AI Adoption in Finance: What CFOs Get Wrong (And How to Do It Right)

00:55 Megan: Today, I'm joined by Ashok Manthena, CFO and Chief Finance AI at Chatfin, where he is helping reshape how finance teams operate through the practical use of artificial intelligence. Ashok works closely with CFOs and finance functions across FP&A, treasury, tax, and controllership, helping organizations automate manual work, improve decision-making, and unlock new levels of productivity through AI. He is also a recognized speaker and thought leader focused on the future of AI in finance.

01:29 Megan: In this episode, we'll explore what AI adoption really looks like inside mid-sized and large companies, where finance teams are seeing the greatest value and how leaders can move from experimentation to meaningful transformation. Welcome, Ashok.

01:44 Ashok: Thanks, Megan. It's really nice to be here.

01:46 Megan: Nice to have you here. I'm really excited about this conversation. Of course, AI is at the forefront of almost every conversation these days.

01:55 Ashok: That's right. It's everywhere, literally. It's anxiety-inducing sometimes, but there's a lot of excitement around as well. And we're all in the midst of this whole evolution that is going to happen for humankind.

02:08 Megan: If I just think back to maybe two years ago, I think that was the first time I'd heard of ChatGPT. And here we are, like, two years later, and literally AI is doing amazing things.

02:22 Ashok: Right. I actually started working on AI in finance way before ChatGPT came into the picture, and it was different times. When we talked to CFOs about AI, people used to ask, "What are neural networks?" How does it even work? Now things have changed after ChatGPT. So the last three years have been totally different, and now I think we are in the peak excitement of AI.

02:44 Megan: I'm excited to see where it's going to go in the next three years. So as you look back maybe towards the beginning of your career, was there an early moment when you realized that technology was going to fundamentally change how finance teams work?

03:00 Ashok: I think by the time I started working in finance, there were already amazing ERPs. There are tools for planning, the tools for ERPs, the tools for AP, AR, and everything in there. Once I saw the complexity of finance and accounting processes, I realized you actually need the technology to manage it. Now I think we just take all the ERPs for granted, but the complexity of the human operations and how they actually made it really simple for us—we don't usually think about how data can be stored or what all the workflows are. We only think about, oh, can I use this specific ERP and get my things done? So I think technology, it's not a question about whether it's going to change or not. It's very much needed. And now with AI, I think it's another big technology leap that is happening at this point of time. Again, the question is about whether it is needed, whether it is transformational. It is actually very much a necessity for us to run these complex operations.

03:57 Megan: And you work with CFOs across multiple finance functions. So what are you hearing most often from mid-size and large companies when it comes to AI right now? Are people scared? Are they excited or confused? What do you hear?

04:11 Ashok: I guess it's everything, but at least a discovery problem. What I'm saying is you don't have to tell a person what AI is and what it can do. Already CFOs and all the finance leaders have seen the potential of AI in other fields. Not in finance, but they have seen it, how it changed our daily lives, even writing emails and writing documents, and how it became easy. And now when they can extrapolate and think about what AI can do to their classes. A lot of companies have started experimenting with it, and most of the finance leaders right now are in a position where they're thinking we should start using AI. I don't think in the last eighteen months I've met the CFO, and they said, "Oh, we don't want to use AI." They are a little skeptical, but everyone is like, we want to bring in AI at various levels. And the excitement is actually at different levels in different industries and different companies. There are few companies where they're all in, and they are building their own products. And we can talk about that, like, building internally with AI. And finance teams are doing it. That's the exciting part. And there's another end where they are using vendors to get AI into their operations and thinking about how that can introduce efficiency and productivity into their teams. So different spectrums of things happening. Again, internal build—it's a long discussion if the internal build can be sustainable. How can they get all the feedback from the users? If you're just getting feedback from one finance user, how do you get all the feedback to improve that specific application or a product? And if you're a vendor, how are you dealing with security? How are you dealing with the flexibility to actually hyper-customize to your needs? So those are the questions that are being involved right now. There are answers for everything, but it now depends on the company, on the CFOs, how they are thinking about AI, and how fast they want to implement. Data has been a big question now. There are few CFOs who are in wartime now, not just for their operations but even for usage of AI. I'm sure, Megan, you have spoken to a few of them. They're like, today, we should start using AI. Find the ways where we can get all these efficiency gains. Find out all the manual tasks that we're doing and see if we can use AI and get those things done. So that kind of a rush is already there in a few of the companies, and a few of the other companies are actually going in a much more calculated way.

06:34 Megan: I used to think that finance was kind of slow to adopt new technology, but these days, it almost feels like they're kind of on the forefront when it comes to technology and AI. Do you feel the same, or from what you've seen, are they still pretty slow to adopt and change?

06:50 Ashok: I think of the companies that I work with; they're really fast. They are going fast. They're making faster decisions to get AI into their process. Once they got AI, they're thinking about how we expand this or how we make sure the specific capability. So I don't see a differentiation between finance and other functions at this point of time in terms of the speed. But, again, some industries I've seen are really slow in adopting AI. It's again would be technology. For example, if you're using a really good ERP or any of the systems that have good REST APIs, it's really easy for you to integrate AI into those systems. And if you're using little archaic systems, then it gets a little slower. So those are the things that are slowing down a little bit in various industries. But, overall, I think we're way ahead in AI adoptions in finance when compared to last year.

07:42 Megan: And what do you anticipate the benefits are going to be for those that are moving quickly and adopting AI and evolving? What do you think the major benefits are going to be for those companies?

07:54 Ashok: I mean, we all know it's the efficiency gains, and more manual work can be done using AI, but those are the benefits that we can get using AI. But what is the benefit of moving actually faster to get this? So, Megan, as you know, I work with mid- to large companies, and large companies have more complexity because there are people, there are systems, there are processes, and there's compliance. If you're a public company, there's more. So for any transformational technology like AI, you cannot do a big-bang approach in a complex process. It's very risky. Because at the end of the day, you still have to close your books. At the end of the day, you need accurate quarter-end 10-K reports. So you can't disturb that process just thinking of efficiency. For any CFO who's thinking, I want to get a change, their probably the second priority is AI. The first priority is always continuity of the business. So what happens is every company needs to go try AI in their process, how it works, and what the changes are they need to do and what is the change management they have to go through. There will be few missteps. How do you correct it? How do you get to the maturity level? And every company has to go through the cycle. There is no shortcut to it. There is no blueprint that you can get from another company and start using it. You have to go. You have to do the baby steps. Go fail. And I'm not talking about the technology. I'm just talking about the change management that happens within your processes. Are you going to move people, or do you have to reorg some of these finance functions? All those things need to be evolved, and that is very specific to each company. So when you are actually moving fast, you are going through that cycle before anyone else. So you will be in a better position in a year or two years than the rest of the companies, where they are now going through that cycle. So that's the biggest advantage that companies get. So these companies, which are moving fast, will probably fail in a few of the use cases, but they will have much more successful use cases and efficiency gains and productivity. And the biggest advantage I always see with AI is it's not about the reduction of human bonds in the process. But particularly in finance, there are a lot of things that we don't do just because we don't have time. We have to do month-end close. We have to do all these audits. So all this work that we have to manually get done, we don't have time for the rest of it. We don't have time to think about what really impacts my profits. What can I do? What is the insight that I can bring in so that my leaders can improve this? Those are the things we are not able to do because of the time crunch. I think this is where once AI takes over a lot of manual work that we do in accounting and finance, now resources can be much more valuable. Now thinking about what else can be done so that we actually support the whole company.

10:41 Megan: And it used to be that the order of doing things was you had to do it, like, work on your process and clean it up and fix it before you would implement a technology. Otherwise, you would just be polishing up something that's not really pretty. But these days, are you seeing AI helping to improve the process, or is it still the case that you really need to think about and clean up your processes before you think about implementing AI?

11:09 Ashok: Right. So I have a thesis for this. I've written an article about this earlier, and probably for the last twenty, thirty years, we've heard this. If you have a bad process, don't put automation on it. You're just making the bad process run faster. I'm sure, Megan, you heard this numerous times. But with AI, now the automation part has become way cheaper, way cheaper than what we have done earlier with bots and with code. Now I think we are talking about one hundredth of the cost, considering the human cost involved, considering the speed and everything. Now if you start changing your process, that means you are allocating resources to change your process first and think about AI doing it. You're spending considerable time and effort changing your process and then putting AI on top of it. The total cost would be way higher than what you can actually achieve by just putting AI on your broken process. I'm not saying you should just put AI on your broken process, but there are a lot of things now that have become way cheaper to experiment with. So even if you have a broken process, you throw AI at it. See if that solves your problem. I'm not talking about getting wrong data, but if you're getting accurate results, but the process is little, you know, you're doing it in multiple ways. Still, if AI can do it, you can still achieve the same results. So the AI is going to redesign the process, and, again, I'm talking about mid- to large companies; I don't think we should just boil the ocean and think, I'll just clean up all my data. I'll clean up my process. Rather than that, you can start introducing AI into various processes. See how those systems are working, and then talk about the systems as humans and the data and all the systems work. You can actually see how things are evolving and then make changes rather than going through the whole process of transformation.

13:04 Megan: Really interesting. So when organizations say that they want to adopt AI in finance, what do they often misunderstand about what successful adoption actually requires?

13:15 Ashok: I'm sure people have understood this, but most of the leaders actually think this is going to be a short-term process rather than an ERP implementation or the multiyear finance transformations. If you look at any large companies, ERP implementations usually take a couple of years. Because you have to just; it's not about technology. You have to validate all the data. You have to do parallel runs during month ends, quarter ends, and year ends and make sure everything is right, and then you take it to production. With AI, it's actually the same thing. Though it's not as big a scale as changing your ERP, you're still taking each and every process. Now you're rewiring with AI, and then you have to test it. You have to test it extensively to see how things are working. And then you could say, okay. Now AI can take over, with humans in the loop, and you test more when you are in production; when you're running it, you are seeing if anything is happening. And when you can think about it, do I really need a human in the loop, or can this operate more autonomously? But for these things to happen, it has to be a step-by-step process. It can't be a big bang. And we have seen few cases where it just disrupts your current operations. And everyone in accounting and finance has to understand that we support the company's business. We support the revenue generation. We support the expenses. So something wrong with the accounting and finance team is going to disrupt the whole operation of the company, and that is not the goal of doing an AI transformation. So we should be much more careful in taking those steps and adapting AI into the process. And that's what we keep telling all the CFOs as well. You start small, but you have to start now. You don't have to think about a really big project. Start small, but then think about how you can expand, and then the learnings will really help you to go exponential in how you can use AI in your processes.

15:07 Megan: Yeah. I'm sure that's great advice because if you want to start off with some grand project, it's probably somewhat overwhelming for a lot of CFOs and maybe causes them to actually never take the first step. So where are you seeing the fastest and most practical wins for AI inside of finance? Is it FP&A, treasury, controllership, tax? Where exactly are these quick wins?

15:32 Ashok: We're seeing everything from AP process to AR month-end close, controller tasks, tax and treasury and reporting, and all those tasks. But I feel like reporting is probably an easy win, but I don't usually see a big ROI in using AI for reporting. People may disagree with me. But I just say, in a lot of companies, and again, mid- to large companies, people have already streamlined a lot of reporting activity. And there's a reason why they are that way because we need controls. We need transparency in how things are done. So a lot of people jump into AI, and they say, oh, we are generating these dashboards and reports. They are nice, but I don't think they are a great ROI. I'm not saying you should stop it. You should do it. But still, it's not a great ROI use case. Really, the biggest ROI comes from the activities that we do manually at this point in time, even though with all the systems in place, there are still a lot of activities that are done. It could be accruals. It could be a reconciliation. It could be an AP process, account receivables. Just as a few examples. If you have a person who's actually sending emails to all the vendors asking them to send invoices. That's just manual work. You send emails, and then they respond. They're going to say, "I'm going to send it tomorrow," or, you know, "We are dealing." And then, again, someone needs to go respond to it. All this process of information exchange is way slower and manual. However, we let AI do that process where it automatically knows who to email. When they respond, it automatically knows what to do with that data that is given in that email. And then is it an accrual entry that goes in there, or do I need a human to actually know about this information? This is just an example. And thinking about reconciliation work, there's a lot of reconciliation we do manually even with a lot of systems right now around still things done manually in Excel, and then you upload them and you're certified. I don't think that needs to be done manually anymore where AI can actually do it. It can also generate all the intermediate data to give you the final results so that it's easy for you to verify all the intermediate data that is generated. So rather than you going and running the data, running the reports, getting the data, calculating it, and putting it in Excel, now AI does all that work and gives you various traces of it so that it's easy for you to verify. So great wins for this, and I've also seen a few of the use cases. Some of the use cases are very specific to industries where you do that only in that industry. It could be an insurance industry. It could be a medical device industry. Those are the processes where you have to go verify contracts and verify a lot of customer data before you can say something about it. All that information is a great win because till now, those processes were not automated. They're being done manually. You can actually pick those processes, and now you can use AI to solve them.

18:27 Megan: And you mentioned that you've worked with companies of different sizes. So, obviously, mid-sized companies have fewer resources than large enterprises. So how should they be thinking differently about AI adoption so that they can compete effectively without overspending?

18:42 Ashok: We have seen mid- to large, when I say mid-size, anything way above probably a $100,000,000 revenue. When a team is smaller, it's actually easy to go adopt a new technology or build on it. And that's what we have been focused on. Now we have multiple teams, and we don't know what the other team is doing for the same issue. And someone from IT, they just don't want multiple tools in the domain, so they are like, okay, can we standardize it? So there are a lot of conversations that happen. Mid-size companies are where things can actually move faster. As I mentioned, actually adopting AI is way cheaper than what we all assume. We all thought it was going to cost like an ERP, and you know how much companies pay for ERPs in mid- to large companies. Now it is way cheaper than having a new ERP implemented because of the human costs involved, testing, and also business resources to subscription costs. Now rather than thinking about how much it costs, I think every project as we do it, I think we should start thinking about ROI. And also, as you start taking these initial steps, it'll become very clear to you what ROI is and how we pick use cases. One of the ways, and of course, we learned this the hard way, is when we talk to a CFO, we always tell them, "If you're going into the AI route, pick the most complicated use cases in your first phase." Don't pick nice-to-haves. Those are not going to set the pace for your company in terms of AI adoption. Pick the hardest use cases and try to solve. And that's what we help them do as well because that actually sets the whole pace in the company in terms of once you take the most complicated case and you automate it or now AI is doing it, that becomes much easier for even champions to actually expand the AI capability.

20:27 Megan: And you just touched on this a bit, but larger organizations obviously are dealing with legacy systems and more complexities. What do you see slowing those larger organizations down the most when it comes to AI adoption?

20:40 Ashok: It's the usual culprits. In a larger organization, there are more teams and more people, but what really helps is having an AI champion at the top of the organization. I've spoken to a few AI-champion CFOs of these really large public companies who are 100% into AI, and they rally their teams to find ways they can implement AI. It's not letting them know to leave security and everything aside. They are saying we need to figure out the use cases where we can implement AI safely and accurately, and now it's your job to go figure it out where we do it. And they are rallying their teams to do it. So having a top AI champion like a CFO or a CEO really helps for the whole company to actually go after this. Because even if the CFO or a finance leader is very much into AI and they have a mandate to use AI, we'll need to make sure how that mandate is actually flowing down to all the levels. How are you taking the ideas from top-down? Are you taking ideas from the bottom up and seeing how things are working? One of the biggest limitations as the company grows is no one in the company actually knows an end-to-end process. When I'm saying end-to-end, even if you think about the whole accounting data to finance, like how the data flows from revenue to various systems, various teams, and then finally goes to the reporting, no one actually knows the whole end-to-end system to think about, "Oh, we are doing this way." Can we actually skip these parts? Can we do this way? So that makes it really complex for someone to think through where we can use AI to speed up things for the whole end-to-end process. That's why large companies have traditionally used vendors to do it, to map those processes. But now with AI, things have become easier to take some parts of it, and you can make it faster. And then you can also think about it overall at a high level. How do I take it from end-to-end and make it really easy? And we could take help of vendors to do that, or you can have an internal team that can go think through each and every process and see where it can be booked.

22:46 Megan: And I've heard from lots of CFOs who are concerned about data quality, controls, and security. How should companies approach governance while still moving enough to capture value?

22:58 Ashok: Right. I've seen this problem with many companies now. People are building white-coded apps with the company data, and they want to launch them. One of the complaints we hear is that who is going to ensure the data is right in all these apps? Who is going to ensure the data is accurate? I mean, I can build that; say, I'm a finance leader. I built an app. Who's going to test it to see if it is actually giving the right data? Someone needs to go test it, and then you have controls. So that's the problem with now; it became easy to build, but there's always that problem: Who's going to ensure that everything, the data, is accurate in there? So making sure, even though giving that flexibility to end users to build things, also having controls on testing and using it for reporting. I mean, for example, if I build my own, we can say I can use it for my needs. But if it needs to be more public, more communal within the company, then that needs to be tested. So having these guardrails, having those controls is very much necessary at this point of time, considering how easy it became to build different apps using AI.

24:03 Megan: I'm curious, what do you think is more valuable, an AI tool off the shelf or building something internally? And maybe it depends on the process or the company, but I'm just wondering what your thoughts are on something that you buy versus something that you build.

24:21 Ashok: I think whatever I say, probably people will choose based on their needs and based on their priorities. At this point of time, I see there is lot of inclination to build internally. It's because everyone is motivated to build. And also, there is fun in building things. If you're a nontechnical person, if you're a finance user, and you can build a dashboard without having the Power BI skill, or you're building a quick app that can do your AP, those are the things that are exciting people a lot. So they are stretching, and they're saying, let's build things internally. We can do it, which is great. There are a lot of simple apps that you can build. You can start using it. But there's a downside to it. As I mentioned, the simple apps you can build within a company, and then you can start using them. But as the complexity increases, all the questions come in. Who's going to optimize it? Because the person, a non-finance user who built it, I don't think he will have time to optimize the whole flow. So then that needs to be done again by a purely technical person or someone in their team taking ownership of it. So building internally, though it seems exciting, as the company size is larger. So I always say if you're a smaller company, I think it makes total sense to build internally. But if you're a mid- to large company and you are in accounting and finance, again, your primary goal is not to build. Your primary goal is business continuity to make sure your data is accurate and make sure your reporting is accurate. So doing that and also thinking about your processes and where AI can be used probably gives greater value to your company and to your job than just building something and not going the total end-to-end of testing and productionizing the stuff. So there are problems with the internal builds, to summarize. And as a company grows bigger, it gets much more complicated to build internally. And one more point is about the feedback that I touched upon earlier. Let's say if I'm building things for multiple customers, then I'm getting feedback from multiple customers on how to optimize a specific workflow. But if you're building that same thing internally, you are probably getting feedback from only one user, and that user as well is very busy doing their month-end flows or quarter-ends and audits, so your feedback is very sporadic. How do you even ensure that you have enough feedback for improving the product? So that's actually a major problem, a major setback for building things internally.

26:51 Megan: And when you think about AI, how do you think it's going to reshape the skill sets needed in the finance team over the next, let's say, three to five years if you can imagine out that far?

27:02 Ashok: Again, this is my thesis because we are not sure what's going to happen in the next four years. Things can change very rapidly. For a finance user or for an accounting user, you do need exposure to technology. You do need exposure to AI, but that's not going to be something that will differentiate you from other people. The biggest differentiating factor would be domain knowledge, and we are going back to the basics of it. If you're great at accounting and finance and you know the specific industry that you're working in and you know what your company is doing and what the factors are that impact it, that is going to be the superpower for an accounting and finance user, not the AI part. And I'm not underestimating the AI impact. I'm just saying when everyone learns AI, it's going to be like an ERP skill. Every accountant needs to know how to use an ERP. Even though you use one, you can easily go and adapt to the next one. Now it became a very common thing among all the accountants and finance people. AI is going to be the same in the next few years, but what will differentiate is that the domain knowledge and your creativity and your analytical skills to ask the right questions to AI, get the answers to prioritize things, and then give it to leadership or give it to teams that actually make decisions. That is going to be a superpower.

28:23 Megan: Last question, but looking at the top of the finance and accounting function, how do you see the role of the CFO evolving during that same period of time?

28:35 Ashok: I've been thinking about this a lot, and it's not just about the CFO. Any leader. Let's say if I take a CFO of a billion-dollar revenue company and if I give him the most amazing AI agents to run its operations, will he actually run all these agents by himself in the next few years? The problem with that is, again, not the AI agents, but I always think, as humans, we have context problems. That means our context is limited. You can't have multiple things going on in your brain at that point in time. We always compartmentalize things, and we do it. And we are taxed by the context switching as well. So one person running a lot of things is, I think, going to be a limitation. That means we still need multiple people, multiple smart people in a company, to run finance operations. Or even if it's a different kind of supply chain, you need multiple people to run it. And a lot of people are assuming it's going to be only the CFO, and they will be able to run it. But just based on what we have seen so far and extrapolating how things will change, I think you still need multiple people. You still need specializations of how things work and a special function that really happens so that they do it perfectly well with the help of AI.

29:52 Megan: Ashok, thank you so much for your insights and for taking the time to be here with me today.

29:57 Ashok: Thanks, Megan. It has been a pleasure talking to you.

30:00 Megan: And thank you all for listening. We'll see you next time on CFO Weekly.