AI Can Build the Model, But Can You Trust It? with Ryan Humphrey
In this episode of Financial Modeler’s Corner, Paul Barnhurst speaks with Ryan Humphrey about financial model auditing, the skills strong modelers need, and how AI is changing the profession. Ryan explains why communication, judgment, and trust matter as much as technical ability. He also shares how model audits work, why poorly structured models create serious risks, and why finance professionals must continue developing their skills as AI takes on more of the building process.
Ryan Humphrey is a Partner at Forvis Mazars in London, specializing in accounting, tax reviews, and financial model audits. A qualified chartered accountant, he co-authored Credible Model Audit for the AI Era and is currently researching sustainability leadership with Cambridge University.
Expect to Learn:
Why communication is essential in financial modeling
How independent model audits support better decisions
Why human judgement remains critical in the AI era
How AI may shift modelers from building to reviewing
Why trust and continuous learning are becoming more valuable
Here are a few relevant quotes from the episode:
"The model really is a communication vector more than anything. It is about telling a story." - Ryan Humphrey
"Trust becomes so important because things can look right very quickly, even when they are not." - Ryan Humphrey
AI may make financial models faster to build, but it also makes careful review more important. The future belongs to professionals who can combine technical knowledge with judgement, communication, and trusted decision-making.
Follow Ryan:
LinkedIn: https://www.linkedin.com/in/rchumphrey/
Website: https://www.forvismazars.com/uk/en/users/our-team/ryan-humphrey
Follow Financial Modeler's Corner:
LinkedIn: https://www.linkedin.com/company/financial-modeler-s-corner/
Disclosure: Portions of this episode (such as the introduction or promotional segments) use AI-generated voice narration produced under human editorial review.
Newsletter - Subscribe on LinkedIn: https://www.linkedin.com/build-relation/newsletter-follow?entityUrn=7079020077076905984Sign up for the Advanced Financial Modeler Accreditation today and get 15% off using the code “Podcast” at https://bit.ly/497oAqW
In Today's Episode:
[00:00] -Trailer
[05:16] - Why Communication Matters
[06:45] - AI and the Future of Modeling Careers
[08:50] - Ryan’s Model Audit Nightmare
[12:26] - How Financial Model Audits Work
[17:36] - Essential Skills for Model Auditors
[21:35] - How AI Is Changing Model Audits
[27:54] - The Growing Importance of Trust
[32:34] - Why Continuous Learning Matters
[37:30] - Excel Preferences and Shortcuts
[41:30] - Rapid-Fire Modeling Questions
[47:35] - Final Advice for Financial Modelers
Full Show Transcript:
Paul Barnhurst (00:42):
Financial Modelers Corner is the world's premier modeling podcast. It is brought to you by the Financial Modeling Institute, the world's leading financial modeling accreditation organisation. Welcome to Financial Modelers Corner. I'm your host, Paul Barnhurst, the FP&A guy. And this podcast is all about the art and science of financial modeling with distinguished modelers from around the globe- not those kinds of modelers, financial modelers. The Financial Modelers Corner Podcast is brought to you by the Financial Modeling Institute. FMI offers the most respected accreditation in financial modeling. And I would argue it's more important than ever in the era of AI. Fundamentals, foundations are more, not less important. I completed my advanced financial modeller because it's a great foundation. All right, onto the show. This week I'm thrilled to welcome Ryan Humphrey. Ryan, welcome to the show.
Ryan Humphrey (01:41):
Thank you so much for having me. Great to be here.
Paul Barnhurst (01:43):
Excited to have you. So now Ryan, you got to tell your audience, what do you listen to when you're running? This is going to be an ad for me.
Ryan Humphrey (01:50):
As if we just spoke about that, Paul, my new running podcast has been all about Parish Modern Corner and everything that you put out there. I absolutely love it. I was on the YouTube channel before. I've been a fan for a while, in particular of the Mod Squad, and I'm disappointed that I didn't get an invite to a Mod Squad video, but maybe this is my plug to be included next time.
Paul Barnhurst (02:08):
We will have to mention that to Giles and Ian because we do a monthly episode, so we may have to bring you on there as well. So I love the plug and thank you for giving me that commercial. I appreciate it. All right, so a little bit about Ryan's background before we jump into our interview. He's a partner at Forvis Mazers in London. He's an expert authority in modeling, in particular specializing in accounting and tax reviews on financial models. We got to have people that love it, so I'm happy. Ryan acts as a signing partner on financial close, operational and refinancing model audit engagements as well as more atypical model due diligence engagements. He originally trained as a chartered accountant. He became fully qualified in 2017. He was a named contributor of the paper, How to Review a Spreadsheet that was published by the Institute of Chartered Accountants in England and Wales and has more recently co-authored the white paper Credible Model Audit for the AI Era, which is something we're going to talk about today.
(03:10):
He is also currently carrying out research with Cambridge University in the field of sustainability leadership and is a passionate advocate of the energy transition. I love it. So thank you for a little bit of your background. Before we get to... Actually, I'm going to start with the career corner, then I'll ask my normal first question. So this is a new section. You're only the third person I've done this with. We're putting whether a little section is aimed toward people early in their career to give them a little bit of advice. So how did you get into financial model audit in particular? Was that intentional or did it just kind of happen?
Ryan Humphrey (03:44):
I don't think anybody who's kind of landed in this was when they were at primary school or elementary school, his name was thinking, and I could be a footballer or I could be a model auditor. I think everyone's kind of found their way into it. From my perspective, I followed quite a typical path. I trained as an accountant originally in a sort of more typical field of statutory audit. And what happened was kind of strange actually. I'd qualified and I was speaking to people in the organisation that I work at and I heard about this team called Project Finance at the time and I heard they specialised in renewable energy projects, which at the time was perhaps not as popular as now, and they also loved Excel. And in my team that I previously worked in, I'd heard about these people and they were told to me as being kind of a bunch of nerds and that was the words that were used.
(04:25):
And I think that was to try and put me off joining them. I'd sort of had a few conversations. What I didn't know is that I'm kind of a bit nerdy and a little bit... I have a kind of hippie tendency as well. So I was thinking, okay, this sounds really interesting. I love renewable energy. I love the green transition, and I've been doing that for the best part of the last decade now in that team.
Paul Barnhurst (04:41):
There you go. Funny how it works. So my last interview, which I think actually will come out after this one, but I recorded one right before we were talking, Dim did refer to it. He goes, "I thought finance people were just a bunch of nerds that sat in the corner before he learned about financial modeling." So you use nerd and hippie. It seems to be a trend. Nerd
Ryan Humphrey (05:00):
Come up. Reclaim it. Yeah, reclaim it, definitely. Yeah.
Paul Barnhurst (05:04):
That's what I say. I own it. Although I told somebody I was a nerd and they're like, "No, you're a dork."
Ryan Humphrey (05:09):
Yeah, actually. Understand
Paul Barnhurst (05:10):
The difference.
Ryan Humphrey (05:11):
Where do geeks come into that? Yeah, I've never really understood. But anyway, I speak about all of it in an affectionate way.
Paul Barnhurst (05:16):
Same here. All right. So what is one skill that you have developed as a modeller that most people would be surprised by?
Ryan Humphrey (05:24):
From my perspective, one of the most important skills is actually communication. When we probably talk about technical skills later, I really feel like interpersonal communication skills are really what's critical. So understanding what the client wants. I actually review models more than build. In that capacity, I'm always working with a third party to really understand how we can help. When it comes to modeling, I know you speak a lot about scoping and specification in your podcast and I completely agree. So clear communication skills are actually essential. And again, in this field of coming back to the nerds and hippies, the offhand comment, I think the best modelers are people who are actually able to really communicate and understand clients and that's something I've really noticed.
Paul Barnhurst (06:03):
It's amazing how often communication comes up. I think that's probably the one that I agree with. People realise the least is that no modelers really have to communicate. It's not just about technical skills.
Ryan Humphrey (06:14):
I mean, the model really is, and like I said, it's basically a communication vector more than anything. It's about telling a story that's fundamentally what model's meant to do. And I think that the modeller who's developing that model has to understand what the story is. So again, that's communication.
Paul Barnhurst (06:29):
All right. So one more question here in the career corner and then we'll kind of move on. What would you say to people I'm sure we'll have some, especially early in their career, college students listen, who are worried AI will make the financial modeling job profession obsolete?
Ryan Humphrey (06:45):
A very big question and it's a question I've heard a lot of people ask me. In my capacity, we have about a hundred people in the team that I run at Fort Mazars. And to be honest with you, it's difficult to map what that career looks like. And if anyone tells you that they know what the financial model is going to be doing in 20 years, I will call them out right now. They are lying 100%.
Paul Barnhurst (07:05):
So the BS meter goes off is what you're saying.
Ryan Humphrey (07:08):
100%. The radar's pinging in my ear when I hear that. But I do think that there is still a role because ultimately what cannot be replaced is judgement . And ultimately the coding skills may become less valuable with time because ultimately AI is already very good at that in my view. But the ability to understand and apply commercial judgement to a model is something that's becoming more important. And what we're already seeing is that I think the role of actually the more junior financial models is becoming more interesting, to be honest. I think there's actually the opportunity to be having more value adding conversations earlier on in the process because as the tools develop and you have to develop different skills earlier on. And again, I'm not sure if that's something you've heard in your many conversations, but I'm yet to believe that that layer of judgement is going to be replaced anytime soon.
Paul Barnhurst (08:04):
That's probably the most common thing I hear is people are, look, the human judgement isn't going to go away anytime soon. And as I like to say, if we had AGI, all bets are off. You could find me on the beach.
Ryan Humphrey (08:15):
Oh, definitely. I'll be there with you as will everyone. Again, I think it's something everyone needs to park to one side.
Paul Barnhurst (08:24):
And I agree with you. Whether we get there, it's a whole different world if we get there. If, when, however you want to look at it. Until then, let's talk about what seems realistic. So I think you and I are on the same boat there and we've had that conversation Mod Squad, and I always kind of joke about the AGI thing. All right. So now we're going to get to the question I usually start every episode with. Since you already know what it is, tell me that nightmare. I know you have one.
Ryan Humphrey (08:50):
I have prepared this one. I had to think about it quite a lot. I mean, one of the things that ultimately I'm re-reviewing other people's financial models rather than building them myself. So I can happily comment on lots of awful models I've seen over the years. But I think the one story that sticks with me is it was actually very early in my career and I was reviewing a model that was sent to me and it was built vertically. The timeline, instead of being horizontal, was completely vertical. I know that some people like to model that way, especially in a non-sort of professional financial modeling environment, but this was very unusual to me. I was relatively on duty. Maybe I've been doing it for a year or two. So when I opened the mode, I though, okay, listen - Was this
Paul Barnhurst (09:28):
By month? Because I could see years, just maybe a couple years.
Ryan Humphrey (09:33):
There was a monthly timeline going. Yeah, exactly. We're talking monthly, quarterly, semi - Oh,
Paul Barnhurst (09:37):
That'd be
Ryan Humphrey (09:38):
Rough. Everything was vertical. And at the time I was thinking, this is unusual. I haven't seen it before, but I'm relatively new in this field as well. So okay, I'll go with it. We spent a lot of time understanding this model. The first thing we did with the client was just asked why they'd done it. They didn't really have a good answer. I think they had a relatively new, let's say, modeling team that built this model. And one of the first things we said is, "Are you sure? Are you sure this is something that banks are going to be able to rely on? " And we didn't hear from them. We gave them our findings. We tend to work iteratively, so we'll review a financial model, give them a list of findings, and they'll come back to us with an updated model. That's how the process works.
(10:19):
In this case, we heard nothing from them, and a month later we came back with a horizontal model. So we had to do all of our work again from scratch. The model was obviously in a better place and somebody had had a conversation, but for me, that was probably my nightmare story just because it took so long. It took us a huge amount of time to effectively re-perform all of the calculations.
(10:40):
Some financial model reviewers will review a model in isolation sell by sell. So they'll go down row by row. They'll look into every formula. The way we do it is we effectively re-compute all the calculations. So we rebuild a model entirely and it's taken us a long time to effectively transpose that logic into a way that we do things. So we've spent all this time building this kind of very fancy transposition for it ultimately to be completely pointless because they went and rebuilt the whole model and didn't tell us. The main thing is they didn't tell us. They just told us the time.
Paul Barnhurst (11:11):
So what was that, Sky? So there's important communication. Was
Ryan Humphrey (11:14):
That one? Something about communication would've been greatly appreciated to my former self when I was reviewing that horrible model. And maybe I'll just say one other thing as well. I guess the theme of the band was really a one-off. And now we do occasionally see vertical models that are annual, like you said, which is not that unusual. But I think the theme of most ugly models is the whole Frankenstein model. Models that have been used by many people over many years. Again, these are normal things that I'm sure you hear a lot of that haven't been taken care of. They didn't have a nice guide, so no one really knew how to use them. They've been passed down generations of finance teams. These obviously - Have you
Paul Barnhurst (11:57):
Been peeking at my FP&A models over my career? Very common.
Ryan Humphrey (12:04):
Oh, yes.
Paul Barnhurst (12:05):
All right. So I know you're relied on by some of the largest banks organisations in the world to audit financial models. You've worked on hundreds of deals. So maybe take us through the process. What is a typical, I know there's not necessarily a typical, but if you had to say average or outline, what's that normal process look like for a model audit for you?
Ryan Humphrey (12:26):
Good question. So most model audits that we do are in the energy and infrastructure space. This is just because we're talking long-term cash flows, potentially contracted revenues to some degree, and effectively that need to build out for the long-term and get certainty over them. I think typically they'll be in terms of deal size about half a billion pounds. So about 500 million US dollars if you want to talk about putting a number on that. And the process effectively is that we will receive a financial model. It'll take us a couple of weeks to run proprietary tools and processes over that. And then we'll go back and forth with the client. I spoke previously about that iterative workflow before ultimately landing on what's the closed model, which is what the deal is underwritten on. And ultimately we sign off on the logic of that model. So if there was an error in that model, if there was something going on, ultimately we would be responsible for that potentially.
(13:24):
And that's something obviously we are very keen to avoid, but that's sort of how it works. And it kind of makes sense because you look at the quantums of the deals we're talking about, if you're a bank and you're going to underwrite a transaction on somebody else's financial model and put out half a billion pounds without that assurance of this model works, I think that would be a pretty foolhardy investment for a funder.
Paul Barnhurst (13:47):
They need some level of comfort. So when you're signing off on a model, what are you validating? I mean, obviously you're not validating that the 102 million of NPV they expect over the next 20 years is right.
Ryan Humphrey (14:01):
We are validating that the NPV calculation works. And I know that you talk a lot about X MPVs and MPVs at times as well, but it's not just about that. What we stay clear of is also validating the assumptions. So it is really about the logic.Fundamentally that's what we're looking to do. Decision makers do not have time to troll through a model which can have tens of and at times hundreds of sheets to understand everything that's going through that. So we're really trying to eliminate the need for that to be done by somebody else. And ultimately we are saying that the different constituent parts of them will all come together so that the outputs can be relied upon. That's really important. It's really output driven. So that might be in the case of a debt raise that your debt is going to get repaid. Ultimately, if we're being brought in by a bank to make sure that the debt is repaid, we don't need to test a thousand different sensitivities or scenarios.
(14:59):
That's not really our job. We're just making sure that according to the base case, the debt is repaid, but that's effectively our role.
Paul Barnhurst (15:07):
Yeah. So you're basically going through the model, making sure the logic is sound,
Ryan Humphrey (15:13):
The
Paul Barnhurst (15:13):
Assumptions are supportable. Correct.
Ryan Humphrey (15:16):
The assumptions we validate to documentation as well. So for example, with debt transactions, we would look to the financing agreements, the credit agreements. Correct. You're
Paul Barnhurst (15:25):
Basically saying, look, this is a sound model, the logic, the assumptions, everything is supportable. You're not saying, "Hey, we guarantee that the 102 million is where you're going to end up." Well that's a -
Ryan Humphrey (15:37):
No, actually that is really important. I mean, ultimately we are independent. So sometimes we refer to our work as an independent model auditor. Some people who review a model might be, I would say, promoting the fact that they're very collaborative, the fact they're very efficient. We try to avoid those terms actually. It's a bit counterintuitive having said communication is the most important skill. But for us, it's about professionalism, independence, because you want to make sure in these transactions that everyone is satisfied that the model works. You can't take sides and that independence point is really critical for us. It's something that differentiates us from, I would say, other people who can review financial models. Yeah,
Paul Barnhurst (16:19):
You're basically saying, look, we're an independent review. We're not favouring the bank. We're not favouring the lender. Yeah,
Ryan Humphrey (16:24):
Or by an online seller in an M&A deal. We will ultimately offer a loan to all parties on the transaction. Works for M&A deals as well. We can do that.
Paul Barnhurst (16:34):
Well, my background is in FP&A. I am also passionate about financial modeling. Like many financial modelers, I was self-taught. Then I discovered the Financial modeling Institute, the organisation that offers the Advanced Financial Modeller Programme. I am a proud holder of the AFM. Preparing for the AFM exam made me a better modeller. If you want to improve your modeling skills, I recommend the AFM programme. Podcast listeners save 15% on the AFM programme. Just use a code podcast. Sure. Yeah, yeah. Yeah. We're not in favour of the buyer or seller. Here's our opinion. Do what you will with it. Which makes total sense. I think that makes a lot of sense in modeling. So what would you say are the most important skills for model auditors? Because I think there are some differences between those that are just building models. I don't want to say just, but those who build models and those who audit.
(17:36):
So what would you say are the most important skills for the auditor?
Ryan Humphrey (17:40):
I mean, it probably depends on how you order to model. I think the skills are closer than you might imagine, and there's convergence because the way that we review a model is we actually build a model. We build all sort of stuff. If
Paul Barnhurst (17:51):
You're building a full model to review it, then they're going to get a lot closer in the way you do it.
Ryan Humphrey (17:56):
Exactly. The skills are more similar. I think it changes though as you go up the seniority. So ultimately more people will look at and review a financial model than build a model. By that, what I mean is a single user who's creating a model is generally building for somebody else. It might be for your own self-validation or interest, but typically in a corporate setting, there'll be many other people who look at it. And I think that a key skill for more senior people, and this is not like a professional model auditor, but anyone reviewing a model is being able to quickly distill what are the critical outputs and what is this model trying to do? It is with alarming regularity that I will open a model and I've been doing this for a very long time and I'm not really clear what the purpose of the model is because it isn't signposted.
(18:41):
The outputs aren't clear. There's a lot happening. One of the things that is a real bugbear is a financial model trying to do everything. A good financial model, in my opinion, should really be doing a single thing. If you want to do multiple things or build multiple models. Otherwise you have a confusing story. It's difficult to understand what's happening. So that ability to apply human judgement , it becomes really critical for financial model reviewers in particular in my view. I don't want to pick sides here. I've got a lot of very competent modeling colleagues and model auditors.
Paul Barnhurst (19:14):
Go ahead. You can make fun of all the modelers.
Ryan Humphrey (19:16):
We are a little bit competitive. That's okay.
Paul Barnhurst (19:20):
I would've never guessed. Speaking of competition, do you compete in the modeling world championship stuff?
Ryan Humphrey (19:27):
I haven't, but a lot of people have been trying to convince me, Jiles and Faye in particular in the UK. And I do really promote it as an excellent activity. I've followed on YouTube, and I think that some of the live battles of the Global Excel Summit were also really impressive. I think that the eSports world is really important because it has opened up Excel to a new audience, which for me is really important. I'm really passionate about this. But personally, I'm terrified. Yeah, that's a good question. Why? Let's
Paul Barnhurst (20:05):
Have a therapy session. Ready? Go. No, I'm kidding.
Ryan Humphrey (20:09):
Do I seem like I need therapy? No, I think the reality is that I think there's a lot of difficult skills to learn, and I think that I'm applying my own learning time and other things at the minute, but I wouldn't rule myself out in the future, definitely. And I've encouraged a lot of my junior colleagues to have a look at that and become involved, certainly.
Paul Barnhurst (20:34):
I'm actually completely with you on your reasoning. I would like to do it more, but it's just a matter of time because I see the ones, if you're really going to get into it, it's a time investment. Now, some people maybe just do the cases, totally have fun, but I know me and I know I'll want to get better and I know there's a real time investment. The level of stuff they're writing and the level of stuff I'm writing right now are just different levels.
Ryan Humphrey (20:57):
Yeah, absolutely. It is a big time commitment, but I think the skills you develop are really worthwhile. I don't know if we'll come to the importance of skills now in the future. I'm sure we might be, but I do think it is time for people to be doubling down on that investment in skill development, whatever that means to somebody. Yeah, absolutely.
Paul Barnhurst (21:19):
So speaking, this is a little bit on skill development, but we'll start here and we'll get more to that. How do you see, first we'll start with auto models changing with the use of AI. Let's start there, then we'll get into skills because people are going to use AI to build models.
Ryan Humphrey (21:35):
Yeah, I think that is a really big question in my field. It's a big question in financial modeling.
Paul Barnhurst (21:43):
Yeah, a lot of people are trying to figure it out. Of
Ryan Humphrey (21:46):
Course. I mean one of the critical distinctions, which is well understood and well discussed, is just the difference between probabilistic systems and deterministic systems. And that's really important. I mean, what we have done over the last 20 years in auditing financial models is that we have actually a solid proof of what we've done. We can point to the shadow model that we've developed. So you have a client model, we have the shadow model, which is our own intellectual property. It's our version of their model based on our knowledge. One of the concerns about using AI without a deterministic process is that you don't really know how it's got there. And if it isn't targeted, then that is a major concern. That said, I am personally very bullish on what AI can do. And I have really observed in the last three months in particular, a major step change in spreadsheet understanding from AI, which was not there a year ago.
(22:49):
That does change the opportunities and possibilities in the model audit field. And I talk about this all day, so probably stop me there. But I don't know if there's anything specific you want to pick out from that.
Paul Barnhurst (23:01):
No, I think that was good. I agree with you. First, I think you mentioned the whole probabilistic versus deterministic. I think second, we've all seen it. I think most of us when Claude four seven came out or four six, I think it was four six, now we're on 4.8. There was a leap forward in what the models could do. I mean, anyone who watched ModSquad a year ago when we first launched, or not quite even a year ago, October of last year and saw some of those first models we went through, they were pretty bad. I mean, there are some of them that if a junior analyst did that, I would be concerned. And now it's like you can look at some of them and go, "Okay, that's better than as good as a lot of people starting. I mean, some of the structure is better than a lot of models I built early on in FP&A." And that comes with excitement and nervousness.
(23:53):
I've talked with so many people, the whole idea of I can build a model in five minutes. You can, but is that really the important thing?
Ryan Humphrey (24:00):
I mean, the real concern is what you just mentioned, the fact that it can seem good. That's the real concern is something now that's produced is so plausible in how it looks. It's become so strong at detecting patterns effectively that it can produce something that looks really credible. But ultimately there can be errors in there. There are errors in there. I have never, again, I don't know if you have seen this, but I'm yet to have received a model built by AI yet. I am fully built by AI. I'm sure people are using AI to understand concepts to help with complicated formulas for both pausing formulas, understanding what's going on. But I'm yet to see someone actually say, "I've built this entirely using, I don't know, agentic AI solutions." I would love to see that.
Paul Barnhurst (24:47):
Bad financial models can lead to bad decisions or worse. So how do you minimize the risk of a bad model? You make sure the models you build are great. The Financial modeling Institute developed the Advanced Financial Modeller Accreditation Programme to help modelers like you. The AFM programme offers a step-by-step approach to building world-class financial models. The programme ensures that you know the best practises in model design and structure and will help you brush up on your Excel and accounting skills too. Be the one on your team to build great models. If you want to impress your boss and your clients, get AFM accredited. Podcast listeners save 15% on the AFM programme. Just use code podcast@fminstitute.com/podcast. So I know someone who got a model, but it wasn't a financial model. Well, I had financial components to it. We'll call it a financial model, but it wasn't like a three statement or what you think of as a typical.
(26:02):
It was an HR or it was a resource model around hiring. Someone who worked in finance, their business leader goes, "I use Claude and I put it together." And the person shared, "And then I spent three hours fixing it all. " That was their description. And he was so proud of it. They're like, "They're so proud that they were able to build their own model and didn't need my help." I was like, "I wish you would've come to me. "
Ryan Humphrey (26:24):
Yeah, absolutely. I mean, it's interesting you say that because that does illustrate an interesting trend, which could be in the future that financial modelers are moving more towards from build to audit. And it could be that financial model, you need the skills of how to build a model. But as you said, as the coding from the AI side becomes cheaper, you need to spend more time refining, fixing, reviewing models. And that could definitely become a key trend. So maybe I've much more by luck than judgement ended up on the right side of the model versus model audit line. But I think they converge anyway. There's a lot of convergence there.
Paul Barnhurst (27:04):
There definitely is some convergence. I mean, regardless, any role, you're doing FP&A, you're doing investment banking, you're doing whatever it may be. If you're doing a lot with modeling, at some point you're reviewing modeling and sometimes some point you're reviewing somebody else's. You may not be formally doing model audit, but at a minimum, review and build converge to a certain degree. Do I expect in the future review to become even more important? Yes. As less and less of the build is done by a human, it's more and more important that we validate all that. Human judgement I believe will become more important I think as we talk about skills. And I imagine you would probably agree, but what do you see as the most important skills for this profession? And maybe how do you see it changing? Just your thoughts in general as we see more and more AI.
(27:54):
I
Ryan Humphrey (27:54):
Mean, it's a really, really big topic of discussion. I do think that soft skills become more important. I've said this earlier, that interpersonal skills become more important than they perhaps once were. Once were, sorry. You might have in the past, you request a model from a model builder, they send you a model and away you go and it becomes very black boxy how they get there. It doesn't really matter. You get a product. I think now there is much more need to develop a deeper relationship between the modeller and the user, which is a good thing as far as I'm concerned. And I think you could argue that certain technical skills become less important. Now that doesn't mean that they're not important because they are absolutely critical, but that the types of technical skills become different. I mean, I've spent a lot of time in the last year looking at Python more than I've ever spent in my entire life before the last year.
(28:48):
So I think that there's a sort of need to continue to upskill, but people need to think about are they using the most up-to-date technical resources? Are they using trusted resources to upskill? And I think that we haven't really talked about it, but what I really value is the trust economy. We're moving away from a product and services based economy to a trust economy where people will effectively pay for things they trust because the actual unit cost of producing something in general has reduced. And I know that you've had Ian Bennet on a recent podcast talking a lot about this and I completely echo all of the sentiment. I mean, we're both coming from a professional services environment, so we have probably quite a similar angle on this conversation. But certainly the value of trust becomes so important.
Paul Barnhurst (29:40):
Yeah. I mean, trust should have always been important, but it comes even more. I think I posted on LinkedIn the other day, and not talking about modeling, I was saying content creation, creation of content has basically become free. Development has become nearly frictionless, not free. You still got the tokens, but even with content you do, but content doesn't use many Tokens , developing code does or developing a model can use a lot. And so development in general has become virtually free and frictionless.
Ryan Humphrey (30:11):
Yeah, absolutely. And also that point about content, we could talk a lot about content on LinkedIn, but it does seem to have just the quality seems to have deteriorated, at least the quality of conversation that I see and understanding who are authentic voices that I can trust and what has become a quagmire of frankly rubbish. I was going to use a different word, but I'll keep things professional. It's important to know who you can trust. And I think that actually when I think about the FMI, what they're doing, and I know they sponsor this podcast is really important. It's actually about building those trusted voices in the community so people don't have to question what to believe in, if you like.
Paul Barnhurst (30:51):
Yeah, trust becomes more important than ever as we're moving. It's funny, I'm sure you've heard of Y Combinator, probably heard of that. And they used to be, if you came to them with the services business, it's like you weren't getting it. Go away. We just want software. This last round, they basically said you need a service component to your AI business because where's the trillion dollar opportunity in the services field? And as agents get better and better, what are they starting to do? Build a lot of those workflows that were performed by people, services. And so it's a whole new world in a lot of ways.
Ryan Humphrey (31:28):
It definitely is.
Paul Barnhurst (31:29):
It's exciting and frankly a little scary, but 100% agree. Trust has never been more important to find voices you can trust. I can put rubbish that looks great on LinkedIn. I could put rubbish in Excel that looks great. I could put rubbish on YouTube. I can put rubbish in software. You can put it pretty much anywhere really quick and it can look right. The hard part is unless you really know that area well, it's very easy to believe it's not rubbish. So that's why I like people to comment on my post to get a free prompt pack. I'm just kidding. So I think I know the answer to this question because we've talked about it quite a bit in different ways, but you mentioned what skills become more important as we go more and more to AI, but as you become more senior in your career, what are those skills that become more important?
(32:27):
I imagine communication. Is there something else? Because we've covered communication pretty well. Is there another skill in there that you would say becomes more important?
Ryan Humphrey (32:34):
That's a really good question. I think what comes to my mind is sort of abstracted away from the financial modeling world just for a second. I'm very fortunate to speak to very, very intelligent, high achieving people, top CFOs, top CEOs. One of the privileges of my job is working with really clever clients. And one thing I've noticed from leaders in general, this is a very personal observation. So don't hate me if you don't agree, but this is my view that the leaders who are at the very top have very few weaknesses and they continually invest in their development areas, almost obsessive in their keenness to never stand still. And I think that becomes a really crucial element to senior people, let's say medium grade seniority people, the highest C-suite as well. I think now is a time where everything is changing and that requirement to continue investing in yourself becomes more and more pronounced or you'll be left behind 100%.
(33:37):
And that's what I said earlier about looking into Python. I'm not a code, I'm not a computer programmer, but you need to work on your development areas. You need to really seek to identify them and then you need to really invest in that as well. And I think that for me is often what differentiates somebody who maybe has a career that's sort of stagnated, still a good career, but maybe they haven't continued to progress further. It's because they haven't taken the time or perhaps invested the money required in that self-improvement and that self-development. And maybe they don't know what to do. Quite often I do quite a lot of coaching and mentoring as well. And a lot of the time people don't really understand what their weaknesses and strengths are. And it's not for me to tell you, it's for you to... Often you know the answer if you think about it a bit.
(34:19):
So I think I sort of abstracted that completely away from financial modeling, but it's also true in the context of financial modeling and people in finance in general. At least that's my view. No
Paul Barnhurst (34:29):
I love the answer and I love just sharing CFOs and these top people you speak to. An example I'll share of I think really hits this home of development. So I interviewed Gary Ridge. He wrote a book called Any Dumb Ass Can Do It. He was rated as one of the top 10 CEOs of the 21st century. He ran WD40 and had over a 90% approval rating over his 20-something years as CEO. So you know the surveys, you're thrilled when you get in the 80s for the employee surveys. He averaged, I think it was 92 or 93% over his 20-something years from his employees. So tell you he was loved and the stock price increased like 12,000%. Revenue went up. It wasn't just that he was loved, he performed. And when he became CEO, the first thing he did is he goes, "I don't know how to be a leader." He signed up for a master's degree in the evening on leadership.
(35:28):
And Ken Blanchard was the one who taught it. Anyone knows him, he's a big one of the foremost leadership people in the world in San Diego at the local university where headquarters were. I was like, okay, that speaks to me being a lifelong learner and I will never stop developing myself.
Ryan Humphrey (35:45):
I totally agree with that. I mean, I feel embarrassed to compare myself to someone of such a high esteem, but you mentioned in my intro about researching with Cambridge University. It is true. I'm studying at Cambridge University. That's a course I only took up about a year and a half ago. And part of it was just to effectively broaden leadership skills. I was challenged by and appropriately cited by people that I trust very much that basically said, "We are entering a new world. You need to effectively broaden your horizons away from just the details of the Excel." I think that's really valuable advice. And I think that parallel you've drawn to senior leaders is really good. I completely agree with that.
Paul Barnhurst (36:23):
And I'll share one with you when you mention the development, because I think you'll like it to broaden your skills. So I interviewed Jorge Rojas, which will come out before this episode. I think it'll be the one before yours is released, one or two. And we talked about using satellite data to improve financial models. I can't say I ever thought about satellite data. He just published a research paper looking at all the research that's been done around earth apps observation and financial modeling.
Ryan Humphrey (36:52):
That sounds very impressive. And I'm looking forward to listening to it.
Paul Barnhurst (36:55):
It's fascinating. So there's some great stuff in it. So I'll put a plug in there when you talk about that because had you ever thought about earth observation and satellite data and modeling? I
Ryan Humphrey (37:04):
Have. It won't surprise you for me to say no, I have not.
Paul Barnhurst (37:09):
So I'm glad I'm not the only one because when he first broached that topic, I'm like... And then he started explaining, I'm like, "I can't wait."
Ryan Humphrey (37:15):
I have no idea what's going to happen. I'm looking forward to it.
Paul Barnhurst (37:18):
It's well worth it. He even shared his research paper with me, so I got to read it. Definitely worth it. All right, we're going to move into our more standard section. We're going to go to rapid fire here in a minute. Before we get there, your favourite Excel shortcut.
Ryan Humphrey (37:30):
I know people say this a lot, but it is control C, control V. Sorry, I design to choose one. I'll go Control C. There you go, because I'm lazy.
Paul Barnhurst (37:37):
That works. The last guest had an alt ESV, so pay special value.
Ryan Humphrey (37:41):
Values. Yeah. But you sometimes don't want the value. Sometimes you want the formulas. So I
Paul Barnhurst (37:47):
Do also
Ryan Humphrey (37:48):
In OESV as well.
Paul Barnhurst (37:49):
But they first gave me a custom lamb. They gave me a custom shortcut. I'm like, no, no, you got to give me one of the standards. What's the most fun thing you've built with the spreadsheet? Most unique?
Ryan Humphrey (38:02):
Oh, I didn't think I'd be talking about being there so much. I guess my favourite spreadsheet that comes to mind, I'm actually not somebody in general who models for the love of modeling. I know there are people out there, full of respect for them, but I've always been about the story, the meaning behind it, the meaning behind it if you like. And that's where I sit on that sort of separate debate, if you like. The spreadsheet that comes to mind is I've been playing a video game quite recently called Blueprints. It's a puzzle game. It's pretty well known actually. It's done very well in the last year or so, I would recommend. And as part of that, you have to take notes. I'm not going to spoil the game because it should be experienced on its own. But effectively there's a part where I needed to build an Excel spreadsheet.
(38:45):
Of course I did. Now again, if this tells you anything about me as a model reviewer, it's a spreadsheet that had zero formulas. So my most fun spreadsheet, and this is probably I assume I'm the only person to ever sign a spreadsheet with zero formulas. It was just a very beautiful kind of illustration within Excel.
Paul Barnhurst (39:01):
So you're going to be like that Japanese artist if you've ever seen him that does nothing but illustrations and -
Ryan Humphrey (39:06):
Something like that. Yeah. My most fun spreadsheet was effectively a spreadsheet with no formulas for the game blueprints.
Paul Barnhurst (39:11):
Love it. That is great. I don't think we've had anyone say that. We get all kinds of unique things, so I appreciate it. I'm sure you've heard a few of the unique ones over the years.
Ryan Humphrey (39:20):
All
Paul Barnhurst (39:21):
Right. So rapid fire, I know you know how this works. If I can speak, I'll get it out. So no it depends. You pick a side, you can elaborate at the end. Are you ready?
Ryan Humphrey (39:31):
Yes.
Paul Barnhurst (39:33):
Circular references, yes or no?
Ryan Humphrey (39:36):
No.
Paul Barnhurst (39:37):
VBA, yes or no?
Ryan Humphrey (39:39):
Yes.
Paul Barnhurst (39:40):
Lambdas in financial models?
Ryan Humphrey (39:42):
No.
Paul Barnhurst (39:44):
External workbook links?
Ryan Humphrey (39:46):
Definitely not. Whatever has to be a no.
Paul Barnhurst (39:51):
Should modelers primarily use keyboard shortcuts?
Ryan Humphrey (39:58):
Yes? Let's go with yes.
Paul Barnhurst (40:00):
Should models always be print ready?
Ryan Humphrey (40:04):
Yes.
Paul Barnhurst (40:05):
Okay. Are merge cells ever acceptable?
Ryan Humphrey (40:09):
Definitely not. No.
Paul Barnhurst (40:11):
No, no. I got my first yes about an hour ago.
Ryan Humphrey (40:14):
Oh, no. I know who's in the interview as well because I stumbled into it.
Paul Barnhurst (40:18):
I know. I won't tell our audience yet because you come out first, but let's just say here in the next month you're going to hear someone say yes on that one.
Ryan Humphrey (40:25):
They're going to completely undermine my answer, but I'm going to stick with a no for now.
Paul Barnhurst (40:32):
I could resist. I knew you'd be like, wait, what? Should financial modelers learn Python in Excel?
Ryan Humphrey (40:39):
Yes.
Paul Barnhurst (40:41):
What about Power Query?
Ryan Humphrey (40:43):
No.
Paul Barnhurst (40:44):
How about Power BI?
Ryan Humphrey (40:47):
No.
Paul Barnhurst (40:48):
Okay. Should modelers be able to build a fully integrated three-statement model? Is that necessary for a
Ryan Humphrey (40:55):
Financial
Paul Barnhurst (40:55):
Model? Yes.
Ryan Humphrey (40:56):
Yes. Okay.
Paul Barnhurst (40:57):
Will Excel ever die?
Ryan Humphrey (41:00):
Yes.
Paul Barnhurst (41:01):
Okay. Have you used AI to help you build a model in Excel?
Ryan Humphrey (41:06):
I guess yes.
Paul Barnhurst (41:08):
Okay. What financial statement is most important? Income, balance, or cashflow?
Ryan Humphrey (41:13):
Cash. Cash is game, definitely.
Paul Barnhurst (41:16):
That's almost the exact same answer I got last time. Favourite LLM?
Ryan Humphrey (41:20):
Can I count mythos or is that banned now or Fable? I would go with Fable if I can say that.
Paul Barnhurst (41:27):
Sure, you could say Fable.
Ryan Humphrey (41:28):
Okay. Let's go with that.
Paul Barnhurst (41:30):
It will come back out eventually, even if it's a slightly modified version. They'll figure
Ryan Humphrey (41:34):
It out. So if we're counting models that are no longer allowed to be used by people, I would go with Fable.
Paul Barnhurst (41:39):
You could thank my government for that. You're welcome. All right. If you could pick only one for all your models for the rest of your life, which one would you pick? Sensitivity analysis or scenario analysis?
Ryan Humphrey (41:53):
I mean, people define it in different ways, but I think scenario analysis probably.
Paul Barnhurst (41:57):
Yeah, how you define it and I get it. It's not a hundred percent clean question. Do you believe financial models are the number one corporate decision-making tool?
Ryan Humphrey (42:07):
No.
Paul Barnhurst (42:08):
Okay. What is?
Ryan Humphrey (42:09):
I think the question for me is that a model typically validates a separate kind of idea. So I guess if you mean what's the key tool to validate an idea or a suggestion someone's had or a though, then yes, definitely it is. But in terms of that ideation piece, I think that has to come from people, from humans, from people in sectors, technologies, companies.
Paul Barnhurst (42:34):
Makes sense. I get that. What's your lookup function of choice?
Ryan Humphrey (42:37):
X lookup probably.
Paul Barnhurst (42:39):
Okay. And any of those you want to elaborate on any of your answers there? I
Ryan Humphrey (42:43):
Think there's a few of them. I guess the Lambda question is interesting because I know that some guests have said yes. And for me, I feel really strongly that financial modeling should be about understandability and readability, ease of use. No all financial models and spreadsheets, but from my perspective, that's a key part. And I think if there's a function which is not understood by someone, that's not a very helpful function for somebody. Imagine if I build a function, effectively a custom function, a Lambda function myself that only I really understood that that would be a problem. And I know they do lots of things, but I'm yet to receive a model which has complicated Lambda functions because as facially a fresh pair of eyes, that understandable piece becomes really complicated. So for me, people have done amazing things with Lambda functions. I probably just wanted to clarify that before I get hate in my DMs or whatever the LinkedIn equivalent is.
(43:43):
I'm trying to think of other quick fire questions. I think actually a lot of my answers come back to that point about understandability and ease of use, to be honest. The same with Power BI. Of course, Power BI is great. It's a great tool, has a place, and is very helpful. I use Power BI every single day, but is it a skill that you have to learn? Probably not. And you can find other ways of clever reporting. So it's about understanding there as well.
Paul Barnhurst (44:09):
I'm going to follow up on one because I think you're the first one who's done this combination, at least that I can remember. You said yes to Python, no to Power Query and Power BI. Give me a little bit of your reasoning why Python. I've had people say yes to that, but I don't know if they've said no to the other two with that yes.
Ryan Humphrey (44:25):
You know what? I was saying to you before, we started recording briefly that I was listening to an episode in preparation somewhat earlier this week. And instead of doing the rapid fire, I was a bit nervous about this section because their answers were almost exactly the opposite to what I was going to say. But almost apart from perhaps merging sales, we think, I hope not as well. Y said one person now said yes to that, but I think overall it was almost the opposite. I think where I'm coming from is what skills do you need to learn from now? Which is a different question. If you asked me this set of questions two years ago, you would've got very different answers. So what I'm really thinking about is where's the future of financial modeling? Less so Python Index out just Python in general. I think as I mentioned in the show earlier, I've been spending a lot more time looking at Python in the last year than before because I think there are interesting tools and ways to essentially review financial models or build models that will require that almost fundamental skill of Python.
(45:17):
So that's the reason why I've said it's purely based on what I've been, if you like, using myself in the last six months, three to six months in particular.
Paul Barnhurst (45:25):
That makes a lot of sense. And as soon as you start to give the answer, I was figuring that's kind of where you're going. Because the more and more you use AI, AI makes Python very easy to use. Python can solve a tonne of problems.
Ryan Humphrey (45:38):
Very machine readable, exactly.
Paul Barnhurst (45:39):
But you want to have some understanding there. I mean I've been having AI do a tonne of building on my website. And at first I had no idea what any of the CSS was. And now I'm starting, as I understand it better, going through and refining and understanding meta. And so I get it. Similar to Python, certain things you're doing, you just have to understand it to a certain level if you're going to trust the output beyond just going it looks right.
Ryan Humphrey (46:03):
So perhaps one thing I would say as well is that I think learning how to do things is really important. So learning best practices from trusted voices is crucial. Make sure anyone that might be listening is interested in developing their own skills, make sure you go to a trusted provider, someone that has a name that you understand that you can fully trust because I also am concerned about the quality of training. You can get, I'm worried about a future where training materials are AI generated, basically a slop. You get a sort of vicious circle where the unit price of training becomes very low because you can just do obvious things asking AI. And I think that what you can't teach as easily is what's best practice. And maybe I stand to be corrected in the future on this, but certainly there's that requirement to look at those things.
(46:46):
And my answer, I feel a bit bad about Power Query and Power BI. I'm coming back to that though. I do feel bad. They are both very valuable skills.
Paul Barnhurst (46:53):
No need to feel bad. We all have different opinions. That's what the show's about. There's not a right or wrong answer here. That's the whole goal of forcing people to say yes or no is one, there's nuance to all of them. And two, very good modelers, some of the best in the world, can disagree on some things that others might think of as fundamental. So I like it. You're all good.
Ryan Humphrey (47:16):
I feel less bad now. Thank you.
Paul Barnhurst (47:17):
You're welcome. That's my job. That's the only reason I show up here is to try to make sure you don't feel bad at the end. Now let me see if I can make you feel bad with this last question. I'm kidding. As we wrap up here, first, thank you for joining. I really appreciate it. Any last parting words? Any last thoughts you want to give our audience before we let you go?
Ryan Humphrey (47:35):
I think maybe coming back to that theme of self-improvement, almost abstracted from Excel, financial modeling is a really interesting conversation. I didn't expect to have it actually. And I think we've obviously both had different conversations with different people and have identified that theme of obsession to always self-improve despite being at the top of your game already. And I think that just has become so important in the AI era of modeling that we are absolutely in right now. So I guess my message is to not stand still, keep engaging in conversations with people like yourself, make sure you listen to the right voices and keep self-developing and good luck.
Paul Barnhurst (48:09):
Listen to the right voices like going, running while listening to modeling.
Ryan Humphrey (48:12):
Just keep having.
Paul Barnhurst (48:14):
No, I think it's great advice. I love that you ended on always developing yourself. That goes just to life in general. Whether you're a modeller, whatever you may be doing, improving yourself, it's really the best thing you can leave behind is you did the best to make yourself the best you could, which involves improving the world for others. And so I think get real deep on that, but I think there's a lot of great advice that goes far beyond modeling in that. So thank you for sharing it. I really appreciate it. People will be able to find Ryan's contact information in the show notes if you want to reach out to him. But again, thank you so much for carving out an hour and spending some time with me, Ryan. I appreciate it.
Ryan Humphrey (48:53):
Thanks so much for having me. It's been a lot of fun.
Paul Barnhurst (48:55):
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.