.
20:54But most importantly, bottoms up is where, where, where greatness happens. And we have folks
20:59within our engineering teams, our go-to-market teams, our support teams, and then within our
21:05GNA teams who are the champions. And they actually do amazing things, bottoms up. And when they get
21:11stuck, then they go to that centralized committee. So we've set up a lot of good governance and
21:16frameworks to, to make, to make it successful at the company.
21:19That's really interesting. Performance reviews will reflect that. And it's the outreach to each
21:25division of the company. Thanks for sharing that, Craig. Michelle, how are you approaching AI at
21:32Greenlight? Yeah. Hi, thanks so much for, for having me. And it's great to be here, Cheryl.
21:37So I, I would, I would say there are sort of three angles that we approach it and, and Greenlight,
21:43just for the, for the sake of, for the audience, we are a consumer subscription, family finance,
21:50and safety company. So we have millions of families that, that sort of use our app on a regular basis.
21:56And so some of the ways, Thad talked a little bit about volume. We don't quite have petabytes of data,
22:01but we have a lot of transaction data that we, that it can help us review, ingest, summarize, and,
22:09and we think about things from a risk standpoint. And, and this is like truly money movement that is
22:14going across many different sort of like millions of surface areas. And so this is, this is helpful
22:21to us from that standpoint, internally, and from an operation standpoint, I would say on the finance
22:26side, there are a number of things that, that we use AI to do to help us review contracts. We have a
22:32number of enterprise partner contracts that can be pretty voluminous. And, and we want to make sure
22:38we're reviewing terms and making sure it's reflected correctly within our accounting practices. I'll say,
22:44generally speaking. And so that's, that's a large sort of internal area that we've seen
22:48a good uplift in terms of, in terms of using this technology. The other two facets that I would say
22:54we look to use it as one, because we are a consumer facing app, we actually try to use AI in our product
23:03with customers. And so one example for this is that our kids, we have, this is a family, family finance
23:10app. And we have a, we have a product that allows kids to invest in different stocks, actually. And
23:18what it does, the AI has sort of delivered content across any of the stocks that our platform can
23:24offer to the kids that help them make the decision and go through the logic on whether this is a good
23:30investment to make from understanding if this is a leading company in a thriving industry to here's the
23:35financials and financial profile. And then whether the stock is overvalued or undervalued, it's a good
23:42company, it has strong financials, but is it expensive? And it sort of walks them through that
23:46process. And AI was a big tool to sort of help us launch this and release it. And our customers really
23:53use it and love it. And then the third angle, I would say is really just democratizing, you know,
24:00a finance person can perhaps be a technology person, a technology person can maybe be a marketing
24:05person. Like, I would say all of these tools, both from everything that I think has been spoken
24:12about in terms of this panel, the quicker and easier sort of like user productivity tools has
24:17really enabled, you know, I would say a flatter sort of surface area for everybody. You don't have,
24:24we don't have to be ad siloed, generally speaking.
24:26You can all upskill together, so to speak, on AI and learn. Exactly. Exactly. Oh, you all mentioned
24:35use cases. Now, I'm wondering, were there any use cases that didn't exactly work out the way that you
24:41thought it would, and you kind of had to pivot? How about you, Craig? Yeah, so there's a few that,
24:48you know, the version one where we tried to use, you know, ChatGVT as a kind of like a junior analyst,
24:58it's not ready for that yet. You know, it's not great at math. So, you know, and so that use case
25:06didn't work. And that's why we pivoted to more of the chat with your data use case. The other one I'll
25:10share is my FP&A team has been iterating on automating their flux and variance analysis.
25:18And the earlier attempts at that, actually, when you put it in, you got different answers every
25:23time, and that's not going to work. So we learned a lot. We had to iterate. We had to test it. We had
25:29to continue to experiment. And now we actually have the ability where it, you know, when you put it in,
25:34you get the same answer every time. And of course, someone's checking it right now.
25:37But I guess the point I would share with CFOs is if you fail the first time, that's okay. Don't give
25:45up. Keep going. Because in my experience, in that FP&A use case, we finally got to something that is
25:51now adding a ton of value. Thank you. Thanks for sharing that. If you fail, don't give up. Just try
25:56again. Try something different. Exactly. Try something different. Yes. So I'll give you an example,
26:04just an early, simple use case. We were considering doing something at Enrix. And I was talking to our
26:10GC. And I was wondering, like, I wonder if there's any case law out there regarding this. And I went
26:15back and I used Chad GPT. And to be fair, this is like two years ago. So it was early released.
26:21But I asked it for case law. And it came back, like, oh, my gosh, it was amazing. All the law
26:27references, the appeals court rulings. And I gave it to our GC. I said, oh, my gosh, look, it's fantastic.
26:32Like, all this stuff's already been decided. She comes back a week later. Where did you get this?
26:36I can't find any of this case law. And I went back to Chad GPT and said, hey, like, where was this?
26:41He goes, oh, I made a mistake. That's actually not real. And one of the things I found out,
26:46we'll probably talk about AI prompting, that you still is a good practice of when you ask your
26:50question, you make sure you tell it, if you don't know, don't make it up.
26:54Or make sure you get a source file, you know, and then click on that source file. And then,
27:02you know, like, yes, there are many instances of this. I totally agree.
27:08Jeff, do you have any questions for our CFOs?
27:11Well, there is a whole thing that is the human part of this within the finance function.
27:19Does the, I don't know how to put it really, but does the
27:26whole feel, I'm trying to think of the right word.
27:37But what changes within the way people
27:41work with each other? Are some people real experts with AI and especially agentic AI?
27:52Others are not and are needing to learn more. How are you dealing with that? And, and, you know,
27:59I want to know what everybody says. Michelle, let's start with you.
28:02I, I, so it's, I, I kind of, I think my answer to this is, I think it's, of course, I think every
28:13human is different and every person is different and sort of the reception to change, to new
28:19technology, to doing things a different way and, and sort of the value that, that, that I bring to
28:25the table and, and that I can bring to the table prospectively, it can be, it, it is very different.
28:31And I actually think the way to motivate people to sort of embrace and, and, and to think through
28:38like, what, what is, how can this be helpful to me, to the company, to my function, to my team?
28:45It, it can be a different argument, I guess I would say, depending on who the person is, like,
28:49like understanding the person, understanding their motivations. Is it more about building an asset for
28:54them so that they can be competitive in the marketplace? Is it, you can help your team out in a
28:59better way and drive and deliver a ton of impact? Like there, there's a number of, I guess, different
29:03angles. And, and I don't know, I think this is just another vector by which, like, it goes to the
29:10performance views that, that, that, that Craig was talking. It's just another vector that we need to
29:13use to sort of support and, and motivate our employee base.
29:19Yeah. Thad, what about you? This is a cultural thing that we're talking about here, really.
29:23Uh, what have you noticed?
29:27So I think, you know, we're humans and there's a natural change cycle that goes on here. It does,
29:34I was talking to somebody the other day. It doesn't matter if you go back to like early days,
29:38right? Switch from horse carriages to cars, no phones to telephones, to cell phones. Every time
29:44you have people like, oh no, it's never going to happen. Nobody's ever going to use it, but it's
29:49going to happen, especially with AI. And now that you layer on the agent, agent piece of it.
29:53And so I'll echo what Michelle said, getting people engaged earlier and sooner and try to find as many
30:00opportunities as you can, um, for them to engage with it. We'll take advantage of just about any
30:05opportunity work gives us on an early adopter program, because it's not a formal implementation.
30:10It's just a way for us to engage with them, share our thoughts on a product, but also expose our people
30:16to the capabilities that are coming down the road. And then, you know, just be thoughtful about how
30:21it's going to impact your team. Um, we're not a large company, but the larger you get, you know,
30:27we've talked about it's, you end up automating initially things that are very repetitive, manual tasks,
30:34and just start planning on how that's going to impact the organization and the people. To the extent that you
30:40can manage that via attrition. So the more things make it a positive approach, it's going to be a
30:45well more received than if you wait till the tail end to have some of those conversations.
30:51Yeah. Makes a lot of sense. Greg, what have you seen?
30:56Yeah, I think, um, you're right. Change management, just like any, any technology or anything is hard.
31:03Um, so the way we've approached it, I talked about the AI fluency framework. Um, so like level one is,
31:10you know, not using it and I don't believe, um, so you don't want to be level one. Uh, so,
31:15but level two is I'm using it for personal productivity. Um, I'm doing, I'm doing things,
31:20I'm learning. Now, when you get to level three, you're starting to get to a place where mentorship
31:24becomes a key aspect of that, where you're actually out there helping others. You're maybe a
31:31champion in your department. So you're actually seen as one of the leaders. The other thing that
31:35we've started to do, and this is actually hard because the tools themselves actually don't really
31:39have the APIs and the technical back ends to be able to do this, but we're trying to be transparent
31:45and measure things. So we have a metric internally. It's one of the ways we measure, um, adoption is
31:51it's not only did you use it in a month, but how many days in a month did you use it? And you should
31:56use it just like an email tool or, uh, or your, or Slack. We use Slack internally like that. My,
32:01I always say like, I want to have the same usage as Slack in terms of depth. So we, we measure that,
32:07um, you know, CFOs, we love measurement. Um, and you know, by having that now we can have a,
32:13we can, you know, and we haven't done this quite yet, but you know, again, we're, we're rolling this
32:17out, but like we may give it to managers just so they have the information and they can have a
32:20conversation and say, Hey, like, let's talk about this and let's see there. So, uh,
32:25transparency is very important. I think here as well, as we're trying to drive adoption.
32:29Yeah. And what about that, uh, that level one people? I mean, if there still are any, uh,
32:35what do you do as the first step? I can imagine that if you just showed them
32:41what it can do, they would have their eyes opened, but are there people who are really resistant? And,
32:48um, Greg, you were mentioning it. I mean, I, in any organization, I would kind of bet there would be,
32:54but I don't know. What do you find? Yeah, I, I think in every organization,
32:58there's going to be some folks who are resistant to technology. Um, you know, we've taken an approach
33:04to meet people where they're at. So we don't assume that, you know, everybody knows what to do
33:10or how to use it or have a use case. Uh, so we try to do show and tells. Um, you know, we, we,
33:15we have sometimes executives actually showing them how they're using, like, I'll give a personal
33:20productivity example. I have a, an agent that runs every day that sets, that prepares me for my day.
33:26Uh, and it's been an iteration to get it to be good by the way, but like I, I show that off to
33:30people. I also trained, I've trained my GBT to be a C, to be the CFO webflow. So maybe I'm not the
33:37CFO webflow. It's actually the, the chat GBT who's my assistant there helping me out. Um, but I've,
33:43I've actually trained it so that when I ask it questions, it has context to be able to do it.
33:47So giving those examples to people, it gives them like very tactical places to start.
33:54Yep. Yep. Sounds good. Um, Cheryl, back to you. Oh, thanks Jeff. Um, so I'm going to ask the ROI
34:03question, uh, how you each are approaching or what your perspective is on that. Because James Glover,
34:09uh, earlier in his talk with Jeff said that, um, some, uh, companies are already seeing green shoots.
34:15So just curious where your, your perspective on that. Sure. I, I don't think I have anything,
34:21um, that different from what, uh, James Glover was saying. It's, it, it is, it's, it is, I would
34:27love, I would love to do a discounted cashflow, project out the, the, the, the, the potential,
34:33um, uh, inflows from, from any project and, and, and NPV to today and, and compare that against
34:39what the cost of the AI tool might be. Um, I would say that, that it's probably not worth
34:45the administrative tax and burden, uh, just yet in terms of, in terms of trying to instill
34:50that rigor in, in every sort of POC or every experiment that we do. I think there's, there's
34:55certainly, again, because I think there are, um, um, on the internal operation side, sort of,
35:02you can look at volume and you can sort of project out, look, if we don't do this and we're not able
35:06to scale, um, this is what it would take to do without that tool. And, and it can be a little
35:11bit more straightforward, um, uh, in terms of, let's say back of the enveloping, something like
35:16this versus by doing this, I can get this many more kids engaged. This many more kids will, um,
35:23will sort of transact. That's going to drive my engagement. There may be referrals from this.
35:27There may be more growth from it. Like this is a, I will call it a little more squishy and, and,
35:31and sort of based on the assumptions that you want to put in. And so this is something that I,
35:35I, I would, we, we, we, we, I think we have to play it forward a little bit in my mind before I
35:40feel, you know, like I, we could stamp something that says, here's the ROI on it.
35:45Thanks, Michelle. How about you, Craig? Um, I, I'm going to break it down into revenue and cost.
35:51I think, uh, I'll be pretty black and white on this one. Uh, so on the revenue side, uh, you know,
35:56we have AI in our product as well. Uh, you know, and it, it does things that I think humans can't do.
36:02Like, you know, we optimize web pages at scale and personalize that experience, um, really well.
36:08So, you know, I measure the revenue that comes from that. That's one piece of it. Now, when it
36:12comes to productivity, um, we're doing a lot in our go-to-market teams in terms of, of allowing our
36:19sellers to be more efficient. Um, and what that means that turns into revenue. So for example,
36:26to prepare for a sales call used to take 60 minutes, an hour. And with AI now and the processes
36:32and the, some of the agents we built, uh, it now takes four minutes. So if, if you can get that type
36:38of productivity gain, then I can say, okay, now my revenue per rep is going up and now I'm actually
36:45talking about revenue. Um, on the cost side, I think it's harder. Um, you know, there's some easy
36:51one. So customer success or customer service is very, a very well-established use case.
36:56I can measure that really well. And when I talk to my board, I say, look, this is what my,
37:00my cost of goods sold line item is going to look like because I can project what I believe that,
37:07that, that, that ability will be. Um, that's an easy one. Engineering is harder. Uh, I have
37:12qualitative data on what I think my lift is in engineering. Um, it's harder there. Uh, and GNA
37:18is very hard. And you, you mentioned earlier that you, you do have some, uh, metrics, um,
37:24that you use in regard to AI. Is that correct? Yeah, we do. So we have a metric around adoption.
37:31That's one thing that we're measuring. That's like the, what I call the input metric. And then
37:35the output metric is we actually, um, like I gave the example of the, you know, the, the, you know,
37:4160 minutes to four minutes. So actually on a per pro program basis, we are measuring success
37:47on the, using the technology to just, just to justify the cost that goes in. Uh, I'll be
37:52honest. We're being, we're being more open about it. Like we're letting people experiment. So we're
37:56less ROI focused on it in terms of, cause we want people to use it. And I don't want to be
38:01restricting that at the top because you won't get the outcomes yet, but we do measure like those
38:07outcomes. So as this, the go-to-market teams are doing things as our engineering teams, we're
38:11measuring all these things to try to figure out how effective our investments in those technologies
38:15are. Thanks a lot, Craig. Uh, how about you, Ted?
38:19I'll follow on it, Craig. There's two areas, right? If we're talking about the company from
38:23a revenue perspective, it, you know, it's easy to see when you're talking about the amount of data
38:27we have spread across an additional called 650 data points, it'd be extremely challenging for a
38:34city planner engineer to try to rummage through that, to find the right insights and for them to
38:40be able to use natural language to just ask the question, like, why is this traffic signal
38:44taking later? What's going on at this intersection and having AI do that for them? That's an easy
38:49business case. Um, on the, on the back office and the GNA, we've actually been able to find
38:56some pretty simple solutions that have worked out great. You know, I mentioned the ARR forecasting,
39:01um, the efficiency that we gained out of that, not only for an accuracy rate, but just the time
39:07it took to perform it was easily paid for. Cause the tool it's a low cost tool on the other ones.
39:18I'm less inclined to try to go build something ourselves. We're just not big enough. And on top
39:23of that, if I can leverage work day, I get two things. One, I get a tool that's already been
39:29tested and built, but on top of that, the way that they structure this, right. And so you get the
39:36tenancy privacy, but their AI models get the learnings across all these companies.
39:41And so it's not just now my process improvements I'm learning from it's, it's across multiple
39:46companies and to be able to leverage that in a meaningful way. I don't know how you do that.
39:53If you're trying to build on your own, especially for the back office, you know, you're talking
39:57customer facing, it's a whole different story, but it's hard enough to create an ROI on the back
40:02office. So the extent that you can leverage something else out there and somebody that's doing the work,
40:05I'm going to choose it every time.
40:09Thanks so much. Thanks for sharing your perspectives. I have a quick question for
40:14Michelle. Michelle, you're both the CFO and the COO. So just wondering how these roles inform
40:21how you approach AI.
40:25I don't think it's, it's actually, I would say it's, it's very similar to what Craig had mentioned
40:31in terms of some of the ways that he, and he is the benefit of, of seeing it across the whole
40:35organization, given, but like I, um, on the operation side, um, there are things like customer
40:42service operations, um, card operations, risk operations, compliance operations. And so, and all
40:49of these have tickets, have volumes and, and are very metrics driven. Um, and this, I do would,
40:55I would say, um, it, it actually lends itself a little bit more to, um, um, uh, probably potentially
41:03more rigor in a lot of the decision-making for, for tools and whether it makes sense. And I do think
41:08there's a lot of low hanging fruit here because of the volumes and because of the transactions and
41:13because it lends itself to metrics. It's, it's a, it's a much easier business case often to justify,
41:19um, particularly on how are we going to scale? Like we are a high growth company and, and we want to
41:26make sure that we're, we're always as, as nimble as possible. And this gives us comfort, um, um, putting
41:32in AI tools and agentic AI tools, uh, that we're better positioned for that than, than other, than we
41:37otherwise might be. Thank you. Uh, well, thank you all who have already submitted questions.
41:44We will be coming to those shortly, but before that, we have a question from Tim Wakefield at Workday.
41:51Tim. Thank you very much, Cheryl. It's a really interesting conversation, everybody. Um,
41:56I have a question that's not so much focused on return on investment or actually on the technology,
42:01but far more focused on trust and the governance that needs to be applied around this conversation.
42:07So agentic AI promises like a technology shift. That's going to, that's going to move finance
42:13way beyond automation towards true autonomy, where systems will act and make decisions on their own.
42:19And so I can imagine that one of the questions that finance leaders are wrestling with today is no
42:24longer can we automate this process? And instead it's, can we trust the machine, um, to act on our
42:30behalf? So my question to you is this, what do you consider to be the most critical non-technical
42:37guardrails that an organization should put in place to breach the agentic and trust gap and
42:42therefore accelerate the adoption of autonomous AI? How about we start with you, Dave?
42:48So I think two things. One is within the application itself, the ability for a human to be reviewing
42:57those decisions and the transparency around what's happening and validating that what, what it's
43:04choosing what it's doing is valid. The second, I'm going to tie it into more broadly in the, the product
43:12that we're using. I have to trust my provider. And I think you have to step back and look at whatever
43:22you're considering purchasing and not just look at that product, but look at the organizations that
43:26you're considering partnering with, um, I've worked with a lot of different, you know, systems across
43:32the years. Um, and you know, not to put a plug for work date, but we've had a lot of opportunity to work
43:41with them, not only on just purchasing and employment from a product perspective. And when you see the level
43:49of care and diligence that these product managers are putting into the products and really trying
43:57to understand the problem and solve it, but with the guardrails around it, for me, it builds that
44:03trust. And so whoever you use, regardless of work date, just make sure you're partnering with somebody
44:08that you can trust that is going to instill those guardrails and the things that you need
44:13to feel comfortable, not only with you, but your auditor going forward.
44:21How about you, Michelle?
44:24Yeah, this is a hard one actually for me. Um, I thought I, I, it's, um, I think I have a slightly,
44:32um, different view. I'll say this, like, I, it's not that I disagree, but I, I sort of the,
44:38when I think about AI and sort of the level, like every single time we're, we're putting
44:43in a tool, we try, we ask the AI tool not to train on our data. Like there's like this,
44:48there's sort of an inherent, like what is ours, what is yours and what are you going to use that
44:53is ours to sort of better your, your, like, and so I, I, I don't disagree with you. I want to trust
44:58the provider to be very clear, but it sort of only goes so far. I don't want it to, to go. And so
45:02like for me at the end of the day, um, Tim to, to your question on, on sort of like, how do I,
45:07how do I like, how do I bridge this gap for trust? I actually, and I, and you know,
45:13I actually think it elevates human value, uh, to some extent. I, I think it is the judgment and
45:19to some extent, the subject matter expertise that is required for the human, like just like
45:22Thad's example, like if you don't know case law and you're not a lawyer, like it's going to be
45:26really hard for you to catch these things. And so there's still some, like the, the human mind that
45:31is the supercomputer and pattern recognition, uh, sort of machine that it is like, like somebody who is
45:36sort of, um, you know, steeped in this, in that subject matter, um, um, area would be able to be
45:42like, okay, this, this, I don't know this, like, how can that be? Like, I, I know the resources that
45:46I can use to check it. I might know people who might have been involved in this case. Like there
45:50are some, some ways that they can sort of confirm. And, and, and I think there's that. And then the
45:55process that, that you use to sort of make sure there are lots of checks and balances, essentially
45:59one person's view.
46:06Do you have anything to add, Craig?
46:08I think they made most of the points that, that I was going to make. I would just reiterate
46:13Thad's point that your security teams are going to be the most skeptical people in the world. Um,
46:20and actually some major providers, I won't name them, uh, when we actually did the work to,
46:24cause you have to integrate these agents with different systems for them to be effective. And
46:29you'd be surprised how some of the security models broke down. So they, these agents would get access
46:35to data that they shouldn't have access to. So I would be wary of those things. I think this will
46:39get better over time. Um, and then in terms of the trust thing, I, I really love what Michelle said.
46:45I, I think, you know, I think us as, as humans, but we're here, we were humans and we, we know our
46:51craft really, really well. And the agent should be making us 10 X better because we know our stuff
46:59really well. And if that's the mindset we have going into this stuff, um, you know, we'll be able
47:04to catch just like a, if you get assigned something to a junior FP&A person and it comes back and you
47:09know, it's you look at it, you know, it's wrong. Um, you just know, right? Like, I mean, you know,
47:14I think that's as CFO, like your gut, like you, you know, so I think that's, that gives you,
47:21give you comfort, um, to experiment with these tools and see what can happen and what works in those
47:25processes. Well, thank you for those thoughtful responses. Um, and thank you, Tim. Uh, Jeff, I
47:36think we have some questions for the audience that we would like to ask. Yes, we do. And I'm just, uh,
47:44going over them now. Uh, here, here's one that just comes up. How often do you expect the AI vendor
47:52to use their, I just jumped, hold on. How often do you expect the AI vendor to use their experience
48:01and help you in building a business case that projects the ROI? Well, it's not, uh, it's not asked
48:14of anyone in particular, uh, Michelle, what do you think? Um, for the most part, I, I, I actually think
48:23it is very beholden on the vendor to, to help us with the business case. We may have data on how many,
48:30how much volume, how much time per, you know, whatever it is, ticket. We like for our own
48:36application and our own use that we'll use to sort of plug into their model, but they're going to have
48:40benchmarks for what it is that they can help us improve, whether it's a workflow or, or, or I'm
48:47going to cut this time down. And so I think it is very much a partnership with the vendor. And then
48:52frankly, you've got, you come out of that exercise for the business case and you've got some great
48:56metrics to hold that vendor accountable to. Um, and so this, that to me is, you know, so then you
49:02can sort of do a check-in at a, on a quarterly basis or, or whatever it is. Sorry about this to,
49:07to confirm that that works. I'm, I apologize. Anyone else want to take a, I'll just, I agree
49:18with Michelle. I don't much add there. I just say that the vendors will come in and do enablement
49:22sessions for you. Like we just had a few of our vendors come in and, um, actually do 45 minute
49:27show and tells and training sessions to, to meet people where they're at. So, um, they're a great
49:32resource for that because they want you to drive adoption because they know that it's pretty competitive
49:36out there. And there's, for every solution that you implement, there's probably five others.
49:41Yeah. I think all the same. Yeah. I think it's just table states. If your vendor partner isn't
49:44willing to do that, then I think you need to question whether this is the right one for you.
49:49Yeah. Uh, here's another question. Uh, interesting. This says, uh, determine the types of workflows
49:59that are actually needed. Deterministic workflow, deterministic workflows, for example, are by design
50:08agentic within guardrails. True agentic means the AI can make the decision for you, which isn't always
50:18necessary. The first question is, do I need AI at all? Well, I'm not the one to ask, uh, Thad, what do you
50:31think? So I think there's use cases out there, easy ones around AP, AR, just to get super transactional,
50:40approving and processing an expense report and assigning the right jail accounts automatically
50:49and routing it to the right person and escalating if necessary and checking its policy and making
50:54decisions off of what it's finding. Right. So I've got now both AI as well as agentic as it makes those
51:00decisions and routes it and does a different thing. Um, yeah. Well, people are already starting to do that
51:08today. So I don't know why we wouldn't. Craig, Michelle, I don't know if you got a different
51:13opinion. Yeah. Yeah. I, I think the, I think we're at like, I see it as a couple of things. I think
51:20we're in terms of phases of AI, we're very early. We're in almost like a phase one where it's improving
51:26personal productivity. There'll be a phase two where we improve, uh, productivity to the point where like
51:32we can replace people in the finance org. And then there's going to be a set of things that we can't
51:36even imagine yet, you know, um, that it'll be able to do. Um, so I think, again, I think we need to
51:43keep experimenting and, um, just see what we, I think there's, you know, just trying to think what's
51:50that's not possible today that could be possible with AI. I think, I think it does unlock a lot
51:55of things. Um, I'm trying to think of a good example in finance. I have an example from what our
52:00product, our product does, which I already gave in marketing, but I'm trying to think of a good one
52:03in finance. Yeah. Um, Craig, let me ask you something else. Uh, to what extent are you
52:11using AI to write code, you know, to write software for you? Yeah. Um, so, um, every engineer in,
52:24in my company, uh, and every IT person in my company has, uh, access to AI and it's an expectation,
52:31not required, but an expectation that they use it, um, qualitatively, uh, it's a 20% lift in
52:38productivity, um, just by having them, have them do that. Um, you know, you know, and my past company
52:46was in, was in DevOps and I think there's a question is, are we getting high quality code?
52:52Um, or is it, or is it not high quality code and, and do you need better testing and stuff? I think
52:57you need that too. And I think AI can help with that too. So I think, I think software engineering
53:01is it, I don't, I don't believe it replaces the software engineer because there's, there's still
53:07a lot of thinking and design and critical out criticality. Actually the vast majority of a
53:12software engineer's job is not coding. It's actually doing every other things. So, so I, so we're using
53:18it. I think it superpowers our engineers. Again, I think it's a combination of taking the human and
53:23human capital that we have, which is incredible and making them even better.
53:28Yeah. Uh, Michelle and then said anything to add to that?
53:31No, we're, we're, we're using it in a very similar way that Craig described as well. There's
53:36one software engineer, not actually our company, a different, a different company, another software
53:40engineer I know, uh, who told me that he has like three different screens that, that he's like
53:45got that running code all the time so that he can review. So he basically has like several junior
53:50software engineers sort of working and, and writing code, and then he's reviewing it all
53:54the time. So it's a review and approve rather than drafting. And so I, I think it goes to
53:59the, how do you 10 X, how do you use it to 10 X your, your, your superheroes? Um, they're
54:04super, super heroes now.
54:06Yeah. Yeah. I, I think every engineer right now is using it. What's been interesting to see,
54:12and he heard an earlier speaker talk about finance and that knowing Python and SQL. And what I've
54:20seen it do, like you can use co-pilot's analyst feature and just working on large spreadsheet
54:26models or data sets and watching it go build the Python and SQL to do the things that need to be
54:32done. And so I think you're seeing everyday finance people use it to write code that they
54:36probably would have never done before, even if they understood it, it just makes it that much easier.
54:40And something that's, uh, related, um, that I don't know that we've, uh, addressed head on,
54:51which is, um, with regard to how agentic the agentic, uh, AI is going to be, are you allowing AI
55:02to disburse money by itself? No, no, no, no, no. Well, that's kind of what I thought. Can you foresee
55:16the day? Yeah. I, I, I, within certain guardrails, I think absolutely you'll get there, especially on,
55:26you know, low dollar, you know, low risk transactional type of things. Um, I can certainly
55:33see that being a case. Yup. What do you think, Greg? I, I think we, as CFOs right now who, you
55:41know, know the number of approvals to release funds out of our bank accounts right now, just can't see
55:45it. But again, going back to my point, I think there'll be a world, I think that's right. I think
55:49it'll, you know, you'll start doing the small little AP stuff and then eventually, you know,
55:54the bit that we'll think, well, we're going to figure it out, uh, you know, so, but we're
55:58not there yet today. Yeah. Yeah. Violent agreement with that. Yeah. I, I, I, I'm just, I'm kind
56:06of using this as, uh, uh, uh, something that will be very important when it finally happens.
56:15It is not happening yet. Yeah. Cheryl, we're almost out of time. Have you got anything else?
56:20Uh, I just really enjoys this conversation, really insightful. And, um, I appreciate you
56:28sharing, uh, what's going on at each of your companies and your perspective on AI. But, um,
56:33yeah, I think we're at time, Jeff. Okay. Well, look, uh, thank you again to
56:38our fantastic discussion leaders here. And thank you the audience for joining with us today.
56:45A special thanks to our partner Workday. Our next virtual emerging CFO conversation will
56:52be January 27th. We'll be discussing governance and building trust in the AI era. Please keep
57:01an eye out for your registration details. We hope to see you there. So long.
Recommended
1:14
|
Up next
58:01
11:34
18:52
16:13
16:03
23:26
12:58
1:13
5:50
Be the first to comment