Adam Larson: 00:06

Welcome to another exciting episode of Count Me In. I'm your host, Adam Larson. And today, we're joined by Pujha Sund, the engineering finance leader at Microsoft. Get ready as we unravel the complexities of artificial intelligence and uncover its game changing potential. Puja shares the key differences between AI and traditional programming, highlighting AI's adaptability and practical applications in finance, from mismanagement to fraud detection.

Adam Larson: 00:37

This episode is packed with invaluable insights and actionable advice for anyone looking to integrate AI into their workflows. So if you're ready to unlock the full potential of AI with expert advice, stay tuned. I'm so excited to have you on the Count Me In podcast. It's we're gonna be talking discussing AI and its effect on the finance and accounting team. But I figure when we start talking about AI, we need to demystify it a little bit.

Adam Larson: 01:06

What exactly is AI? Maybe how does it differ from traditional machine learning into a program that's been around for a while?

Pooja Sund: 01:14

First of all, thank you so much, Adam, for inviting me to your podcast. I'm very excited and looking forward to share my learnings and experiences. When it comes to AI and how it differs from traditional programming, there are a couple of aspects that I would like to spend some time on. 1st is around, AI, the first 1 traditional programming in terms of its approach itself. So in traditional programming, Adam, developers, they explicitly code rules and instructions for the machine to follow.

Pooja Sund: 01:39

They find specific conditions and logic. They create a set of predefined rules that guide the program behavior. While in case of AI, you are actually letting machine learn from the data and make, its own decisions. So instead of hard coding rules, AI systems are using complex algorithms such as machine learning, reinforcement learning to learn from some of the patterns and adapt autonomously. So first, difference is on the approach.

Pooja Sund: 02:06

2nd, I feel like there is a big difference in terms of data handling. Traditional programming on 1 hand deals with just structured data, and it follows some parts that are pretty well defined. And traditional programming struggle with unstructured data, for example, natural languages or images. While if, you are following the news nowadays, AI can help you process large datasets. Doesn't matter if it is structured or unstructured, if if it includes images or even videos.

Pooja Sund: 02:35

So second big difference according to me between traditional programming and AI is in data handling. And 3rd, which is a key, is around transparency and control. When it comes to traditional programming, Adam, developers have full control and transparency in traditional programming. They can trace the logic and flow of the program, making it easier to understand and maintain. When it comes to AI, sometimes there could be some mystery behind it, and their complex programmings and algorithms may not always reveal the reasoning behind the reasons.

Pooja Sund: 03:07

So that lack of transparency can become a concern for those who need clarity and accountability. And people like us who are in the finance discipline or who have worked in the finance discipline before, they would like transparency and control. So that's where there is a blurry line between AI and traditional programming.

Adam Larson: 03:25

There really is that blurry line, and and people always wanna make sure that they have that ability. You you can't have artificial intelligence still without human intelligence. You need you kinda need both to make sure that they function properly without human eyes on certain aspects. Obviously, the the AI can, analyze things more than the human brain can do, but you still need that human side to be able to tell the right story.

Pooja Sund: 03:48

Absolutely. Yes. And story is what differentiates us from the machines.

Adam Larson: 03:53

Mhmm. So can do you think AI can truly learn and adapt? I mean, obviously, as it has all those complex algorithms and it seems like it's learning, but is it actually learning?

Pooja Sund: 04:05

It is, Adam. And there are 3 points again that I would like to highlight to make sure that, we can understand how AI can learn and adapt. Of course, there is that storytelling telling us that empathy aspect that would take AI to catch up on, and it would take more time. But when it comes to traditional, learning and some of the innovative forms of learning, AI can adapt, learn, and improve as it encounters changes in data and its environment. So unlike traditional programming, which follows set rules and algorithms, adaptive AI systems can modify their behavior based on the experiences.

Pooja Sund: 04:41

They can adjust their own code without human input, providing a level of adaptability. And second is, you can actually allow adaptive AI to process and analyze new information. You can train it so it can it can acquire knowledge from multiple sources. It can identify some anomalies and make predictions. It can also adjust its own algorithm and decision making processes.

Pooja Sund: 05:03

This flexibility makes them practical and relevant even in dynamic and unpredictable situations. And 3rd is around how AI can help us self improve and, by using the weak points or inefficient areas when AI would identify those, you can refine your algorithm in response. So by learning from those experiences, by learning from when the financial projections were inaccurate, it can actually bake those, into the assumptions and can come up with the right assumptions. So you can use adaptive AI to come up with sophisticated algorithms that can solve your problems. So I do believe AI goes beyond complex algorithms.

Pooja Sund: 05:43

It can learn based on your given inputs, conditions, and environment. It can adapt, and it can evolve based on the real world experiences, making it a powerful tool for for financial, folks.

Adam Larson: 05:55

So speaking of financial financial and auto speaking of financial and accounting folks, you know, what are some key capabilities that we can focus on with AI that can be leveraged by those teams? Because, you know, obviously, there's the initial automation that we've already discussed, but I know that there's so many more capabilities out there.

Pooja Sund: 06:14

Yes. You can ask me questions around how, different capabilities of AI can be, implemented, not just in finance, but in technology, in in retail. It's just like there are multiple use cases. And for those of us who are on LinkedIn, we have been following influencers. I mean, there are tons of use cases.

Pooja Sund: 06:33

For professionals like me, I have actually already applied, implemented, and, leveraged the use of AI within finance. So the first 1 is that I can talk about is around risk management and fraud detection, Adam. AI algorithms have helped us in detecting anomalies, including fraudulent activities or unusual patterns. By analyzing historical data with respect to sales deals, made by customers, with Microsoft, AI models are helping us identify potential risk and preventing financial fraud, enhancing security and trust in our company. So an example would be, like, before a customer would come in and, engage in business with us at Microsoft, we can actually use AI to identify if it makes sense for us to do business with that customer or vendor looking at their risk profile, looking at where which country are they coming from, things like that.

Pooja Sund: 07:27

So it definitely helps in risk management and for detection. 2nd, transparency and compliance, the 3rd approach that I talked about when I was explaining the difference between AI and traditional programming. AI helps companies in maintaining compliance with regulations by automating some processes. So some companies have a long way of looking at transparency in reporting, anti money laundering. So AI can help ensure that there is a transparency in anti money laundering checks and adherence to legal requirements.

Pooja Sund: 07:58

3rd is around cost reduction, which is the favorite for people who are in finance. AI actually helps streamline routine tasks such as document processing, and there can be structured, unstructured loan servicing, customer onboarding. So by automating these processes, we can actually use AI to reduce operational cost and improve efficiency. 4th, that you might have heard is around personalization. So no matter who we are talking to, AI can help creating personalized experiences for those customers.

Pooja Sund: 08:29

So for for us in the technology company, we use AI to come up personalized services and recommendations for our customers. For a retail, for a finance, team, it could be customer looking at customer data, analyzing it, looking at their preferences and behavior to offer tailored advice, whether it's, coming up with the new banking products or insurance options, or it it might involve coming up with the right products so you can entice the suppliers. And AI can help suggest personalized portfolios as well based on risk tolerance and financial goals. And the last 1 that I can think about is around sentiment analysis and investment insights. So for those of us, Adam, who are in the financial industry, we also play with stocks.

Pooja Sund: 09:11

Right? We look at what are what is the financial sentiment of the stock market at that time, what is, Fed guidance on the interest rate. AI can help us look at the broader market, analyze those sentiments coming in from the external market, also look at what customers are tweeting about, the particular product on Facebook or Insta, and then can actually help us look at is the overall customer sentiment positive or negative for for this. And then based on that, we can actually, make a change in the pricing or the marketing strategy. So these are couple of advances advancements and capabilities that we can all leverage in finance and in other comp, industries as well.

Adam Larson: 09:52

It seems like the possibilities are endless, and it gets it gets really exciting when you think about the application of so many of the the things that you mentioned. And just and just thinking about how far our businesses can go. But I start to think, you know, you know, AI can do all these things, can analyze all these different things. Yeah. Do you think it'll help us better tell the future?

Adam Larson: 10:13

You know, obviously, science fiction has given us very fantastical, ways that AI can go bad and all those different things. But how can we better make decisions for our organizations with AI?

Pooja Sund: 10:26

Definitely, we can make better decisions thinking about security and privacy and keeping that in mind while thinking about the future. So think about all of those advancements that we are making. We can keep making personalized recommendations for customers thinking about not just what is the customer looking at it right now, but what would be the customer lifetime value? How should we think about that? What could be the needs that that customer is gonna ask in the future?

Pooja Sund: 10:53

So if you think about a a company's long term vision in North Star, they are always interested to know what could be the other potential business model that they need to come up with to keep earning, to keep retaining their, current customers, and to keep preventing customers to go to another company. So AI can actually help you in understanding not just operational efficiency metrics, but also able to analyze sentiment as I mentioned in the previous question and can help you understand how you can retain that customer for a lifelong. And that can actually add value to the business because you are not just thinking from the short term. You are thinking from long term.

Adam Larson: 11:31

We really are. We can't just look at the small things now. We have to look at the big picture and as as part of the as part of the strategy team and being business partners as finance professionals, we have to be a part of that conversation. And maybe there's ways we can use AI. Our our the traditional ways of looking at profit margins in our eye are important, but are there new matrix that we can use to kinda see our financial health in a better way?

Pooja Sund: 11:57

So by no means, I'm saying do not look at the traditional mean metrics that you just quoted along the including profit margins in Android. What I mean, by the new metrics that we can look at is it's think of it as a supplement. Think of AI as your personalized assistant who can help you look at other metrics in addition to traditional metrics. So couple of them that I would like to mention it here and what people would be able to think about when they think about how AI can help in enhancing their financial financial health would be, around, Adam, the first name that comes to mind is around customer lifetime value. So CLV is what we call it.

Pooja Sund: 12:38

CLV means, representing the total value a customer would bring to a business over their entire relationship. AI models can help predict customer lifetime value by analyzing customer behavior, their preferences, and historical data. And this helps business allocate resources effectively and focus on only high value customers. So 80 20 rule, 80% of your value is gonna be come from is gonna be coming from 20% of customers. So customer lifetime value is 1 of the metric that AI can help you look at.

Pooja Sund: 13:06

2nd would be the churn prediction. Churn meaning customers leaving a product or service. AI algorithm can help forecast churn by analyzing patterns and identifying AI risk customers, so customers that are at risk. This allows businesses to take proactive measures to maintain valuable clients. 3rd, I would say, Adam, is around we call it as, risk adjusted return on capital.

Pooja Sund: 13:31

I know it's a big 1, so you can remember it by using RAR0C, risk adjusted return on capital. What it means is you are measuring the risk adjusted profitability of an investment. So no matter if an investment $100 investment is gonna be $1, 000, you're actually looking at that $1, 000 in, relation to inflation, in relation to the risk. So AI models enhance these calculations by incorporating complex risk factors and adjusting returns based on the risk exposure. And the 4th 1 that I can think of would be operational efficiency metric that can help us evaluate the efficiency of internal processes that are being executed within each and every financial company.

Pooja Sund: 14:15

AI impact would be around how it can help process automation reducing cost, improve accuracy, and speed up operations. So metrics like resource utilization, error rate can benefit from that. So I truly believe that AI can help us look ahead. Do not lose focus on traditional financial metrics, but look ahead in, with respect to customer lifetime value, churn prediction, risk adjusted return, operational efficiency, and sometimes even sentiment analysis score, which could be positive, negative, or neutral.

Adam Larson: 14:48

Mhmm. I it's amazing as you go through all the different, levels and all the different elements and how we can apply it. It seems it continues to grow. It's like every time you mentioned something, Oh, that's new. That's something.

Adam Larson: 15:00

That's something. And it gets it like, I just keep saying exciting is the word because the opportunities are are continuous. You know? And and as we look at action, putting putting this this AI into action for your the workspace. You know, you've already mentioned so many different examples that we're talking about.

Adam Larson: 15:19

But, you know, things like bookkeeping and auditing, I think those are some of the biggest places. And how can AI kind of really attack those and help help that that part of the organization really become more efficient?

Pooja Sund: 15:32

It has so many good, points that you just mentioned, Adam. You said opportunities are so many. So I actually like to think, like, opportunities exist here in our mind. They're not outside. With AI coming in, there are bunch of opportunities that are untapped, and we have not even thought about exploring it.

Pooja Sund: 15:51

So first of all, all of us needs to keep in mind that generative AI can help us lead productivity growth of not just 0.1%, but even more than 0.6% through 20 40 depending on technology adoption rate. And specifically with respect to bookkeeping and others, just just keep that in mind when you're thinking about making sure that every and every single digit gets recorded in the books. At times, accruals are going to happen. You need to understand if you are doing overaccrual or underaccrual. And if you're not sure of how you can take care of it, you can use AI to come up with projections.

Pooja Sund: 16:29

So the traditional bookkeeping was just keeping track of what's coming in, what's going out. Using AI, you can actually do the better projections, and you can use safe approach, traditional approach, and you can use an aggressive approach. And then you bake in the external factors, sentiments, market analysis, industry trends, and come up with better projections so you can actually estimate how much do you have to keep it in the next year, when would you be able to spend it, and if something would drastically change from the expected assumption, what is going to be the impact on the bottom line? So these things can also be done using AI.

Adam Larson: 17:06

Mhmm. What about security and risk? Because when it comes to using AI, you're putting your information into a system. And I a lot of people are using AI within softwares that they already apply for because so many companies are running and using it. You know, how worried should we should we be about our data?

Pooja Sund: 17:26

We should be very worried about our data, Adam. And you are actually asking a question which is a multimillion dollar question, security and privacy. Mhmm. With the cyberattacks that are happening at an alarming rate, this becomes the top priority for every single company. I'm not sure if you have heard, but there are big corporates that are not investing time in coming up with new features or products.

Pooja Sund: 17:51

They're actually investing time in going through the end to end security review. So when you think about AI, yes, incorporating AI into your own products can become easy. And later on, you can think of, looking at the security aspect, but you would fail. So you have to really understand that you're not gonna think about security privacy at the end when the the AI products have already been built or AI models have already been implemented. You're gonna think about AI security and privacy when you are dealing with data to create that model because data breaches can happen when your data is being handled by your data analyst and data scientist or even by financial professionals.

Pooja Sund: 18:37

So thinking about the persons who is who are using the data, do they have the right access? What is gonna happen when they're gonna move to the next role? How many, what kind of access are we gonna give it to that person? Are they going to use multifactor authentication? These are the great steps that you can take to prevent data breaches is to happen, to prevent your model to be vulnerable.

Pooja Sund: 19:00

Implementing robust security protocols, Adam, including encryption and access controls can help the teams to build saw solid AI models that can avoid these vulnerabilities and these risks later on. So I would say definitely, there is 1 thing that I would like to suggest people before even thinking about learning about AI, start thinking about security concerns. Think about legal and ethical risk as well, and keep that in mind while building AI models because data is very important, and you do not want your data to be leaked.

Adam Larson: 19:35

What about bias? You know, that's something I've been hearing a lot when talking about AI and the large language models and and different responses that have been coming back. How do we avoid bias when we're trying to look at all their data and look at you know, and trying to analyze in all the examples that you've been mentioning?

Pooja Sund: 19:51

Great question. So I would say definitely that's another 1 that comes to my mind. Accuracy issues and bias in models. AI models can suffer a lot from inaccuracies and bias leading to suboptimal decisions. So human supervision in that case, Adam, is absolutely crucial to address these issues.

Pooja Sund: 20:10

Do not think that when you have built the model, the the confidence score is 90 or over 90%. You do not need a person on your team to supervise it. You would always need it because human supervision can help us avoid those issues. So regularly evaluating model performance, identifying biases. Are the people who are building it, are they only representing 1 part of the population?

Pooja Sund: 20:34

If they are, then definitely your model is gonna be biased. Are they looking at just 1 stereotype? Are they looking at 1 part of data? So make sure your data is whole, and you can identify those biases early on. You can fine tune their your algorithm later on, but you cannot identify bias, later on.

Pooja Sund: 20:52

If you'll do that, then all of the work you have to restart. So implementing fairness aware techniques at the get go can help to reduce bias and ensure equitable outcome. There are cup couple of companies that I'm aware of, Adam, that have come up with their responsible AI principles and, ensuring that there is no bias in models is 1 of them. At Microsoft, we have come up with these principles. We are calling it as fairness, accountability, reliability, performance, and accuracy.

Pooja Sund: 21:23

So there are 6 principles. Google also has couple of them, and, Amazon have their own. So think about no matter where you are, you can either come up with your own responsible AI principles, or you can think of leveraging a governance framework that's already out there in the market that is being followed by other companies. But do not, think of ignoring looking at bias and handling bias and then incorporating, strategies to mitigate the bias in your AI systems.

Adam Larson: 21:54

So from everything that you've been saying, it sounds like if somebody's just getting into this and and looking at using AI in their systems, and and if you're not using AI, you should get started. And but the best way to to do it is to lay a good foundation for your organization and yourself before you really dive into it deep. Because if you're not, then you're gonna be backtracking as you're trying to, cover all the different things that you were just mentioning.

Pooja Sund: 22:20

That's absolutely what I'm saying. Yes. Thank you for reemphasizing the notion of thinking about it right at the get go. Do not get, lurid or do not get, enticed with the outcome efficiency of AI and then just go and let your team just go and follow the route to get to the destination. Take your time.

Pooja Sund: 22:42

Think about how you're gonna build it, who is gonna build it, what kind of processes do they have? What kind of profile do they have? How do they think about security? Have a governance team. Have them do thorough review throughout the process, not just at the beginning and in the middle and at the end, but, consistently, cadence of reviews should happen.

Adam Larson: 23:04

I really like that. It it you have to you have to see the big picture, and that's that's something that we talk all about a lot is looking at the whole strategy, not just your 1 siloed of your team, but seeing the big picture and how it affects everybody is super and super important, especially because AI affects your whole your whole organization, not just maybe your 1 team that you're over, but it affects everybody. And if your IT team doesn't have a policy, say, hey, guys. What's our policy on this? Otherwise, you know, things could go awry very quickly.

Pooja Sund: 23:34

Yes. Absolutely. Thank you.

Adam Larson: 23:38

So as we, you know, as we wrap up this conversation, we've covered so many other things. You've given us so many wonderful examples and shared so many things. What are some key takeaways that you want our audience to remember as they as they leave this conversation?

Pooja Sund: 23:51

I'll say audience needs to keep that in mind that AI definitely drives automation, and the AI driven automation offers immense benefits. Everyone should proactively address all of the challenges that we had just talked about to ensure successful implementation. So go and reap the benefits. Do not forget, proactively take measures to keep security, legal, and ethical concerns, plus learning and creating training programs for your teams in mind. 1 of the aspect, Erin, that we didn't, touch base on was maturity and change management.

Pooja Sund: 24:24

So all of the technologies are evolving rapidly, making it challenging to keep up. So companies needs to make, make sure that they'll actually find out a way to manage change, address employee concerns, and foster a culture of continuous learning. So 3 big takeaways from this session, develop a culture of continuous learning. 2nd, proactively think about security and privacy concerns. 3rd, always leverage responsible AI principles and make sure your data is, free from bias.

Adam Larson: 24:54

Well, Puja, thank you so much for sharing all your knowledge and expertise, and thank you. And I just really appreciate, you coming on the podcast today.

Pooja Sund: 25:03

You're very welcome. Thank you so much, Adam, for having me. And for those of you who are listening or, looking at the podcast, please reach out to me or Adam. We both are on LinkedIn, and we are happy to share our knowledge and experiences with you all. Thank you.

Announcer: 25:16

This has been Count Me In, IMA's podcast providing you with the latest perspectives of thought leaders from the accounting and finance profession. If you like what you heard and you'd like to be counted in for more relevant accounting and finance education, visit IMA's website at www.ima net.org.