In this episode of Financial Modeler's Corner, host Paul Barnhurst welcomes Lance Rubin, Founder of Model Citizn, to discuss how AI is transforming financial modeling, accounting, and FP&A. They explore the risks of AI-generated financial models, the importance of human oversight, and why strong financial modeling skills remain essential.
Lance Rubin is a Chartered Accountant and financial modeling expert based in Melbourne, Australia. As the Founder of Model Citizn, he specializes in financial modeling, data analytics, AI-powered solutions, and data-driven decision-making. He also provides fractional CFO services and helps businesses improve their financial processes using Excel, Power BI, and AI.
Expect to Learn:
Why AI-generated financial models can look accurate while containing serious errors.
How human oversight helps prevent costly financial mistakes.
Why Excel and Power BI remain essential in AI-powered finance.
How AI skills, workflows, and guardrails improve reliability.
Why fully autonomous finance systems present significant risks.
Here are a few quotes from the episode:
“The human has to be in the loop. The human has to check it.” - Lance Rubin
“I like to translate all AI work into formulas, into deterministic logic that I check, then I can trust it. Otherwise, I can't.” - Lance Rubin
Lance shares insights from his AI-powered accounting book and training program, explaining how finance professionals can automate repetitive tasks, build reliable AI workflows, and improve productivity without sacrificing accuracy, transparency, or critical thinking.
Follow Lance:
LinkedIn - https://www.linkedin.com/in/financial-modelling/
Website - https://www.modelcitizn.com/
Disclosure: Portions of this episode (such as the introduction or promotional segments) use AI-generated voice narration produced under human editorial review.
In Today's Episode:
[00:00] - Introduction
[03:50] - Career Corner
[07:41] - Lance's Financial Modeling Career
[14:35] - Writing an AI-Powered Accounting Book
[21:52] - Five Levels of AI-Powered Accounting
[28:12] - Why Excel Skills Still Matter
[34:28] - The Risks of Uploading Financial Models to AI
[40:46] - Rapid-Fire Financial Modeling Questions
[47:43] - Closing Remarks & Thanks
Full Show Transcript:
Host: Paul Barnhurst (00:41):
Financial Modeler's Corner is the world's premier modeling podcast. It is brought to you by Financial Modeling Institute, the world's leading financial modeling accreditation organization. Welcome to another episode of Financial Modeler's Corner. I'm your host, Paul Barnhurst, AKA the FP&A guy. And this week I'm thrilled to have with me Lance Rubin. Lance, welcome to the show.
Guest: Lance Rubin (01:06):
Thanks Paul. I think it's my third time. So yeah, great to be back.
Host: Paul Barnhurst (01:10):
I interviewed you first on FP&A Today. That would've been what, four years ago now? Then you came on the show and then you were part of the great debate where you and Danielle argued whether AI could build models. Danielle said they'd never be able to. You said they would. And I brought that up recently, and Danielle's like, "You had to remind me of that. That didn't age well." And then this time, we're excited to have you. As everybody knows, the show is sponsored by the Financial Modeling Institute. The reality is in the age of AI, knowing how to model is more important than ever. And one of the best ways to demonstrate that is the Advanced Financial Modeler. So I'd encourage you to take that if you haven't taken it. We'll jump into our episode. So before we go into career corner, I'm going to ask a couple of questions about your career.
(01:54):
Can you just give a quick bio, a little bit of rundown of yourself to the audience?
Guest: Lance Rubin (01:58):
Sure, Paul. So Lance Rubin, I'm the founder of Model Citizn Financial Modeling Consultancy. I started the business almost a decade ago and I've branched out and I've gone into the data world, doing data engineering, Power BI. So I don't just talk Excel; I talk Power BI as well and data. Also started to launch a fractional CFO offering because I realized that we would build good models for clients, but they couldn't get the accounting and the bookkeeping right. So we sort of now packaged the whole lot. I've been a trainer and a speaker on a range of different programs and training. And then I wrote a book called AI-Powered Accounting with Excel and Power BI and launched a program called the AI-Powered Accountant on Maven. And I think it's an exciting time, but it's also a frightening time for everyone. And I just want people to listen and understand what AI actually is and how to navigate it because you just see so much slop out that you and I spoke about recently.
(02:51):
And I ran the finance lab, which is a real look at the hype and just calling it out and actually looking at the practical steps of how you actually get value. And I've seen it. I've implemented in my own business. I implement it with clients and I'm running AI projects now as well. But there are concerns and there are things that we need to remember. The foundations are critical.
Host: Paul Barnhurst (03:13):
100% agree with you. So just so everybody knows before we jump into the episode, Lance in this episode is going to be wearing the hat of a fractional CFO, CFO, that's what the perspective he wrote the book from. We're going to go a little deep there. Ever wish you could get career advice from the best financial modelers in the world? Well, now you can. That's why we created the career corner. Let's go ahead and see what today's guest has to say. So Lance, what is one skill that you have developed as a modeler most people would be surprised by?
Guest: Lance Rubin (03:46):
The skill that most people that get into financial modeling would be surprised by, I think, is the ability to go into detail and stay high level. I think as a financial modeler, you sort of hone your skills in the detail, in the cells, in the sheets and the dashboard and all of that sort of stuff. But the ability to really hone and zoom out and communicate complexity simply is a very critical skill. Now, people might say yes, well they know that, but I think that's probably people that are later in their career. I think people that are starting out, particularly people wanting to get into financial modeling, which I think a lot of your listeners are, that don't know how to build financial models or not familiar with art and the skill of financial modeling, which is both art and science as you always say, which I agree with, is how do you articulate that higher order thinking and explanation?
(04:35):
Models can be very simple, but they don't cut it because they don't do the calculations properly and so they miss the mark. And very detailed mechanical models that are mathematically on point are complex to decipher and communicate and understand. And so you need to bridge the gap. You need to have a robust model that has scenario management, that has three statements, there's cash flow. Obviously I'm coming from a fractional CFO perspective, cash is king. I need to understand cash. I need to know my balance sheet balances. And so all those things, I then need to talk to someone who may not know accounting, may not even know Excel. And so how do I communicate something that is as complex as a multi-entity global rollout, which is a client that we're working with and explain this stuff simply? Now they know in their heads that it's complex, but they look to me to actually explain it simply to them, not wanting to know the detail.
(05:32):
And so there's that skill of how much do I tell? How much do I not tell? To what level I go into detail? Sometimes I share too much and then it freaks them out. I'm like, I've lost you. Sometimes I don't share enough and then I'm like, oh, I didn't know you assumed that. What does that mean? Or what is the scenario? So clients have asked us for very simple two pages on the model and we now use AI. My Bob digital worker builds a mechanic explainer sheet, which basically explains the entire Excel engine logic in one sheet, explains the engine, explains the assumptions, where's the core? Now we just didn't have time to do that. So I think those are the things which I think are going to help with that skill that often people think, well, being good at financial modeling means I just need to be good at Excel and be very technical and in the detail and check every cell and that's it.
(06:28):
It's like, no, well that's only half the story, half, not even three quarters. It's half the story because you could build the most complex model. And if you can't communicate that at a level which is above the formula level in a way in which that actually is understood by non-financial people, then you're missing literally the other half of that value that you can add as a financial modeler. And that's the skill that I think is not taught enough and it comes with experience. So people who've got experience will say, yeah, yeah, that's expected of course. But I think people that are starting out won't necessarily see that in the beginning because they're so honed in on the detail, on the cell level, on the formulas, on all the things that they've got to learn now, which is a lot.
Host: Paul Barnhurst (07:13):
Great point. I think it's one that you get more experience as you're younger, it's hard to learn. Alrighty. So next question I want to ask, coming out of school, did you have any idea that financial modeling could be a career path?
Guest: Lance Rubin (07:26):
I had no idea. I mean, when I came out of school, Microsoft Excel wasn't really around. I saw Lotus 1-2-3. My father used to work on a Lotus sheet in DOS, which is a green screen. Windows 95. So I finished high school in 93. Windows 95 was the big change. So yeah, I had no idea that financial modeling even existed at school. I just knew that I like numbers. I wanted to work with businesses and understand business, understand money. So I figured, well, accounting is a natural fit for that. So I joined PwC in the graduate program and eventually became an auditor and an accountant and worked through PwC in Johannesburg and in Sydney. And so for me, it was only then in that role that I had to audit an actuarial model. They didn't even have a financial modeling division back then. This is way before Ian Bennett's time.
(08:35):
Ian was still in the UK. So I worked with PwC in Sydney and the modeling audit work was thrown to the auditors, not to the modeling team because it didn't exist.
Host: Paul Barnhurst (08:46):
93, you are old.
Guest: Lance Rubin (08:50):
Thank you.
Host: Paul Barnhurst (08:50):
I could only say that because I graduated in 94.
Guest: Lance Rubin (08:55):
Well, I have passed 50, so I think that does make me a little bit old, but why?
Host: Paul Barnhurst (09:01):
You and me both.
Guest: Lance Rubin (09:01):
Hopefully wise, Paul, not just old.
Host: Paul Barnhurst (09:05):
Yeah, hopefully we've learned a few things along our years. Yeah. No, I remember the pre-Excel days, the Lotus and all the different tools, Quattro Pro and different things.
Guest: Lance Rubin (09:16):
Never touched Quattro Pro or was just Lotus and Excel?
Host: Paul Barnhurst (09:20):
Yeah, I touched both Lotus and Quattro Pro a little bit. And then I used some of Netscape had a Office Suite and sometimes I'd also use Corel's because I didn't like Excel in college. I tried to avoid using any Microsoft product as much as I could. Go figure.
Guest: Lance Rubin (09:40):
Yeah. I look with you now, right?
Host: Paul Barnhurst (09:42):
Now I use it more than I probably should. All right. Instead of asking you your worst financial model ever, what's the worst model you've seen? So not the hype, but worst model you've seen someone's built with AI. What's the worst thing you've seen?
Guest: Lance Rubin (09:57):
The worst thing I've seen is, I guess it's hard to pick unless you experienced. So I think the thing is what's worse for me is something that looks good that is fundamentally wrong. To me, that's the worst. That is the worst time because if something doesn't look good and looks messy and it's all over the place, well, you can kind of guess straight away that it's going to be bad. That's obvious. That's not the worst to me because if it's obviously wrong, that's actually easy. When it's not obviously wrong, and particularly if it's given to you by a client that says, "Hey, we've got this model, this is what it is," and you look at it and you start pulling it apart and you start asking questions, you kind of realize that fundamentally the model is not wrong. However, the way in which it's being constructed and the logic that has been applied is misleading.
(10:55):
So wrong is not... There were no formula errors that I found in the model. In fact, the formulas were fine, but what was incorrect was some of the assumptions and the back solving and the prettiness and everything else. So when I started to pressure test some of the assumptions and there was a backsolve number to get two numbers to balance, I'm like, "What is this percentage?" I could see him sort of shaking like, "I'm not quite sure. I have to go back to my notes," was the response. And then like, "Well, hang on, that's a fundamental input into this calculation and it's just sort of a backsolve number that AI's created with no real basis other than, well, the purpose of this number is to get this number and that number to match." So it's just like that's the basis of the number. I was like, "Okay, well what does it actually represent?" And then even when I ask a simple question like, "How much capital do you think you need?
(11:56):
How much equity?" This is a startup. So startup had a financial model, sent it to us, looked all good on the surface. And so as we started to peel away the layers, we realized, no, it's got some fundamental flaws in it and is not investable in its current form. So the model would not give any form of investor confidence. And the order of magnitude difference on the capital was significant. So he said, "Oh, well, we need..." I'm not going to disclose the number, but it's whatever that number is, it's double. So the reality is that the number that AI told you in that model is double. And the reason it's double is because it's put something in the opening cash, which actually doesn't just magically arrive in opening cash. It has to be invested in, and then there's a subsequent draw of more cash. So you've got to add those two together.
(12:53):
So the numbers were there, but it was just hidden. And I think that's the horror story. That's the thing that should make us concerned, because if we're going to stand behind these AI-built models without kicking the tires and seriously kicking the tires, we are risking our reputation, we are risking money, and with all the hype that's out there about AI-built models, the human has to be in the loop. The human has to check it.
Host: Paul Barnhurst (13:20):
Every great modeler on this show says the same thing. Simplicity is a key skill. The AFM, the Advanced Financial Modeler, is built around exactly that discipline: clean structure, sound logic, defensible work. Join me in becoming an AFM. Go to http://www.fminstitute.com/podcast. That is fminstitute.com/podcast, and use code podcast for 15% off.
Well, as Ian Bennett has said, and I've had him on this show and I've quoted him many times, "We trade in trust and it doesn't matter. People have to be able to trust that model. If they can't trust it, you got a problem.
Guest: Lance Rubin (14:08):
Massive problem. Yep.
Host: Paul Barnhurst (14:10):
I'm with you on that. All right. So you wrote a book. First off, did you ever think you'd be an author, that you'd be coming on a podcast to talk about a book you wrote?
Guest: Lance Rubin (14:18):
I think at one stage, about two or three years ago, I did want to write a book. It wasn't the book that I wrote. I figured that there's a lot of... And that book's still sitting in my draft pile when I'll get back to it at some stage. The work which you were part of, Paul, was the financial modeler's guide, the knowledge hub. It had lots of different perspectives and putting that in a book was something that I did want to do and I still want to do. So did I think I would write a book? Yes. Did I think it would be a book on AI? Absolutely not. In fact, about this time last year, I almost canceled my AI training program altogether. And if you asked me 12 months ago, you'd write a book on AI, tell you crazy, this stuff's useless, it doesn't work, it hallucinates, it's terrible.
(15:15):
And then Opus 4.5 dropped. It was around towards the end of last year and suddenly -
Host: Paul Barnhurst (15:20):
And November, December last year, whenever, I think it was November when Opus October.
Guest: Lance Rubin (15:24):
Yeah. So Opus 4.5 for me was the watershed moment on where everything changed. And then that's when I created the training course, the training program. I implemented it in all my workflows as a fractional CFO. I've trained my team. We've now launched a new product called Financial Intelligence as a Service, so fixed fee service for FP&A, CFO work as well. And then the book. And the book's just a combination of all the training that I do in the program, but also the experience I've built over that time. So the short answer is no, I didn't. And I think that's the exciting thing about AI is kind of what things it's going to create that we are not aware of.
Host: Paul Barnhurst (16:08):
So how long did it take you to realize when four or five came out that everything's different? Because I think that was a watershed moment for a lot of people in AI. I remember on the mod squad when we were like, "Wow, this is actually good."
Guest: Lance Rubin (16:23):
Almost that week or the week after, after it dropped. So we run an eval process. We have an evaluation process of all the latest models as they come out. And actually I'm talking about it on the global conference tomorrow. So every time a new model drops, we run it through our eval. And so the training program for AI-Powered Accountant with Excel and Power BI, a lot of my actual use cases, client use cases were being applied to all the current models using Copilot, using Tab AI and using... Back then I was using Claude, but not in Excel because Claude in Excel hadn't come out yet. I was using Claude in the desktop and browser. So very different. Again, another change in harness engineering, which is again something else which has a big impact. So I noticed it straight away because the training program that I was trying to develop kept on giving me too much randomness.
(17:30):
And I just thought, "I'm not going to risk my reputation on this training course. It's just not going to work." I was literally at the end of my tether. I was literally at the end of... I'm like, "I'm going to give this up." And suddenly it started working that next week after that new model dropped and I'm like, "Oh, okay, that's different now." So I saw it pretty quickly, and then that's when I just accelerated the training program, the course, the work that I did as a fractional CFO. And Tab AI was the best at the time for me, which is why I sort of used and jumped on and worked with that tool and the team a lot because it was far superior to... And I know there were others, there was shortcut, there was Tracelote, there's a bunch of others, but there was a couple of things that really caught my eye with Tab AI.
(18:19):
So I think that's kind of how quickly I saw it because I was already in that mode of testing all the new models as they came out.
Host: Paul Barnhurst (18:25):
That makes sense if you're testing all them. So for our audience, who's this book written for? Who did you write the book for?
Guest: Lance Rubin (18:36):
I wrote the book for accountants, and I say accountants, which is a very, very broad, very broad church, and I'll probably say it's more probably accounting and finance people, but it all comes back to accounting. So it's called AI-Powered Accounting because accounting and finance people always come back to the books and records. And so it's anyone who's financial modeler, CFO, FP&A, financial controller, management accountant, in practice, out of practice, public practice, you name it. If you know what a journal is and know how to do a debit and a credit, this book's for you. Because basically everything that we do in accounting and finance about getting numbers right, closing the books, looking backwards, and then looking forwards is painful to do on a repeatable basis. And so this book was written for those people to want to level up on that work, but level up in a way that doesn't create more work.
(19:46):
Because if you don't use AI properly, it actually creates more human work rather than less. And so I felt that this was a gap in the market, which is everyone says, "Oh, prompting's everything, prompting, prompting, prompting." No, no, it's not. It's actually skills, it's harness, it's understanding the concepts, it's understanding the guardrails and putting in proper processes in place to learn how to do that. So it's not just prompting. Prompting is level one. I talk about five levels of AI-powered accounting capability. And so that's something which is not out there. And I figured, well, let me go write the book given the program is covered and focuses on that, and it's a very practical book. You can download all of these 10 modules, one to nine are actual practical modules on anonymized client workflows.
Host: Paul Barnhurst (20:42):
That is helpful. I'm totally with you. Early on when AI first came out, prompts mattered a lot more than they do today. When I do training, I tell people, frankly, tell AI what you want to do, what you're trying to accomplish, and let it write the prompt for you and then edit it.
Guest: Lance Rubin (20:57):
Yeah, exactly.
Host: Paul Barnhurst (20:58):
It will do a better job. Create a skill using a bridge or whatever methodology and have it write the prompt and then edit it and you'll get a better prompt than you can write yourself. It's like you said, we went from prompt to context. Now you're seeing the harness, which there's some overlap between those two, but that kind of continued shift. It's about so much more than just what you write in that simple prompt to get anything repeatable.
Guest: Lance Rubin (21:28):
Absolutely. I mean, for me, just to quickly run through it, those levels are critical when you're talking about prompts, level one, level two being skills, level three being MCP and connectors, and then you start to talk about level four, which are artifacts, and so you want persistent artifacts that they refresh and update. An artifact can also just be an Excel file as well. It doesn't have to be an HTML output. It can be just a markdown file. And then level five is digital workers, which I've got now. These are not agents. These are not autonomous. These are multi-skilled interdisciplinary orchestration of agents using a skill paradigm, which effectively is what we've built. And then you've got autonomous agents that hack into places and all that, and I don't believe in... I mean, I know they exist. I'm not a proponent of supporting them, and I think they're very dangerous because you're sort of suddenly handing over all your critical thinking to an autonomous agent.
(22:34):
And I think we need to have the human in the loop. You can get very far -
Host: Paul Barnhurst (22:37):
So you're not buying into the autonomous finance craze right now.
Guest: Lance Rubin (22:40):
I'm not, Paul. I think the autonomous finance craze is over-hyped. I think it's a real risk. It's one of the reasons why the Global Leadership Council said not one of them would trust an AI built model without review. Not one, not one single person. Why? Because it is the nature of the technology to be probabilistic and therefore have randomness in it. And no, you can't change the temperature to zero, which means there's no randomness. You can do it on an API, but you can't do it in Excel and some of that is proprietary and who knows. But yeah, I think it's an exciting journey, but it's just not going to end in a really good result for you if you're going 100% autonomous.
Host: Paul Barnhurst (23:28):
Anyone can claim to be able to build a model. Not everyone can defend one when the questions start. The advanced financial modeler puts your modeling under real scrutiny, so you know it holds up. I invite you to become an AFM. Go to http://www.fminstitute.com/podcast. That is fminstitute.com/podcast and use code podcast for 15% off.
I like to say on the autonomous, I'm like, "Find me any CFO that would trust a fully autonomous finance department." There's risk levels. I really like a framework Albert Lee shared on my FP&A podcast where he is like, "Think of it in green, yellow, and red. If it's low risk, if a mistake was made, there's no real financial, go ahead and run it autonomous if you want and if you can live with the risk and it's not a big deal, it's cheaper to just deal with it after the fact, run it." He's like, "Okay, if there's some cost or some risk, flag it as yellow and maybe it's a spot check.
(24:43):
It's a periodic review, not everything. But if it's red, if there's any high level of risk into it, as you know, do you let anything go through that's red in your system and just let it go to a customer?"
Guest: Lance Rubin (24:54):
No way.
Host: Paul Barnhurst (24:55):
And that doesn't mean AI didn't help you build it and save a ton of time. You just know better than turning it over to AI.
Guest: Lance Rubin (25:02):
Absolutely. And big strategic decisions around how much capital I need to raise is not a decision that I would pass entirely off to AI. I might get parts of that model to be built and reviewed, but ultimately I'm checking to make sure that it's right and I need to make sure that it's correct, and I need to design the scenarios that reflect reality and context. Context is everything, and that's the thing that people forget. AI actually doesn't know anything. The only thing it knows is the context you give it. And if you don't give it context, it's going to make it up. This wonderful word called hallucinations, which again, people are not familiar with that term. I think that's the critical thing. That's why skills are so critical when it comes to managing hallucination risk, because you give it pre-prescribed definitions of what it should do and what should it not do.
(25:58):
Now, even a skill file can sometimes miss what they call gotchas. And a gotcha is where a skill misfires and gives you the wrong answer resulting in that red alarm bell going off. Well, that's a problem. So how do we resolve that and how do we make sure that we've got all those guardrails and we've got all those gotchas? It's not just a, hi, just throw the spreadsheet to an AI agent and I'll sit on the beach and it'll all be fine because it's just not going to be. It's just not the nature of the technology.
Host: Paul Barnhurst (26:30):
It's why you see all these review agents and things. But I was going to say, when you said hallucinations, I was talking to somebody who, they have one of these modern AI tools they share. She's one of the co-founders, and she was talking to someone in finance, CFO, and she started talking about hallucinations and they weren't even aware that AI could hallucinate. I was just like how scary that is, the level of education for some. Others, it's much higher. It's all over the place right now. That's why books, training, the things you're doing are needed.
Guest: Lance Rubin (27:05):
Hallucination is the unknown. I mean, it's the thing that we don't know what AI doesn't know. And because we don't have that mechanistic interpolation, so we don't actually understand, we can't trace back all the nodes of the decision-making process on AI, we don't know how AI makes decisions through the neural network. No one has built those safeguards to say, "Oh, it did this, this, this, then that, then that, then that." We don't know. It's a black box. I like to translate all AI work into formulas, into deterministic logic that I check, then I can trust it. Otherwise, I can't.
Host: Paul Barnhurst (27:49):
I'm going to guess that's the reason you've leaned so much into Excel, into Power BI, in the Microsoft tech stack in your book and in your training is that provides transparency and the ability to validate things. Is that a fair statement?
Guest: Lance Rubin (28:05):
100%. The AI-Powered Accountant is made up of six pillars of skills and capability. Pillar one is Excel fluency. You have to be fluent at Excel. I'm not just saying I know what a SUMIF and a VLOOKUP is. No, no. You need to know what a Lambda is. You need to know what a dynamic array is. You need to know what a LET is, or you need to know that having a really complex LET is not good enough. I need to get that explained to me. And so you need to know what you don't know. And yeah, transparency is the only way we get that understanding of the decision-making process, which is why Excel is still important.
Host: Paul Barnhurst (28:45):
And what's your feeling? Do you feel like the newer generation, we're seeing them not wanting to learn Excel as much? Do you think AI is causing people to be like, "Ah, I don't need to really know Excel?" What's your take where we're at? Because there's a lot out there for sure.
Guest: Lance Rubin (29:01):
I think, again, it's like AI's going to take our jobs. There's a lot of this hype. Are we not going to need to know Excel? Well, how else on earth, and maybe I'll give Google a bit of a shout out here. I think how else are you going to not be able to see anything if you don't have some form of spreadsheeting technology? Back to my point is we can't see AI's decision-making process, but with the spreadsheet, we suddenly can. So we have to have that as a part and parcel of what people do. So when the next generation say, "Well, I'm just going to vibe code my way through this and I don't need to know how it makes its decisions," that's a problem. So they might like Google Sheets, but the reality is you speak to any investment bank or any large financial modeling shop, they're using Excel, and there's a reason for that because of a range of different things.
(30:00):
But the younger generation are believing all this misinformation, disinformation, manipulation around what it actually means. And that's, again, why I wrote the book, because quite clearly on the front cover, it says with Excel and Power BI, it's on the front page. It's not hidden. It's on the front page because I say absolutely we need transparency with AI. And the only way you're going to get that is by seeing it in a spreadsheet. And my spreadsheet of choice is Excel. Yeah. I think a lot of people are going to have some problems doing their job properly by giving up spreadsheeting skills or Excel skills.
Host: Paul Barnhurst (30:44):
And so how do we... I mean, advice to managers, people working in modeling as the new generation comes up and maybe they don't think they need to learn Excel. How do you manage that? Because I think management in general is going to have to change with AI. So I'd love your thoughts there.
Guest: Lance Rubin (31:00):
So look, I spoke to one of the leading partners in a big four firm in Australia about what he's doing with AI in their team. And he's basically saying AI is only given to the most experienced people. I'm not giving it to the junior staff. I want all the junior staff to learn. I learned the basics and the foundations because I don't want to breed laziness and I don't want to breed intellectual atrophy. I want people to be able to have that critical thinking, build the logic, build and do the hard yards because that's where the learning happens. And so take away the hard yards, then you lose that. So I think the good managers and the people that get that, that understand that will see that and will act accordingly. Sadly, I think a lot of people are believing the hype and executives are believing the hype, boards are believing the hype.
(31:55):
And you only have to have one scenario where they say, okay, we're banning Excel. We're just going full AI and native AI. We can have a native AI function in finance, no spreadsheets, just all AI. And then you walk up to the board meeting and it says, we're going to run out of cash. And what's that based on? Oh, well the AI told me. So that's where that story lands up where management get a rude awakening and a shock because they realize that doesn't work. We don't have that. So I think management have got to be really, really careful what they do. I think also they got to be careful at what the messaging is to people about AI. We're going to use AI because we want to reduce jobs. Well, that's not really a message that staff are going to take on well. However, what you should be saying is there's a lot of robotic boring work that you do today that we want to automate with you in the loop so you're going to learn the skill and use the technology because there's a whole backlog.
(32:59):
It's called the infinite backlog. It's the backlog of stuff that like my book that I still want to write on the Financial Models Guide to the Galaxy and those sort of things that I don't get around to doing. That's what you want to share with people, not it's going to take our jobs because they want to return on investment and I guess token costs are not cheap. So I think management are going to learn the hard way. They're going to make some bad decisions. Ford, the car company have a classic scenario where they got rid of all their engineers and just put AI fully on the production floor and suddenly they had all these defects in cars and they had to rehire all of the engineers because guess what? AI makes mistakes and no one's checking it. And then suddenly you've got a recall on a motor vehicle.
(33:56):
So yeah, management are going to
Host: Paul Barnhurst (33:57):
Make
Guest: Lance Rubin (33:57):
Stupid decisions. Smart people make stupid decisions. And I think AI is going to cause a lot of that. And I hear a lot of people, a lot of finance people taking an entire model and uploading it into the chat. I'm like, you've just breached every single... It's like kind of taking a spreadsheet and posting it on LinkedIn. Would you ever do that? Would you ever post your public information, your company information into a public domain? The answer is no. It's like, well, you don't have an NDA signed with Anthropic, so why are you doing it? Oh, well, there's no other way. I know there is. You can bring the AI to the workbook inside Excel and use it within the context window and actually change the name of the file and change and de-identify it so that it doesn't even know the client or the company.
(34:47):
And if you actually look at what goes across, quite often what goes across is formulas and logic and structure, not the whole model. So there's ways around this, but people just don't know how to manage it. And that's why, again, back to the book, this book is also written for managers to think about the frameworks of change, of growth mindset, of capability, of the human aspects of this which need to be invested in as well.
Host: Paul Barnhurst (35:15):
The advanced financial modeler is becoming the way modelers signal that the work is real. I took it, it made me better, and it opened doors. What are you waiting for? Go to fminstitute.com/podcast. That is http://www.fminstitute.com/podcast and use code podcast for 15% off.
Yeah, a lot of good stuff said there, makes a lot of sense. So I want to ask, I still think a lot of modelers are really trying to figure out how best to use AI. There are some that have leaned all the way in like you and others are using it quite a bit, but what would be your advice for those that are saying, "Look, I realize it's got good enough. It can do a lot of the work I'm doing. Obviously at the human in the loop." How would you advise they get started?
(36:16):
So if you're going to give some advice on that journey, if they're still really early, what's the advice?
Guest: Lance Rubin (36:21):
Commentary. Start with variance and commentary. It's low-hanging fruit. If commentary's wrong, no one's going to get fired, hopefully, depending on where the commentary's going. But I think just start to dabble and start to think about some of the basic things, get some of the drudgery of the work that you're doing. So I'd say start with that and then build skills. So if you're in Claude, build skills, I know you can use Claude, the Opus model and Sonnet inside Copilot as well as opposed to the Claude harness, but the skills environment is not yet mature in Copilot. I think it will become, I think it will evolve and it will become better. But irrespective, skill files are Markdown, they're human-readable, they're transportable. So start building skills in Claude and you can use the skill file inside Copilot or upload the skill file as part of a process.
(37:25):
So I think definitely start with a basic process that you know is not going to burn you, like commentary, and then start to build skills. So the first skill that I did was the commentary writing skill for a board pack because it's just like I spent so much time comparing this number to last month to decomposing it. But if
Host: Paul Barnhurst (37:49):
I've
Guest: Lance Rubin (37:50):
Got all the foundations right and all the structure and the data right in tables that I've pulled in and I've got dynamic arrays and I've got tables and I've got the structure and I've got the hygiene, we can talk about the hack framework, I've got the hygiene, then I can leverage the automation.
Host: Paul Barnhurst (38:05):
Makes a lot of sense. I think that's a good area to start because like you said, variance commentary is pretty much low risk.
Guest: Lance Rubin (38:12):
Yeah, exactly. Particularly actual versus budget, right? Yeah. Because you're comparing against something that was set 12 months ago anyway, so it's like, well, the world's moved on since then.
Host: Paul Barnhurst (38:25):
Exactly. As we're heading into Q4, for sure. January, maybe not as much, but we're going to have you answer rapid fire. You mentioned you're going to answer that from your fractional CFO versus modeling hat. So if anyone wants to see the differences, test them. They can go back and listen to an earlier episode. But before we do that, anything else you want to say about the book or this AI-Powered Accountant before we move on to rapid fire here?
Guest: Lance Rubin (38:54):
No, I think thank you to you, Paul, for I guess reaching out and talking about it and sharing thoughts and I guess also on the Maven course as well that obviously the book was based off. Yeah, I think it's really focused on the how to, the practical side. And I think that's what people want. I just don't think that we want another overhyped, overpromising the value of these agents and this autonomy, which I think is just dangerous. And then where do we sit? Where does the human sit in that process? No one's articulating that all of a sudden. That's what the book does as well, which talks about having that human in the loop and that review process.
Host: Paul Barnhurst (39:42):
There's so many talking, like you said, about the hype, the autonomous, but there's a few really good books out there. I know what you've done. I know Glenn Hopper, his new book comes out, I think next week or the week after on the AI ready CFO where he makes the argument that the CFO should be the one leading AI adoption, not IT.
Guest: Lance Rubin (40:03):
100% agree with that.
Host: Paul Barnhurst (40:04):
So I think it'll be a really good book. And he's already on his fourth book, which is going to be helping people build an agent, not necessarily autonomous, but help them understand how to build it because that means a lot of things to a lot of different people depending on how you define agent and agentic and all this stuff. Rapid fire incoming. 20 questions, yes or no answers, no hedging aloud. Circular references, VBA, the future of Excel, and more. Let's find out what our guest really thinks. Circular references.
Guest: Lance Rubin (40:40):
Absolutely not.
Host: Paul Barnhurst (40:41):
VBA?
Guest: Lance Rubin (40:43):
Yes.
Host: Paul Barnhurst (40:44):
LAMBDAs?
Guest: Lance Rubin (40:45):
Yes.
Host: Paul Barnhurst (40:47):
External workbook links?
Guest: Lance Rubin (40:49):
Absolutely not. Unequivocally, no.
Host: Paul Barnhurst (40:52):
Should modelers primarily use keyboard shortcuts?
Guest: Lance Rubin (40:56):
Yes.
Host: Paul Barnhurst (40:58):
Should models always be print ready?
Guest: Lance Rubin (41:02):
Yes, they should for the sections of the model that are going to be shared externally.
Host: Paul Barnhurst (41:07):
Okay. Are merge cells ever acceptable?
Guest: Lance Rubin (41:10):
They are. Controversially, they are. Only for certain situations for headings and things like that, but yes, they are allowable.
Host: Paul Barnhurst (41:20):
Frankly, I do agree in some situations, but my favorite answer is the one who said his saying he has is, "Use merge cells, go to hell."
Guest: Lance Rubin (41:29):
Yes, that's the one controversial one out there. That's very, very strong opinions.
Host: Paul Barnhurst (41:35):
Well, we got to get some. Should modelers learn Python in Excel?
Guest: Lance Rubin (41:38):
No. Why? Because I mean, AI is going to do all of the work. I just want to check the logic so I can understand. Python's a black box to me. It's not something that I'm going to check the Python code. I want to check the results and the formulas, so not -
Host: Paul Barnhurst (41:53):
Fair enough. Power Query.
Guest: Lance Rubin (41:54):
Absolutely, yes. As a fractional CFO, my go-to for ETL.
Host: Paul Barnhurst (41:59):
As an FP&A person, yeah, I love Power Query. So you and I are on the same page. Power BI.
Guest: Lance Rubin (42:05):
Yes. If I'm starting to work across multiple clients, multiple environments and multiple data sources and I'm connecting to a data warehouse, this is a data tool. Absolutely.
Host: Paul Barnhurst (42:16):
Should every financial modeler be able to build a fully integrated three-statement model?
Guest: Lance Rubin (42:21):
I think if you're going to call yourself a financial modeler, I think you must have the ability to at least have, not saying you're going to build it every time, but I think to be qualified or at least be calling yourself a financial modeler, absolutely you should have that ability. So yes, it's like saying, do I need to actually get on a bike to ride a bike or can I just sit on a tricycle? You can get by without it. It's just going to be a lot more painful, two-wheeler speed. It's different.
Host: Paul Barnhurst (42:56):
Alrighty. Okay. We're going to hit the rest of these real quick. Will Excel ever die?
Guest: Lance Rubin (43:02):
No, not in my lifetime.
Host: Paul Barnhurst (43:05):
What financial statement is most important for modelers?
Guest: Lance Rubin (43:08):
So as a fractional CFO -
Host: Paul Barnhurst (43:09):
And you're going to answer this as a fractional.
Guest: Lance Rubin (43:11):
Yeah. So I'm answering this as a fractional CFO, cash flow for sure, because I want to know how cash is going and where it's going. As the technical modeler, does the balance sheet balance and the cash flow is reliable? Then I'd say balance sheet, but no. As the CFO, I want to be -
Host: Paul Barnhurst (43:26):
To make sense. I already know the answer to this one, but your favorite LLM?
Guest: Lance Rubin (43:31):
Claude, without a doubt.
Host: Paul Barnhurst (43:33):
I was sure you were going to say Grok.
Guest: Lance Rubin (43:38):
Look, I've dabbled with Grok a little bit, but it's not up to scratch yet.
Host: Paul Barnhurst (43:42):
So you're now bringing it into Copilot.
Guest: Lance Rubin (43:44):
Oh, awesome. So I'll test it there. Yep.
Host: Paul Barnhurst (43:48):
It's already released for some. I checked today. I don't have it yet, but -
Guest: Lance Rubin (43:52):
Cool.
Host: Paul Barnhurst (43:52):
If you could pick only one for all of your models and you can only pick one, you go in sensitivity or scenario?
Guest: Lance Rubin (44:00):
Absolutely scenario analysis because it's multiple and it's more real world than just one sensitivity.
Host: Paul Barnhurst (44:07):
All right. Are financial models the number one corporate decision making tool?
Guest: Lance Rubin (44:11):
I'd argue it is. I think for proper strategic and operational decision making, there is no better tool and framework to do that, particularly if you're raising capital or you want to check actual to budget or you want to make real decisions. Sure, there's many other tools you can use to have sort of on the fly conversations and decisions, but I don't think that's the best. So it is by far the best.
Host: Paul Barnhurst (44:39):
What is your lookup function of choice?
Guest: Lance Rubin (44:42):
Back in the day it was INDEX/MATCH and I still prefer it because it's more efficient, but XLOOKUP probably because it's more dynamic array friendly. So I'd say XLOOKUP.
Host: Paul Barnhurst (44:53):
I thought you might say filter. I'm starting to get that more and more.
Guest: Lance Rubin (44:57):
Okay.
Host: Paul Barnhurst (44:58):
So what is your favorite Excel shortcut?
Guest: Lance Rubin (45:01):
I'm going to be a bit boring here. It's Ctrl+C, because really I'm just doing a lot of it. I'm copying and pasting into prompts, so I'm copying and pasting stuff that's not going away, particularly if I'm writing VBA code and I want to review it and I'm copying and pasting into the VBA editor. So I thought it would die, to be honest. Now I'm not copying and pasting to move data that I'm using Power Query for. So I think what I'm using it for is obviously very different. Control page up and Page Down and all the navigation shortcut keys are a close second. But I think if I think about every time I have to run scenarios or do something, copy just to the basic copy, I think it's definitely a lot gets used a lot.
Host: Paul Barnhurst (45:40):
Yeah. No, Ctrl+C, I think everybody uses that one a lot. Alrighty. Well, we ran through rapid fire there pretty quick. Any last parting word, how people can get in touch with you or anything you want to share? Floor is yours.
Guest: Lance Rubin (45:53):
LinkedIn, of course, Paul. And of course if you haven't registered, come for the Financial Modeling Global Conference. Obviously by the time this goes out, it'll already happen, but if not, go back and watch the FMI. And I guess the content that we're putting up as a Global Leadership Council is incredibly useful in today's times. So if you haven't read the report that came out around financial modeling and AI and some of the research that was done by the council, I think is valuable. Back to your earlier question about what management decisions are being made, management decisions need to be made in the context of what that report was referring to and understand that a lot of people are using AI, but very small percentage are actually getting value from it. And again, I think it's something which if you haven't seen the book or looked at it or the training program that we run, we solve that problem.
(46:44):
We solve that, okay, how do I actually translate it into real work to get real value? And I think that's the key because otherwise it's just fun when people don't want to just have fun. People want to actually get real work done.
Host: Paul Barnhurst (46:57):
I'm okay with just having fun. I'm just kidding.
Guest: Lance Rubin (47:01):
Well, yeah. I mean, if we all didn't have to work for money and it was all fine and we had universal basic income, then let's all have some fun. It's just not -
Host: Paul Barnhurst (47:08):
Oh, I hear you. Yeah. If we could all have that, that will be the day. All right.
Guest: Lance Rubin (47:12):
Universal high income, as Elon likes to call it.
Host: Paul Barnhurst (47:14):
Yeah, I would like the H in it, not the B.
Guest: Lance Rubin (47:17):
Yeah.
Host: Paul Barnhurst (47:19):
Alrighty. Well, thank you so much for carving out some time. Always fun to chat with you, Lance. I'll let you go so you can enjoy your day. I know you're at the beginning. I'm at the end of mine, but thanks for coming on the show.
Guest: Lance Rubin (47:29):
Thanks, Paul. Thanks for having me again. And yeah, it's wonderful. I think it's a wild ride. I think it's great to have people like yourself getting the message out there and just having a balanced discussion around this topic because it's so polarizing around what AI is going to do and there is value, but I think we have to make sure that we are prepared for it because it's not just a plugin and she'll be all right mate. It's not. It needs real thought leadership and you're offering that, so thank you.
Host: Paul Barnhurst (47:54):
Well, thank you. Appreciate that. You're doing great work too with your book and your course, and so good luck on growing that and thanks for coming on. Financial Modeler's Corner was brought to you by the Financial Modeling Institute. This year I completed the Advanced Financial Modeler Certification, and it made me a better financial modeler. What are you waiting for? Visit FMI at www.fminstitute.com/podcast and use code podcast to save 15% when you enrol in one of the accreditations today.