It can be daunting to navigate the everchanging technological landscape. But Glenn Hopper, CFO of Eventus Advisory Group, has been a tech pioneer from the start of his career as a Navy journalist, learning to write code and build webpages in the early days of the Internet. Now, with the rise of generative AI, he has jumped in headfirst, building his own models.
In this episode of “Secrets of Rockstar CFOs,” Hopper sits down with Jack to discuss the power of large language models, where they fall short and how we can foster cautious innovation.
Listen by clicking below. The Q&A, lightly edited and trimmed for clarity, follows.
Listen to the podcast here
In this episode, I’m excited by the guest we have. He’s a good friend of mine and arguably the most technically astute CFO that I’ve met in my career. Glenn, welcome to the show.
Jack, it’s great to be here. Is this going to turn into a roast where you and I are hurling insults at each other or are we going to stay on topic here?
I’ll try to go as far as I can without throwing any insults. You’ll do the same. The contest will be about which one of us cracks first. My shirt is the gift that we’ll keep on giving should we decide to start insulting each other. Glenn, by background, is the CFO at Eventus Advisory Group. What’s interesting is he’s the author of a book called Deep Finance: Corporate Finance in the Information Age, which is one of the best books, perhaps even the very best book, I’ve read for financial leaders who recognize the need to stay technically savvy.
Mastering technology is the defining trait of future financial leaders. If you’re in that category, I would recommend picking up the book. He’s also become a must-follow on social media and one of the few experts that I’ve spoken to. I’ve interviewed quite a few who are both experts on financial leadership and generative AI. It is going to be a great podcast. Glenn, I know you’re busy so thanks for making the time.
I’m happy to be here. I’m always looking for an opportunity to hang out and talk.
Before we get into this, I want to talk about your career progression because it is fascinating, a little about your career progression. You are one of the few CFOs that I know who started as a journalist in the Navy. I’d love to find out how a Navy journalist became a CFO.
As a journalist, a big part of my job through the late ‘90s was moving the magazine I edited to an online presence. I was in technology there. I got married and realized that as a journalist, that probably wasn’t going to be the best way to support a family long-term. I would love to say that I was a sage predictor of the future and knew that journalism was going to be a tough career road but I wasn’t quite that prescient in what I was saying. I saw that I needed to move into the business world. I did have a business interest and did a lot of business writing as a journalist.
I went and got an MBA. My first job after an MBA was not in finance. It was in marketing because I thought, “I know the tech side of it. I can write.” I was a product manager for this website builder. It was a cutting-edge tool at the time. Think of all these website builders you use that are template-driven. This was in 1999. There weren’t many of them.
I was working at a telecom company and I was the product manager for this tool. I was super excited about it but it was hard to get the support that I thought the product needed. With my freshly minted MBA, I decided that I was going to find budget dollars. I have a sales and marketing budget. I started moving that around and advocating for my product.
Somehow, in that role, I became the finance guy for the sales and marketing group. I did that for a while until I was poached by not the CFO but the COO, who wanted someone like me to manage his budget. From there, I moved from a FinOps role and RevOps at one point. I started tracking metrics early. I was managing about a $200 million budget back then. I started interacting with finance and accounting a lot. We started doing telecom in the early 2000s. We’re doing a lot of M&A activities. I became part of the due diligence team and got thrown into the fire early.
I moved my early career through telecom at increasingly higher levels. I got a lot of trial by fire through all the M&A activity that was going on. Once I got a smoothly running department, I had 30-something people working for me. We were running BI for the whole company. I had this real-time financial reporting that we did back then. I decided that was too comfortable and easy. I left that and moved to the startup world. That was my first CFO gig after that. I spent 15 years at four or five different companies as a CFO in the startup world.
There’s no such thing as a conventional path to becoming a CFO anymore. I had a conventional path for people who started their careers in the ‘80s, which was Big 8 accounting firm, accounting manager, controller, and CFO. There’s a lot of us like that. To become a CFO, there are a lot of different ways you can get there because the job itself is so diverse. It’s not all about finance and accounting. The F is for Financial and it will always have that element to it but it’s strategic. It’s about mastering technology and communications.
It’s great that you had a different story than a lot of other people. You’ve had a great career as a CFO, five jobs in several years, which is fairly typical for CFOs, depending on whom you believe. The average CFO tenure is three or four years. What’s a nice finance guy like you writing code to create bots? How’d that evolution come up? It wasn’t the most obvious thing to me.
Necessity is the mother of invention. Here’s how it started for me. When I was in telecom, I was surrounded by tech geeks. Everyone would brag about how great they were at Excel and try to come up with new visual basics, macros, or whatever we were doing back then. This is a million years ago. Everyone was technically minded. I worked with a lot of engineers. I was tracking metrics for engineers. We had such a big team that we were reporting. We were trying our best to do predictive analytics back then.
Churn and telecom are a big thing. We were focused on that and all these metrics we were tracking. When I left this big team to go to a startup and try to raise money, I realized for an early-stage startup that the way that you could communicate whether it was equity or debt financing you were raising, you had to be able to speak the language of either the private equity group, the banker, or a combination of both. To get their attention, you had to present like a bigger company. To do that, you needed more data.
It was my first CFO gig. It was me. We called her a controller but she was a senior accountant. We had a part-time person doing AP, AR, and bookkeeping-type stuff. I knew the level of reports I wanted to get and there were only so many hours in the day. As a CFO in a startup, you’re doing other things beyond. I was managing IT and running HR. We didn’t have people over those.
People underestimate how hard it is to be a startup CFO because there’s so much stuff you have to do. It’s wonderful but it’s more challenging than people realize.
It was like in the early days with websites when I was a journalist. I knew what I needed. I didn’t have time to gather all this data so I started trying to figure out technical ways to get information out of our point of sale systems and project management systems. I’m trying to aggregate that. I got into database stuff. I got way down the rabbit hole.
At that time, we were $5 million or $6 million in revenue but we were able to present like a much bigger company and PE people loved that. I was rewarded for doing that. I realized how much more efficient we could get. We even got into doing real-time reporting. We had a dashboard back then. From the beginning of my career, I’ve realized how important it is to be data-driven.
The only way you’re going to get there is through automation and having as close to real-time access to all the data as you can. The two have been linked. If I’d had someone else to do the heavy lifting of the engineering and how to get this, I would’ve loved to have leaned on them. I haven’t had that luxury. I had to learn and figure this out as I went.
I was a little curious. You’ve created several bots. The first one was Otto the Auditor. That caught a lot of attention. I thought it was interesting unless I missed something. You’ve never been an auditor in your career. Tell us a little about Otto and why you started with audit first as opposed to some other things.
I don’t know if this means I’m petty or vindictive or if I was trying to prove a point. As you know, I’ve been traveling around the country talking about generative AI and the potential for it. For some reason, I’ve had several people in different non-related venues where I’m speaking to finance and accounting professionals. It’s always the auditors who say, “My job could never be replaced. You couldn’t have a software algorithm do an audit.”
As someone who leads companies through audits, I have no ill will against auditors but if you think about what an auditor is doing and where AI and generative AI or the suite of AI, in general, is going, it’s hubris for any of us to say, “My job can’t be replaced by AI.” I work with a bunch of people in audit. I ask them, “What are some key guiding principles, documents, rules, and regulations that you use in audit? If you have to go look something up, where do you go?”
With Otto the Auditor, the idea was, “Let me fine-tune a model on the specific audit regulations that we use and see how good it is at answering questions.” That was a prototype. It didn’t have the ability to analyze data. It was just a chatbot. You could ask it questions about SEC, ASC, or whatever rules, regulations, audit standards, securities, and exchange commission and get answers that were based not just on what these giant language models know but something that was fine-tuned.
Think of the LLMs as generalists and experts, to some extent, on all of human knowledge. If you ask that LLM in a non-specific context, what does a SC stand for? There may be two dozen things that an SC could stand for. That may be a bad example but the idea is that with this specific documentation, when you ask something, it can refer first to that and filter all of its knowledge through the audit standard, in this case. I did all that.
I’m not the best programmer in the world. It took me a few days to build a model. First off, I fine-tuned it. It’s called retrieval augmented generation, where I uploaded actual audit documents but in the fine-tuning, I took a bank of CPA and CFA exam questions and answers and fine-tuned it on a couple hundred of those. Fine-tuning is interesting because it’s almost like prompt engineering. The fine-tuning part of that project was to get the model to be in a mind space where it’s answering finance and account or primarily accounting questions. Once you have it in that mindset and you point out these specific documents, you have a tailored specialty bot.
You said LLM. That was an acronym I didn’t immediately recognize. What was that?
Large Language Models. All these ChatGPT, Anthropic’s Claude, or Google’s Bard are chatbots that all start as a large language model. We don’t want to get deep into the weeds on what they are but the model itself, think of it as like the predictive text when you’re writing an email or typing on your phone. It’s that predictive text. You go through some reinforcement learning, a human feedback thing where you train the model to be a chatbot and interact rather than auto-completion. The easiest way is ChatGPT is at its space and a large language model.
Rumors are floating around that you named Otto after the blowup pilot from the Airplane movies. Is that true?
I love that reference but the blowup pilot that I remember is from the movie Airplane, which was based on the show first. I liked the alliteration, Otto the Auditor. The icon for it is an otter. It’s Otto, the Otter Auditor.
It’s funny because I was thinking, “Why did you come up with this name?” I was thinking of Otto Preminger, the director, or Otto Bismarck, the leader. I’m overthinking this but cool. You live with Otto but you’ve created nine.
It’s 12. Here’s the crazy thing. This is the second time that this has happened to me. I’m okay with it. I spent a weekend building Otto the Auditor. I rolled it out on a Tuesday, and on Friday, OpenAI had its dev day. What had taken me all weekend to make anyone could make one of these in five minutes. You can build these zero codes on the ChatGPT website. You can interact with it as you do with ChatGPT itself. You can say, “I need to create a specialist that does XYZ.” If I said I need to create one who’s a specialist at audit, it could make an extended prompt. It’s more prompt engineering that guides the chatbot to answer in an accounting way or be in that mindset.
The next thing you can do is the retrieval augmented generation that I talked about before, where I uploaded the documents. Now, you drag and drop them into this GPT that you’ve built. You can fine-tune the prompt, which is what I did with my CPA and CFA exam questions. You do this by writing out sentences and telling it what it’s an expert at. You back it up with actual data for whatever you’re building. It’s a template. You could use this for finance and accounting but you could do the same thing for customer service or selling real estate. Whatever documents that you refer to, you have a tool trained to use these.
Once I saw how easy it was to do that, I thought, “Let me build out the full finance and accounting suite.” I’m still refining them but I’m happy. I’ve run not scientifically because I have a day job and don’t have time but I’ve run tests where you use the fine-tuned GPT model versus the off-the-shelf ChatGPT or Bard. I am seeing better results. I shouldn’t say that when it’s anecdotal but I’d love to get more people out there using them and see what experience they have.
We’ll make sure people can reach out to you afterward and get involved with some of this stuff. There are three that I was initially aware of and I didn’t know until you’d added the nine others but it’s Rosie the CFO, Claire the controller, and Francine the FP&A Bot. Each one has its specific knowledge set. I ran a little experiment and it was good.
I asked Rosie the CFO. I described an economic scenario, which is fairly current to what’s going on, and maybe a remedial user compared to what your ambitions offer this. I said, “We’re in a time when capital markets are slower than what we’re used to. Inflation is rising and interest rates are higher than they’ve been in a while. We need to prepare for supply chain disruptions. Help me optimize my working capital strategy.”
It comes out with great stuff. It’s like, “Enhanced receivables management.” It gave some steps for that. Optimized payables, inventory management review, financing options, cost management, supply chain resilience, risk hedging, and strengthening cashflow forecasting. Each one of them individually may not be rocket science. It came up with a thoughtful strategic plan for a highly specific, and in this case, what is an accurate thing. To me, it’s fascinating that you were able to create something that did that.
I am still refining these and the technology will get better. I’m trying to keep from going dorky with how I talk about this but there’s a concept of agents where you have to give an algorithm and a goal. The algorithm sets out on its own to answer that goal. This agent could employ other agents to help with the goal. I’ve fine-tuned Rosie with some specific instructions and documents.
One of the bots I created is an economist. What if you had this one agent that was trained like a CFO? Ideally, the CFO knows everything but we know we can’t all be experts at everything. You give it the tools that you need but what if Rosie could call on her controller for certain questions, call on the economist for other questions, or an FP&A person, and you have these agents working together, whereas instead of the user prompting at every step, “Tell me the back and forth there.”
You could ask that question to Rosie. She could gather her team, take the insights from all of them, and give you a consolidated report of what you should do. Our economist says the leading indicators. It’s not there yet. I’ve got a day job. It takes plenty of time but I also feel like people need to see what’s coming. I’m advocating for this. I’m not saying people should go out, fire their staff, and turn it all over to my bot but it is the time for us to familiarize ourselves with this.
When there are accessible tools for these issues with hallucination and all the problems that people have with generative AI, as those are worked out, we need to be in a position where we’re ready to use them. That’s what drives me to do all this stuff. All the bots I created are free. I’m not out there trying to sell these but I do think it’s important for finance and accounting people to get out there, try it out, and see what’s possible. Build your own if you can build a better CFO than Rosie.
That’s the advice I always tell people. You have to stay on the cutting edge in any job, particularly CFOs and technology because if you’re not on the cutting edge, there are a bunch of other people who are going to be on the cutting edge. They’re eventually going to take your job if you think that you can’t evolve. It’s an evolve-or-die type of stuff with technology.
The tools are fascinating. It’s a little crazy that the bots are able to talk to each other and a CFO bot can consult with an economist bot and a controller. How does this change day to day? You’re a CFO. All of a sudden, you’ve got all these amazing tools at your disposal. How does that change what you’re doing day to day? It probably leaves you more time for strategic thinking in leadership and less time for traditional finance and accounting school skills.
There’s not any unless you were using one of the applications, over $100,000 a year. If you’re a big company, it may not be a lot. For SMBs, which is the space that I’ve primarily worked in, it’s expensive. I am telling clients and anyone I talked to about this, “This is not the day for you to upload your company’s proprietary information into the Bard, Claude, or ChatGPT, the publicly available tools that are out there. It isn’t the time. Go ahead and grab public company information and sample data. Play around with them and understand because the day that they get here is coming soon.”
I keep wanting to laser in on generative AI but that’s one slice of the pie. It’s the one everybody’s talking about. We’re at the peak of the hype cycle on generative AI. It’s the most accessible and the ones people understand the most. Outside of that, we’ve been using AI itself at scale for several years. Think about everything from spam filters in your email to recommendation systems to self-driving cars.
All this is AI but it’s happening in the background and we don’t understand what’s going on with it. With generative AI, we can interact with and see what’s happening in real time. The suite of tools that will come out of AI is going to be used for all the repetitive and menial tasks that we’ve had to do as part of our job, whether it’s information gathering or data entry. That’s going to be the first to be eliminated.
I understand the impact that’ll have on certain entry-level jobs but in a time when fewer people are going into the accounting profession where we have more at the top retiring, there’s this gap that we can’t fill and you think, “If I go get a Master’s in Accounting, I sit for my CPA, and I get that, do I want to start at this low-level job where it’s a lot of data entry? Do I want to start adding value right away?” It makes it maybe more appealing. Once you get through the education and certification, instead of reconciling bank accounts and doing depreciation entries, you’re taking this information that was already done and adding some value to it.
Are you saying that bank reconciliations and depreciation schedules aren’t exciting for a person starting their career?
As someone who’s been in the startup space, it has been surprisingly not very long since I’ve done depreciation entries or reconciled the bank account. It’s one of those things that I thought in my 50s, I would be beyond that. I’m still getting deep into that but that’s not the best use of my time.
I apologize. I disrupted the greater point that you were trying to make about young people and the future of the profession.
Here’s an interesting thing and maybe I’m meandering too much here but we’re drawing from the same pool of people who are being drawn to business analytics or data science. Some people who tip historically would’ve gone into accounting or moved more to finance because salaries are a little higher in finance and even higher still in business analytics or data science. We’re drawn from the same pool. It’s not all journal entries and bank reconciliations that come into accounting because you’re going to be able to provide value earlier. You’re going to be the recipient of this information. Come up with a treasury plan, a risk assessment, or whatever greater value you can add.
If we can automate that lower-level stuff, it makes the profession more appealing. It makes the human-in-the-loop part more valuable. As humans, there’s always a chance for an error for fat-fingering something, and the efficiency that’s gained by having a computer do it is significant. We need to keep finding as more steps of the finance and accounting process get automated. The challenge for us is to keep learning more so that we can keep adding value on top of that.
We’re at the point when computers can do everything better. I know people who are reluctant to use self-driving cars. I understand the concern. There are about 50,000 car fatalities per year in the United States. Rather than being nervous about it, we should look at this as an opportunity to get that number down to eventually close to zero. I’m not one to think it would ever be zero but if we can get 50,000 a year to under 500 a year, that’s a significant improvement. I’m one of those people who are constantly working. If I can get an extra 45 minutes a day to check my email when I’m not driving, that’s a good thing.
I want to ask you a slightly different tack on this. CFOs are not the early adopters. In my view, most of them are expected to be the ethical leaders of their companies. Are there any ethical considerations that CFOs need to keep in mind when developing technology? I leave it to the answer but what I was thinking of is data privacy and security issues.
It is a great question and one that we have to be aware of as we navigate this. Let me address data privacy and data security first. I think about SOC 2 and SOC 1 audits that you get where if you’re getting information out that you’re providing to auditors who are making business decisions, it has to be repeatable and explainable, whatever that system is doing.
Everybody has heard the black box term with AI, especially in generative AI. You can ask the same question 100 different times and each response is going to be slightly different. Most of those will be similar but if the AI goes down a different path, it could start hallucinating. By asking the same question over and over and logging the answers you get, you start to see, “How can I be sure that my answer is the correct one?”
Understanding the models, how they work, being able to have guardrails in control, and the output are some of the developments that are being made by the large language model, by the Googles and OpenAIs of the world to make sure that when you enter a question in, it’s putting out the same information every time that you understand why it gave that answer.
In doing that, I’m not telling any companies to upload their proprietary information directly into ChatGPT or any of the other tools that are out there. You can read the disclaimer from the websites that warn you about putting in your information. A lot of them are saying, “What you put in here isn’t being used to train our models but someone at OpenAI or Google could go in and see everything that you uploaded. They could get hacked.”
The risk with all these large language models and the training that’s going on with them isn’t necessarily that your information is going to be used to train the model. You wouldn’t put your Social Security number into a website unless you had to. Until these systems are able to be enclosed in an enterprise system, the level of security guarantees AWS, Microsoft, or Azure would have.
Until it has that level of lockdown on it, I’m telling people to not put in their information. When it gets to the point where there are tools that are available where the information has SOC compliance and these privacy guarantees, it’s wide open to use. You treat it the same way you would anything else. We can either store our internal information on our internal data center that we manage or trust it to AWS or another hosting company. That concern will be the same, whether it’s AI or anything else that we’re doing data-wise on the cloud.
The other thing I want to mention quickly on ethics is every AI algorithm has the potential for bias. When you design the algorithm and put the data in, you try to limit this but bias could be whatever data set you have. I’m mentioning Amazon. Years ago, when they tried to build an AI tool that would select employees that were like the current employees that they had at Amazon because they liked the employees they had, they tried to replicate this.
They loaded all the current employee’s information into an algorithm and said, “Hire people like this.” What they didn’t count on was that the model became biased and discriminatory because it was hiring mostly White men. It happened at that time. By not taking that into account, they built a bad model but that can happen in every model.
I’ve heard stories like people have said, “Create an image of like a nurse,” and it’ll create a female. Create an image of a doctor,” and it’ll create a male. It’s not the fault of the model. It’s playing the averages. You do need to understand that it’s got some biases that aren’t necessarily anyone’s fault. It doesn’t mean that the biases aren’t real and you can ignore them. You have to be aware they’re out there and adjust.
One I always think of is, if you don’t think about the consequences of what you’re training it on, how you’re using the data, and how you’re letting a machine make a decision for you, redlining is a perfect example of bad data that is discriminatory in housing loans based on a ZIP code. They can tend to be discriminatory. You need to be aware of this.
That’s why I preach how we have to understand. We don’t have to become engineers, data scientists, or developers but if we’re going to use these models, which may be sooner than later, we’re all going to be using them to some extent, you need to understand at a fundamental level the technology that’s behind it because otherwise, you throw your fate to the wind. You might as well shake up one of those magic eight balls. Ask your financial question, shake it up, get your answer, and rely on that.
With driving a car, I never felt like I needed to understand how the internal combustion engine worked. It’s not relevant. It worked but this is a little different. You need to look a little bit under the hood here, particularly when you’re the leader of a company and dealing with the most sensitive of data that any company has under your auspices.
We’re right to be risk-averse. That’s in our DNA as finance and accounting people. That’s great but being risk-averse doesn’t mean ignoring the biggest technological innovation, at least since the World Wide Web, the smartphone, and electricity, if I wanted to be so bold. This is going to be significant. We can’t ignore it. We’re not ready to dive in head first but it is time to start learning about it and plan around it.
I’m a little older than you. I was thinking of the wheel or fire. That’s a generational thing going on. It does have a World Wide Web feel because at this point in my career, when the Internet came out, I was like, “That’s crazy. That’ll never catch on. What is this?” It was the biggest game changer, not only within our careers but the way we live our lives. It would be hard to overstate the impact that the Internet has had. This feels like it has the potential to change everything.
It’s still so unknown. If we’d ignored the Internet in the early days of the mid-‘90s when the first web browser came out, we would’ve done at our peril. We might’ve avoided the dot-com boom if we let it shake out a little bit. That’s a lesson learned about AI. Learn about it and don’t invest your life savings in it at this point.
With the Internet, a lot of people got fabulously wealthy from it but there are a lot of businesses that should never have gotten funded that did and some of them even went public. If you look back, it’s like, “Why was this a savvy business model?” People are placing their bets. They figured, “If I make 10 bets, all I need is one or two of them to come out, and they can certainly pay for the losses on eight.” It was a crazy time, to say the least. I want to ask you a question on gen AI and putting on your crystal ball. You’re as qualified as anyone to ask this. What do you see the CFO role being several years from now? Is it the chief data scientist? It feels like it’s going to change everything but I don’t know what that means.
I have a slightly different view on this from a lot of people. It’s because I came up through FP&A. When you’re building your models, you’re trying to go, “What correlated data, internal or external, that I can use to make my model better? How can I get more predictive? How can I make my forecasts better? If we’re going to be held to this, what can I base it on other than a hunch or doing a straight linear regression for prior performance?”
For me, to be good at financial analysis, you’ve had to be a data scientist before but the expectation is that you would be good at finance. You could write code and query SQL databases and know what JSON was and all these other tools to get data. That’s too big of an ask. One of the things that generative AI brings is these tools and algorithms are already out there.
If I think back to school and try to remember, how did I build an arima model to do forecasting in taking my time building a model versus I could do the same thing with generative AI? I can say, “Here’s five years of data for revenue cogs and expenses.” Whatever time series data I want to feed it, I want to say, “Based on ARIMA, deseasonalized, detrend this data, or do a linear regression and run 1,000 Monte Carlo simulations.” I can do this all in words, not writing a program.
CFOs aren’t doing this themselves but in my time, I get shifted from either building a report or asking for a certain report and waiting for it and all this information gathering. Instead, I have all these complex models that I got instantly. I can shift more to a data-driven decision-making process and bring in more data, both internal and external so that my team would be able to.
In executive-level jobs, not just CFO, COO, CRO, or whatever you’re doing, you’ll still lead the way you do but you’ll have many more tools to make decisions. You’ll see trends and correlations. You’ll understand the big picture. I’ve been doing this for a long time. I have a hunch as to how this will work. I’ll have access to even more data so I can turn that hunch into a hypothesis, run through it, test it, and get results.
We’ve got to have our domain experience but we also need to understand what is possible with data science, analytics, and algorithms so that we know the right questions to ask. I do see there’s still the human need for input but we have a lot more information and tools at our disposal. We need to understand how to use them.
It’s fascinating because, at the start of my career, the best accountant would become a CFO. If you were a good accountant, you had a good shot at becoming a CFO. It is still a useful skill but when you think about everything you said about how it’s going to change within five to 10 years, that’ll be the new reality for chief financial officers. I’m curious. What do you see as some of the biggest obstacles that prevent, specifically in the world of finance accounting, from adopting these tools in cultural business, whatever it might be?
The biggest one for me is, at the end of the day, the numbers we give have to be right, whatever decision we make on them. Maybe there’s room for error there but you have to be trusted when they get the numbers from you, whether it’s a private company or a public company. Sarbanes-Oxley is signing off on the financial statements. You have to give good numbers that are explainable and that you understand.
Generative AI is not there. That’s why I’m not telling anyone. All the bots that I’ve built are proof of concept. I’m not saying fire your team and use these things but see what is potential there. I’m the biggest advocate out there but I will not even trust but verify. I’m going to start with don’t trust, go verify and say, “You got it right this time.” That’s where we are with the technology.
It’ll be solved for them. If any geeks are following, there’s this program called QStar. That is no relation to QAnon. This program, QStar, is doing some basic rudimentary math. Once they expand that, the technology gets better. Whatever your measurement is, when you get to what the machines are doing, you can trust it as much as your controller or whoever else on your team is giving you information.
We’re at trust but verify. I’m not telling anyone to rush in and use it now. Once that’s overcome, there’s a fear factor to it of, “If I give this over to our new robot overlords, what am I going to do with my time?” Trust is the biggest cultural change. There’s an automated close but it’s still human involved. Are you going to turn your entire month-end close process over to automation? Where are the checks going to be? It’s a lot. At this point, we type into a calculator and get our answer. We don’t question it at all. With the pace of this technology several years from now, we’ll be at that, if not several months from now, with the pace it moves.
I know a person who is trying to write a book on generative AI specific to finance people. He can’t because every time it gets through the editing process, it’s changed too much. He’s like, “It’s going to be out of date the day I publish it.” How do you do that? It’s exciting but it’s got to be frustrating that everything you think you know is going to be wrong.
One thing that is obvious to me is we live in a situation where regulations and business standards are always evolving. Some people would like to say, “Gen AI should be exempt from that.” I’m not one of those people. How do we keep gen AI updated and relevant in a complex regulatory environment that’s always growing? Are there things we should be thinking about?
I think about it from the regulator’s side. I keep going back to SOC 2 and SOC 1 because they’ve done a good job of understanding technology. It’ll be interesting to see how they evolve for rules around AI. On the congressional lawmaker side, I don’t know if there’s anyone in government who understands the technology enough to regulate it at this point. The danger is making rules against something that you don’t understand. We don’t want to stifle innovation but we do want to be careful.
This is the whole dust-up at Open AI. This technology has the potential to make big societal changes. How are we going to prepare for it? Are we going to open the floodgates and let it run amuck? Are we going to cautiously and carefully move forward? Ninety-nine percent of the population says cautiously and carefully move forward. As laypeople, because we’re not machine learning and AI engineers, we don’t understand everything that’s going on there so how do we regulate this?
There’s got to be one base level of education that everybody needs to understand. Everybody who’s going to be using it or making regulations about it needs to fundamentally understand what’s happening. There needs to be this partnership between the companies, the engineers who are building, and the governing bodies to figure out how to set these guardrails in place. It’s not just about accounting standards. It’s also about laws, military applications, and terrorist apps. There’s a lot that could go sideways with this. We need to figure it out. I’m talking in circles to say I don’t have the answer but somebody smarter than me needs to be working on how we come up with this.
Glenn, that’s only five or six people who are smarter than you if you’re windowing that down. I’ll cut this part out but I’ve been closing these things with two questions. I don’t think I prepared you for them. I did it on LinkedIn Live, which we did a couple of years back. One of them is, do you have a go-to joke or a hidden talent? Is there a fun fact about you that we could ask? I can ask you that question, or given that I believe I forgot to tell you about it, we can skip it altogether. The final one is some advice for the next generation of financial leaders.
Let’s go with a fun fact. I’ll mention that I have a 200,000-word space opera that I have not finished. It’s been sitting on my hard drive for several years. I’ll throw that out as an off-the-wall thing. I can give advice.
Glenn, this has been fascinating. I learned so much and I know our readers will too but I always like to close on a little bit of a more personal note. I’m wondering. As a finance guy, do you have a fun fact or something about you that people would find surprising?
My default on this is I once shot a man in Reno to watch him die. That’s not true.
That’s no time in life or something from our childhood.
It’s a 200,000-word space opera that has been sitting on my hard drive since COVID. I haven’t done anything with it because I’ve been focused on all this AI stuff. I hope to get back to that. One day, it may end up having to be self-published because it might be absolute garbage. I have a 200,000-word novel called The Baralian Problem. It might be a little bit about AI.
Is it space, as in Star Wars, the Jetsons, and that type of thing?
Yes. It’s a drama set in the future.
You jokingly said that you shot a man in Reno to watch him die and then the space opera. I’m not sure which of those two is more believable. I’ve known you for a while. You’re not a violent guy. I’m siding towards the space opera but that’s crazy. What inspired you to write a space opera?
It’s the same thing that led me into journalism initially. I’ve always wanted to be a writer. That’s how I ended up writing the finance book. Years ago, I wrote and produced a movie. I have a hard time sleeping and I have some weird hobbies. They’ve been hobbies and haven’t netted me a nickel but it’s a fun right-brain and left-brain switch from daily work to a creative outlet.
I enjoyed writing both of the books that I wrote. They’re not great books and I certainly didn’t make much money off of them but it’s such a different process from what we do in our day-to-day things. I encourage anybody who has that creative juice to go ahead and write it. With my second book, I don’t even know if 500 people have read it. I enjoyed writing it and maybe the 500 who did learn something from it.
I’m going to give a plug for your book. With the second book, I didn’t know what to think when I got it. I thought it might be a tongue-in-cheek corporate overview. I was pleasantly surprised at how great and astute it was. It made me think about things looking back at my career and the CEOs I’ve worked with. I highly recommend The Psychopathic CEO.
On Goodreads, I can read the reviews of it. Sometimes, I torture myself doing it but you can not only see on Goodreads who reviewed it but also other books that they reviewed. There are people who are classic. They’re basing it on the cover or the title. They think that they’re buying a work of fiction. They’re buying all these mysteries, murder stories, and true crime.
My business book has a catchy title. I was like, “No wonder you don’t like it. You didn’t even read the description of the book before you bought it.” It does have an intriguing title. On a closing note, I’d love to get your thoughts on some advice you can give the next generation of financial leaders. What are things they need to be thinking of, whether it’s technology or other areas?
I do think beating the same drum is the best thing if you’re starting in your career. It’s already part and parcel with the position but if it’s not, whether you’re on the FP&A finance track tracker or the accounting and audit track, understand the interest in the technology that you’re using and where technology is going. Not that you have to become the technologist yourself but understand what that technology brings.
The one that I go to immediately is always data science. If I have all this data and I can easily get reports on it, how do I know the questions that I need to ask so that I can add value to the data? The way you’re going to get that is having an understanding of business analytics and data science in general so that you can say, “I’m looking for a leading indicator. I need to find this correlation.”
When you see a correlation matrix, you understand what it means, you know how to wait and build a model. You can think of the features to put in your model. Ways to use this technology are going to be important to make your job better. Understanding analytics and working with large data and small data sets to create more value is going to be an important focus. I see a lot of finance or MBAs that are paired with analytics. Finance and accounting people are taking Computer Science and Analytics courses. That’s going to be table stakes going forward.
How can people follow you online?
LinkedIn is the best place. I’ve got a Twitter account but I don’t understand Twitter.
If you want to follow Glenn online on LinkedIn, it’s fascinating stuff. It’s technical but it is done at the level that the CFO needs to digest it. He’s a must-follow, in my humble opinion. Glenn, thanks again for your time. I appreciate it. I know how many things you’ve got going on.
Thank you, Jack. It’s always a pleasure.