Agentic AI in Finance: From Manual to Autonomous CFO Offices | The BizTech Pulse Podcast

Krishna Hari (00:06)
Welcome to The Biztech Pulse Podcast — where technology, strategy and leadership converge to shape the business of tomorrow. Today’s episode takes us inside the CFO’s office, where a new wave of agentic AI is quietly transforming how finance operations run. To unpack this, we have Vikram Gollakota – Vice President of Global Channel and Strategic Initiatives at HighRadius.

HighRadius is a pioneer in autonomous finance platforms that optimizes order to cash, treasury and record to report functions for enterprises around the world. Vikram brings a unique perspective at the intersection of AI, automation and finance transformation. So let’s dive right in and explore how agentic AI is enabling CFOs, especially in the enterprise market as well.

Welcome to the podcast Vikram.

Vikram Gollakota (01:00)
Thank you, Hari.

Krishna Hari  (01:00)
Let’s start with — What is AI in simple terms?

Vikram Gollakota  (01:04)
OK. So let’s go back to the basics.

So I’m going to break AI into 2 broad categories. One is called machine learning and the other is called large language models. So what is machine learning? Machine learning — and by the way, when we talk about machine learning, the actual algorithms that are used today in machine learning were written by Alan Turing — this is like back in the 60s. If you’ve all watched some of the movies, probably have seen him breaking some codes. And so this is way back multiple, multiple years ago where algorithms were invented. It’s just that in technology now, the compute required to go through all these permutations and combinations has finally caught up to make those algorithms real.

Broadly, AI has 2 categories, machine learning, and then large language models. Machine learning talks about taking mathematical and computational data and converting it into predictions. So a good example — You should be able to use machine learning to predict the lottery numbers for the next Powerball, MegaMillion, etc.

That’s where you can use machine learning. It takes numbers, takes past behavior, and predicts the likelihood of a number in the future. Now, the second part is what is called as a large language model, which is all the buzz you see right now, right? — ChatGPT, Perplexity, and all these other companies out there. So what does a large language model do? So given these algorithms have been invented a while ago, now, newer versions of algorithms are coming into play. Newer technology is coming into play. A lot more compute is available. A lot more processing is available. So now a large language model is basically taking inputs of text, images, videos, and manipulating them and translating them into a better version or an output, again in a text, image, audio or video.

7 years ago when Alexa or Siri came out, I’m pretty sure, a lot of people, a lot of the audience were like — “How… How did this computer or this phone or this small little device understand my questions?” But if you observe, there were only a set number of questions it could answer. If you give a wild question, by default response is “I do not know” or “I don’t have the answer to this question.” Now, with large language models there’s a lot more training going on. There’s a lot more different kinds of algorithms, neural networks, things like that, that is trying to mimic human behavior. So big picture AI is machine learning and large language models. Depending on the context, you use one or the other to derive an outcome. I hope I didn’t go too technical, Hari.

Krishna Hari (03:47)
No, no, no, that’s fine. You’re absolutely fine.I think the message which I get is… AI AI has been there before, now with LLM coming in, it has changed, made it much more intelligence-specific to any particular sector.

Vikram Gollakota (04:01)
Especially for consumers, right? For consumers, the real application of AI you see right now, the whole Gen AI buzz, is when consumers are able to very quickly adapt, use that information and technology versus you expect an end user to figure out a decision tree or a random forest algorithm, it’s going to be hard. So it went from a theoretical, mathematical, probabilistic set of algorithms to a much more user-friendly model, which is LLM.

Krishna Hari (04:29)
Makes sense. Makes sense. Coming to the CFO’s office, what are the common challenges we have seen in finance operations in companies? Let us not focus on one segment.

Vikram Gollakota (04:41)
Understood. So as a company, one of the cultures that our founder has set for us is no opinion sharing. So I will go by experience sharing. Our experience being with HighRadius from almost the beginning is that all the clients that we have done transformation work for, whether we use RPA, whether we use machine learning, or gen AI, it doesn’t matter. We have transformed these businesses.

So my experience comes from about 1,100 different clients, a little more than 1,100 clients. Across, broadly, we classify our customers into 12 segments. Like, think about the large ones being technology companies, manufacturing, CPG, food, beverage, retail, insurance, health care. So a lot of these industries. The only industries I don’t have experience is government and healthcare service providers like hospitals. But to make sure I’m answering your question, it is an experience sharing, not an opinion sharing. 

Broadly, when you listen to all the shareholders and asking questions or investors asking questions, a lot of public company CEOs are under tremendous pressure to do something with AI. They have to say, “I am doing ‘blah’ AI, I am doing ‘blah, blah’ AI,” because the market is reacting based on what they say. Most CFOs, they are very, very practical people. They want to see “What is my return on these investments?” And unfortunately, with all the hype around AI, I mean, think about it. NVIDIA was a billion dollar company, a trillion dollar company in 2021. Now it is 5 times that.

I mean, I’m surprised. Let me take a step back. Goldman Sachs published a report, I think in 2023. If you find it — you’ll be able to Google it and find it — where it was very clearly called out that a third of finance and accounting, FTEs can be automated or AI will impact them one way or the other. This is in 2023. So, there’s been a lot of buzz.

So everybody is hyping around the buzzwords of AI and most CFOs are like, “Great, I would love to leverage it, but show me how.” There’s hardly very few companies that have shown me the path. Very recently, I don’t know if you’re following the new buzz around OpenAI. I actually think this is kind of getting into more of a Ponzi scheme model, pretty close to that, maybe I’m wrong.

But think about it right? There is one company that designs these chips, doesn’t even manufacture them — NVIDIA, and then who are their biggest customers? Almost all the infrastructure, all the cloud companies, right? They buy from them and then there are tens and thousands of AI startups that are using these… Most of them are not even profitable, right? They’re burning cash like no tomorrow.

So all of them are funding OpenAI, NVIDIA. NVIDIA decides to invest in OpenAI so that OpenAI can buy more of NIVIDIA’s chips. So this is becoming a most close to, maybe it’s not, but time will tell. Guess where will all these companies make their money? From enterprises like you. They will come and they will say, “I have an NVIDIA chip”,

“Your Google Cloud storage space is going up”, or “Your AWS cost is going up”, or “Your Microsoft Office cost is going up because they’ve embedded AI in it”, they’re going to squeeze money back from you as an enterprise. So CFOs are still struggling to identify what the true return on investment of AI is. In fact, there was a report that came out of an MIT study that said 95% of AI projects do not deliver any value. 

Krishna Hari (08:02)
They failed. 

Vikram Gollakota (08:03)
I mean, if you Google it, you’ll be able to find that as well.

So the common challenge is, if I have to go all the way from CFO down below, CFO is under pressure — to show something in AI. It’s shareholders, investors, everybody wants the company to get onto the AI bandwagon. They’re looking for, “What can I do?” You go talk to the CIO. CIO says, “OK, let me get some company to do it.” They all come. They all bill hours They find something. Project isn’t going to be successful. So that is a common problem at a CXO level.

But operationally, finance teams are still struggling. I have been to so many client conversations where things are still manual. They still use basic paper, Excel models to do their business, which is not efficient at all. So yeah, there’s a big gap between what the CXOs are expected to deliver with AI versus what the operators are actually getting.

Krishna Hari (08:49)
Absolutely. Absolutely. I think I’m with you there. 

The other thing which I was thinking is — With these changes coming in so quickly, what does the future organizational structure look like? What I mean is, obviously there is a buzz, there is some level of change management happening in organizations, right? They are looking at automation, like for example, RPA… and there are some smaller automation tools as well, which are bringing a lot of results. So my question is how does the future of CFO office look like? 

Vikram Gollakota (09:24)

So I’m going to give you a consulting answer. It depends. Okay. I hate those phrases. “It depends”. Let me tell you how we as a company are driving the adoption of AI and then the agentification of human tasks. 

So I’ll take one example for finance and accounting professionals, you probably understand collections management. Collections management is where you call on your customers to get paid for the past due items. So if you think about a collections it’s usually broken down into 2 segments. Large customers and SMB customers. Large customers — you don’t pick up the phone and call anybody because it doesn’t work that way. So you end up usually going to an AP portal, some kind of website, right? Where you log in, you upload invoices, you check invoice statuses, you see which ones are disputed, not being paid and you work off website. So website is your interface to the other company for your large customers. For your SMBs — you have to call them, you have to remind them, you have to address their questions. It’s mainly email or phone, right? 

So if you break it down, you have large customer collections and then you have SMB, where multiple personal conversations and emails exchanges happen. So how are we transforming this end-to-end collections process and agentifying it? Each of those tasks needs certain technology to drive automation. And when I say automation, there’s assisted and there is autonomous. 

Autonomous is where 90% of the work can be done by the machine with the human in the loop. Assisted is — it’ll do 50% or less than 80%, but you need a human to do the remaining part. And as a combination of, let’s say I need to upload to an AP portal. Why do I need Gen AI? I can use RPA, right? To log in, upload invoices, submit an invoice copy. 

Now, I do want to predict the likelihood of a customer not paying an invoice on time. That has to be some kind of proprietary algorithm based on your experience. You can’t expect a large language model to do this for you. It’s an AI algorithm for sure. It’s a machine learning algorithm that does some predicting based on past experience. So you need the machine learning, you need some proprietary algorithms. There is also a segment where you need large language models, which is — when you’re sending these emails, the conversations, what is he (the customer) saying as a response to your email context. So if you think about it, that is what agentification is all about. It’s taking combination of the right AI technology, RPA, machine learning, proprietary algorithms or LLM for the right task, but an end-to-end process.

With that in mind, what will the future of a CFO’s organization look like? They will have most likely 70% of their staff as AI agents and 30% as humans to make sure that 70% is doing their job effectively and doing any human-in-the-loop. 

Krishna Hari (12:14)
Sure. See all this has so many data components involved. So that’s my next question. What is a CFO’s organizational challenge in having quality data and what are your thoughts on that?

Vikram Gollakota (12:28)
So let’s make sure everybody gets this right — That strategy will be a myth. You will never ever be able to fix your data forever. It will always be corrupted one way or the other. 

It is very similar to cyber security. You have this amazing anti-virus program, but guess what? Somebody’s smarter. So you have to keep upping your game. So it’s very similar to the data front. So if you’re expecting, for a magical wand to fix your data problems, and then figure out how you want to bring automation to your organization, then you’ll never going to go beyond fixing your data problem. That is my experience, again. 

You will have to do what is called as a continuous improvement. It is a everyday job. It is more than a job. It is a discipline. How do you make sure you do not create a hundred payment terms? “Well, that customer will only do business if you do this payment terms.” Do you have the decision making to say “No”, or will you succumb to the pressure of sales and make it happen?

So those are all discussions, but my take is data will always be a problem. The question is “How can you work through the data challenges you have yet drive automation for your business?”

Krishna Hari (13:39)
Following to that, I think if this, what I’m hearing from you is it is going to be a continuous improvement, what foundation should you have?

Vikram Gollakota (13:47)
So there’s multiple ways to skin this problem, foundation-wise, right? I’m guessing this is foundation for how do you identify opportunities within your organization? Is that where you’re going?

Krishna Hari (13:55)
Well, that is one. The other one is — There is a continuous set of changes or is there data quality softwares which are coming up or is there an automation process where you can identify these sort of steps?

Vikram Gollakota (14:08)
Let’s go with your hypothesis of data quality software. That’s when there is a magical data quality software that will fix your data once and for all. You as an organization are not going to not grow. Every organization exists to grow. Absolutely. A part of your growth strategy is to do an M&A. Will you not acquire the company because the data is not right? So, the moment you acquire, you’re inheriting data points again.

My take is this is a continuous improvement. It has to be a discipline. It will continue to be on, just like how finance is a forever function, data quality management will be a forever function. So that’s my take on building that foundation for data management.

Krishna Hari (14:46)
Makes sense. The next one is human+AI collaboration. Where should they start to begin a huge transformation journey? Should they do a process assessment? Or should they let IT come back and propose what they have in technology, companies that commit to automation?

Vikram Gollakota (15:03)
You know, my boss always tells me, “Do not delegate the most important thing for your business”. And if I’m a leader, if I am the head of AR or AP, the one thing I will not delegate is what do I do to improve the efficiency of my organization? 

So here’s an example. I think we can share with you some sample time studies that we can do. I mean, think about it, we offer softwares in AR, AP, R2R, treasury, payments, all these areas. And within each of them, we have different products. For each product, we have built a time study or an analysis because as a business, we absolutely build a business case for every client. So we can share some templates and maybe you can share with your audience when they need it.

But let’s take going back to the collections example. Here’s a simple time study. Probably we’ll have about 9 to10 tasks in collections end to end. Almost there, these 9 will almost be a template. Maybe a few exceptions, few things you change here and there, but 80% there, the tasks. You as an owner of that department, of collections department, as a manager, you figure out, talking to your analysts, your 10 to 20 analysts, 30 analysts, whatever, right? What percentage of the time is spent on each of these tasks? It doesn’t take more than a few hours to get this spreadsheet filled up. Then you put your time spent on these tasks because both are relevant. 

Once you get all these things completed, then you go see what the best-in-class software can drive automation for you. Go talk to vendors. There’s a ton of reports out there. There’s a Gartner report, there’s IDC report, there’s so many reports out there. Ask the vendors, saying, “Here is how I am running my business today. How much can you automate?” Usually when they come back, they give you some numbers. You take the min., you take the max., you take the average. Your average will tell you how inefficient your organization is. \

Now, remember: Software salespeople are still salespeople. They will spin a story. So make sure you ask them to not just quantify, but underwrite these things. That’s when the real game will begin. But anyway, long story short is as department leaders, I would do the time study myself, I would invest the time to understand what I can do. This is hard efficiency. There’s other benefits also you can think about, but that’s my first thing I would do. Not wait for… these big 4 companies to do the same thing and charge you millions of dollars. So it’s up to you. If you have the luxury of spending and affording the time, please, if not, do it yourself.

Krishna Hari (17:32)
Makes sense. Makes sense. Absolutely. I think this comes to the next question. Where do you think the world will be in 3 years from now, given the AI buzz or whatever is happening right now?

Vikram Gollakota (17:43)
Yeah. Good question. And unfortunately, I don’t know the answer to this question. If I did I would definitely be betting on those companies and those stocks. 

But I will give you what our vision is. What is HighRadius as a company driving towards, right? And I’m very, very focused on finance and accounting space. So as of now, across all the products we have to offer, I think we have over 20 products in total, 60 different product suites, 186 agents…

There are a few like cash application, a little bit of cash forecasting – is what we call as autonomous. We think in 3 years, finance and accounting will be 90% autonomous. How do we get there? And I’m only focusing on finance and accounting. We can see the path and we can see the mission. We can see the target setting that our Founder/CEO is setting for us. How do we drive to get 90% automation or autonomous, which is 90% automation, at an end-to-end product level for our clients? So how are we doing on the journey? We are heading full force, heavy investments to get finance and accounting to that level. There will be more disruptive conversations.

Now, very recently, there was a new buzzword after Gen AI, which is autonomous AI or some other… It’s AGI. But there’s multiple versions before AGI, where AGI is like the holy grail, right? So that will continue. That will continue. You focus on, as an end user, as a business owner or a CFO of a department — What are the tasks you can automate? How much can I contribute back to grow? That’s my take. 90% is our bet by 2027. Hopefully sooner.

Krishna Hari (19:37)
That’s pretty good. I have one more question. I know you talked about the definition of autonomous AI is 90% of it is automated or without intervention, right? How many products HighRadius has today in those areas?

Vikram Gollakota (19:50)
All of them. All of our products use a combination of machine learning, RPA, AI, and GenAI capabilities to deliver end-to-end process streams. Now keep in mind, we only do the CFO space — AR, AP, R2R… There may be other areas, like customer service, which is heavier on LLM models, because it’s mainly user interaction or ticketing systems…

But for us this is what we are seeing.

Krishna Hari (20:18)
Great. Well, coming to closing remarks, any takeaways for the audience?

Vikram Gollakota (20:23)
So my take as always — I’ll keep it very actionable from my point of view. 

Number 1 — Don’t get carried away by the buzz. Always double click, ask the questions, ask the WHY questions, and ask the HOW questions. When someone says, “I can do…” Vikram talks to you tomorrow and says, “I can do 80% automation.” You ask him me “HOW?”. “Show me proof of concept or proof of value”. Pick one or few agents and then say ask him to do a proof of value, number one. 

Number 2 — Ask him to underwrite the commitment. The company, the software company that will underwrite value for you in contracts, when one way or other is truly going to transform your business. Everything else is hype. 

That’s my first takeaway.

My second takeaway is don’t wait for somebody else to tell you the organization is inefficient. Take the time studies, do it yourself, identify opportunity areas, and go out in the market. You can use search for the best AR software or the best AP software. You’ll find the link. Talk to them. Get them to do these projects for you. Those are the 2 actionable takeaways I can talk to you about.

Krishna Hari (21:24)
Great. Great. Thank you very much Vikram. We unraveled the basics – where it is going and where HighRadius products are also valuable tips for the CFO’s office, how we can take this forward.

Krishna Hari (21:38)
Thank you Vikram. This has been an incredible, insightful conversation. You have helped us reimagine what is possible when finance operations become intelligent and autonomous. 

Also, to our listeners, if you are a CFO or a transformation leader, wondering where to start… Begin with clarity. Identify the repetitive data-rich task. And explore agentic AI and release human bandwidth for strategic thinking. 

You can also engage us who are an implementation partner of HighRadius. And also we will bring a lot of playbooks of HighRadius, success stories with similar industry companies who have done HighRadius deployments using agentic AI.

If you have any questions, feel free to comment and we’ll love to help you with agentic AI journey. Feel free to subscribe to us. We are listed in Apple, iHeart, as well as Spotify – in all podcast platforms. Thank you.