Forecasting the Future of AI in Financial Planning and Analysis

The role of artificial intelligence in finance is just starting to be understood, and more importantly, how AI mixes with other tools like predictive analytics and robotic process automation in building a solid financial forecast.
 
In this sponsored episode of the FEI Weekly podcast, we speak with Justin Croft, Vice President of Cross Solutions Architecture at QueBIT, about the promise and practicality of AI and financial planning and analysis.


 
A transcript of the discussion is below.
 
FEI Weekly: Justin, thanks so much for joining us today. A lot of discussions we're having with the professionals and financial executives around AI and how it's changing the landscape and landscape and the career paths of everyone involved in accounting and finance, but I wanted to start off maybe a little bit about you and your company and your background.

Justin Croft: Sure, sure. Thanks for having me today. QueBIT is an FP&A strategy services organization. We're really focused on delivering excellence in financial and operational planning and leveraging machine learning and artificial intelligence in ways that really improve productivity, efficiency and accuracy of the planning process. We've been doing this for almost 25 years. I've got about 100 consultants across the country that work with thousands of clients each year delivering excellence in planning and FP&A.

FEI Weekly: Great. So I want to really dive into it and always like to go big to small. So maybe you can start by giving us a brief overview of how artificial intelligence is currently being used in finance, and one thing that I've really struck by is how quickly it's changed over the past 18 months.

Justin: Yeah, it really has. I would say broadly, artificial intelligence is not being used in the office of finance. We have a lot of conversations about how to leverage AI and machine learning, and people are interested, but they're not taking that next step. So that's for better or worse. The most common answer is we're evaluating it, we're looking at it, we're not doing it yet. And that's a real opportunity for a lot of organizations to adopt AI, to adopt machine learning and to automate a lot of processes and improve the way that they operate. So that's exciting. There's still a lot of growth coming.

Certainly it's changed in the last 18 months with the advent of tools like ChatGPT. That's also spurred a lot of interest, and I think the way that modern AI is used in finance is really still unfolding. I know we've experimented with, for example, using some large language models to analyze financial statements and to produce analysis and results, and that's interesting, but it's not accurate enough for production yet, and so it's really an evolving space, which is exciting. Overall, the changes in the last 18 months have just spurred on interest in the overall AI field.

FEI Weekly:And that's interesting because when you read anything or you talk to anyone, all they talk about is AI, so the fact that they're not actually implementing it or practicing with it is notable. What do you think? Is it just because they're not ensconced in the services they use? Are they going into ChatGPT and using it or why is it in practice right now?

Justin:Yeah, so the easiest way to leverage AI is to pull up ChatGPT and use it for something like public filing analysis. If you're trying to analyze some 10Ks or 10Qs or compare the financial performance of a set of competitors, ChatGPT and other tools like that do a phenomenal job. That's more of a one-off analysis type activity that's absolutely being done. What's not being done is the integration of AI into systems and processes programmatically to improve outcomes, and that's where the office of finance really struggles is defining what are those outcomes that we're looking to improve and achieve. I think that's what's really holding the office of finance back, and what we really focus on is this idea of improving planning and improving your forecasting capabilities through the use of AI. That's our answer to that question is this is maybe the lowest hanging fruit there is in finance is to get better at forecasting and planning using AI.

FEI Weekly: So yeah, that brings up another question, and this something that I always ask someone when they talk about AI, but specifically, in what ways does AI change the game in forecasting specifically and what can be done with AI that hasn't been possible before?

Justin: I think the best answer to that is this idea of automation. I'd love to say it's all about accuracy, but it's really all about automation. Traditionally, the office of finance, the FP&A team is going to spend hours and hours, days and days, weeks and weeks putting numbers in boxes and just filling in a forecast so that their metrics can calculate, and the remaining 20% of their effort is put into fine-tuning and adjusting those numbers and analyzing those numbers. So the bulk of the time is literally spent just creating the forecast.
When I talk about automation, what I mean is letting algorithms do the work, letting a tool plus algorithms fill in the numbers and the boxes so that you've got an order of magnitude improvement in terms of speed and efficiency in creating the forecast. So that's the first thing. That's the first benefit that we really see within our clients is that idea of scale and automation to say, "Now we can spend our time adjusting and analyzing rather than creating." So that's a real game changer.

And then the second piece is this idea of improving accuracy and being able to use a variety of different algorithms based on the subject matter or the data that's available to forecast out the future. The ease with which that can be done now is really astonishing. 10 years ago it was an accomplishment just to build a model and you felt great about that. Now you're building dozens or hundreds of models in one pass trying to forecast one single line up, so that's a real change in terms of improvement and accuracy. So it's about those two things, the scale of automation and improvement of accuracy.

FEI Weekly: So yeah, and I want to delve a little bit deeper into that because there's been a number of other new and exciting technologies that have come up with the past, like RPA and even predictive analytics. How is AI different from those two things that really a lot of the dialogue has been around over the past couple of years? How is AI different from what's been said before and what can it do now that hasn't been possible before?

Justin: So I think of AI and machine learning as the field and the discipline of teaching computers to learn and make decisions, and that's a really powerful tool. Automation and an AI really sort of blend together in some ways, but automation can happen without AI, and AI can happen without automation. So what's really powerful is when you bring those two things together so that you're not only making decisions, but you're automating the entire business process around those decisions.

By way of example, we've got clients with multiple general ledgers, multiple CRM systems, and being able to manage and master that data across those different ERP systems can be a real challenge. Just keeping the data in sync across those systems can be a challenge. RPA with some intelligence from the AI side is a really powerful tool.

FEI Weekly: Can you discuss a little bit about the role of predictive analytics in finance and how AI enhances that in terms of accuracy and effectiveness?

Justin: Yeah, yeah, so predictive analytics is really a subset of AI overall, and predictive analytics is sort of the tip of the spear in finance. When I talk about things like forecasting and time series forecasting, that's all part of predictive analytics, and it's this idea of moving beyond traditional business intelligence where you're really looking in the rear view mirror and trying to understand what's happened last month, last week, yesterday in your business, and switching that to say, "What's going to happen tomorrow? What's going to happen next month in my business based on this historical data?" So the leverage of historical data to make a decision about the future that's going to be predictive analytics.

FEI Weekly: You mentioned the relation of models and building models. What would you would say the comfort level is now with the people you're talking to out in the field about and the financial executives delving into this concept of building a model or even partnering with others to build a model?

Justin: Finance has been building models for years. They just do it in spreadsheets, and so it's a pretty easy transition to say, "Now we're going to have a machine learning model that is separate but related to our Excel-based model of our business." Where a lot of people start is, "I've got my Excel model for my business. Let me use some basic statistical forecasting algorithms in Excel to forecast out next month." And that's built into Excel. You can drag and drop that and put it together. It doesn't scale very well. There are reasons not to do it that way, but it's a great way to get started. And from there, it's a hop, skip and a jump to say, "Well, let's have a separate statistical forecasting model that's really built around machine learning that we deploy in such and such a way," so people are open to that concept. They're just on the cusp of doing it.

FEI Weekly: One question I had as a follow-up to that, I know in a lot of discussions about these models in AI, there are large language models, and excuse me if this is an ignorant question, but they are language models that are built around words and phrases, and Excel is built around numbers. How do you overcome that or not only overcome it, but how do you integrate Excel-type spreadsheets into a large language models, or is it two different things?

Justin: It has been two different things, right? Large language models do great at predicting the next word. That's literally what they were built for is to predict the next word or phrase that comes behind a prompt. And so traditionally, large language models are not really good at numbers, but that's starting to change. People have done the research and figured out how to apply some of the same concepts to numbers and time series forecasting. So there are large language models that have come out in the last few months that are really focused on numbers, and so I think that's a really exciting development that's just going to make all of this even more accessible.

FEI Weekly: In a lot of discussions when you talk about on the science side of models, the phrase is garbage in, garbage out, and so talking about the data that's feeding the models, how would you describe the ways data is being sourced within finance AI applications currently? Where are the blind spots in that?

Justin: Yeah. Yeah. So that's a real gap today, and that's an area where companies like QueBIT can help deliver a lot of value is in sourcing that data, mastering that data, transforming that data so that it's in the right format at the right level of granularity for some sort of application. What we typically see is the finance operates in a silo with its systems and its numbers, and then operations and the rest of the business are in one or more other silos.

And so what we spend a lot of time on is breaking down those silos and combining financial with operational data so that you don't just have the numbers, you've got the business justifications and the rationale and the explanations behind those numbers. So that's what I'll call the traditional data management approach, which is still really relevant.

Then you've got this idea of what we talk about as external predictors or external factors, exogenous variables. There's lots of different names for this, but this would be data that typically comes from outside of your organization that is closely related, that's correlated to your financial outcomes, and that could be things like GDP, the price of oil, consumer price indices, any number of things that are captured and measured and reported typically by governmental organizations, more by third parties, people like the Nielsens of the world who capture and curate and collate this data and make it available for purchase and analysis. These external predictors are a great source of accuracy in terms of financial forecasting, and so that's another area that we see a lot of interest in it.

People are excited to talk about this, they want to talk about this. Most people aren't using external data yet, but they want to move in that direction and leverage that external data to support the business decisions that they're making. So that's a really exciting part of what we do, and it's evolving just as quickly as the AI side.

FEI Weekly: Let me ask you something about the data. When you're implementing and we want to leverage AI and you think about the data within finance, what do you think is the lowest hanging fruit internally to leverage AI? And to your other point, what's the lowest hanging fruit in terms of external data that you can use almost out of the gate or is that even a dialogue you're having right now?

Justin: Oh, it is. It very much is. So I'm probably biased, but I think that the lowest hanging fruit on the finance is this idea of forecasting. It's an activity that every FP&A group does. They're responsible for budgeting and forecasting, rolling forecast, quarterly forecast, latest estimates, whatever you call it. There's some sort of forward-looking, numerical prediction happening, and that's really ripe for leveraging a statistical or machine learning-based model to analyze historical data and to come up with a view of the future. You don't have to go invest a ton of money into a system that does that. You can start very small and scale up. There are lots of ways to attack that problem, but I think that's going to be the lowest-hanging fruit.
And when you take that approach, what you typically get is improved accuracy, improved efficiency and speed that I talked about earlier. You get repeatability, you get explainability, you get justifications to say, "Well, why is your number this?" If you're just putting numbers in boxes, you really can't answer that question. But if you're leveraging a model for a piece of your business, you're able to say, well, that's what the historical data and the trend tend to indicate. You're able to point to the algorithm and say, "Well, this is what we're using. This is what the math comes up to."

And so that level of auditability is really important, particularly for some industries. So I think that's the lowest-hanging fruit. And like I said, you don't have to convert your entire process. You don't have to convert your entire business into a model. You can start with a few line items that are hard to predict and apply that science and that rigor to a part of your business to get started. So I think that's the first piece.

The second piece around data, I should say there is a lot of data that is freely available from places like the Federal Reserve Board, and that information is really invaluable. They have thousands of time series data points that line up really well with a monthly forecasting process, and it's pretty trivial to say, "Well, let me pull in a few hundred data points, run some analysis and determine which ones are related to my data and which ones aren't."

I will tell a story. We did a project with a client who was really convinced that certain external drivers were related to their business, and we had to prove to them that in fact, these things that you think run your business aren't correlated to your financial results at all. And so we had to sort of dispel that traditional business myth and replace it with some external drivers, some external data that was related to their business. So that was a dramatic change in terms of the way that they thought about and ran their business. It was all based on science, right? It was all based on provable mathematical relationships that we were able to show them. So it can be a really powerful tool, lots of interest in this external data piece, and just like the rest, it's pretty easy to get started.

FEI Weekly: And are people pretty open those discussions] not built in and say, "This is the data I need," they're open to having that sort of conversation about thinking of it differently?

Justin: I'd say most of the time, yeah, most of the time everybody's interested in talking about external predictors and drivers. The limiting factor is typically what's available. Like I said, a lot of this is available for free. Depending on your industry, you may be interested in things that are more subscription-based, and the risk there is, well, how do I know if it's related to my data? How do I know if there's a correlation if I have to buy it to find out? So there's sort of a chicken and egg problem right there, but a lot of that is available for free and that's where we typically recommend people get started.

FEI Weekly: So you talked a little about reaching across the aisle, so to speak, to operations and other business lines sales. What are you seeing as best practices when it comes to finance partnering with these other business lines in order to create a more robust data set to feed to the models?

Justin: So what we see a lot of, and what we try and encourage is the idea of finance really partnering with supply chain and the operation side of the business, but especially supply chain to drive some business outcomes and just having a better forecast is not a business outcome. What you want to do is say, "Well, we want to reduce inventory and take that money to the bottom line," or, "We want to improve service levels and reduce stock outs," right?

So you want to work across the aisle, as you said, to really hone in on some of those business outcomes that people care about. And that's really critical is these have to be things that really matter to the organization and are tied into the business strategy. From there, finance is really well suited to help drive the data gathering, the data curation, the modeling, the analysis, and the deployment and development of new decisions across the organization. So that's why we like to see finance partner with supply chain is because it's really well suited to the office of finance's skill sets, but it takes both working together to get there.
What we see a lot of is organizations with functioning S&OP processes, but that are lacking in terms of efficiency or accuracy or all of the above. And we help clients improve that process across both finance and operations. And just one example, one of our longest term clients, we helped them with their global demand planning process, and when we started with them, it took about 15 days for them to do a full cycle of their global demand planning process. Very large organization. People were comfortable with 15 days, but the proper application of data science, of data and data management techniques and the automation of the forecasting process that I spoke about earlier and really flipping that 80/20 on its head, we were able to eventually reduce that cycle time from 15 days down to one day.

And that's not just an improvement in speed. That's a significant change in the way that an organization can make decisions because now instead of spending a full month, right, what's effectively a full month, 15 out of 20, 22 working dates, coming up with the forecast, now they're able to run different scenarios and run different versions. And so now their demand planning process doesn't just spit out numbers. Their demand planning process is a decision-making tool because they're able to run so many different scenarios and versions and be informed about the trade-offs that they're making. That's a real powerful tool that we help clients evolve into. That, like I said, helps them make better decisions.

FEI Weekly: One thing I wanted to follow up on that is, it's interesting when you mentioned about the supply chain, and obviously since the pandemic, supply chain has been the forefront of tying that into forecasting. Has that focus on the supply chain, has the approach to supply chain and tying into finance significantly changed since the pandemic, or is that just not a proper assumption?

Justin: Supply chain is very much on people's minds. So I think that has changed since the pandemic, and there's definitely an increase in interest in improving demand planning processes, improving the overall effectiveness of an S&OP process. We've worked with a number of clients over the past year that did not have an S&OP process in place, and so we helped to implement that with them. And so I would agree with what you're saying. Yeah, it's definitely seen an increase because people recognize the importance of supply chain on the business now more than ever before, and the same way that people are willing to invest time and effort in improving a forecast in the office of finance, now people are looking at the same thing in operations. So I would agree.

FEI Weekly: So I have a final question to wrap up. I always like to save the big one for the last, and it's something that our members are constantly discussing is the skills gap, especially with all those new technologies, especially with AI coming on board, because there's a couple different ways to look at it. There's the skills in order to implement AI and models and the skills that are needed to support it and how that's going to change if these models have a greater role in the finance function. So how do you talk about the skills gap when you're speaking to some of your clients when it comes to AI and finance?

Justin: Yeah, yeah, great, great question. So I've said for the past couple years that this is not really a technology problem, it's a business process problem, and I really believe that over the last 10 years, the technology has improved to the point where it works and it works well and it's easier to use than ever. And so that skills gap, it's certainly a real thing that needs to be addressed, but it's a smaller hurdle than it's ever been. And that's really encouraging, particularly if you start small. Start with forecasting one lineup. Start in Excel and then move into a more statistical-based approach using an outside tool. Tools are low code, no code now, so they're really built for business users. They're drag and drop configure.

What I'll say is what we've seen that doesn't work particularly well is when you have a standalone data science organization group within your organization, I should say. That creates some challenges. So I really like to embed ownership of the AI and of the machine learning and forecasting within the line of business. That seems to work the best because there's that ownership and it really forces let's say the financial analysts to understand, where do these numbers come from? What is this algorithm doing? How are we making these decisions? You really drive the ownership of that when you locate the AI within the line of business. Supported by IT for sure, but we really like to see it driven in a line of business. That seems to be where it's most effective.

And closing that skills gap. Do you need to know a little Python? Absolutely. Do you need to know a little SQL? Absolutely. But it's more important to understand the business problem, to say, "Here's what we're trying to achieve. Here's my data. I really understand my data now. How do I apply a tool to this data to make a decision?" And there are lots of ways to learn that and close that gap. Some are traditional academic routes where now there are analytics degree programs. There's a whole host of certifications and online training as well. And typically your vendors are going to have some training to help you get up and running with their tools, so lots of different ways to solve this problem, but I think ownership within a line of business is key.

FEI Weekly: Great. Those are my questions. I really appreciate you taking the time today. It was a very interesting conversation. Thanks so much.
Justin: Thank you, Chris.