AI Builds Obsession for Financial Modelers to Refocus on Specification Testing and Trust Now
In this episode of The ModSquad, Ian Bennett, Partner at PwC, joins Paul Barnhurst, Giles Male, and Ian Schnoor to discuss the rapid changes happening in financial modeling with the rise of AI. The team explores how AI tools like Claude are revolutionizing the industry, the critical role of strong modeling skills, and the ongoing importance of trust in financial modeling. They also touch on the future of Excel in the age of AI and how professionals can adapt to stay ahead in this rapidly changing landscape.
Ian Bennett is a Partner at PwC, leading Deals Modeling in Australia and globally. With over 25 years of experience, Ian has built and reviewed complex models for transactions, infrastructure, finance transformations, and BAU forecasting across Australia and the UK. Ian is a founding member of the Financial Modeling Institute's (FMI) Advisory Council and a member of FMI's Financial Modeling Global Leaders Council. He is also the lead author of PwC’s Global Financial Modeling Guidelines.
Expect to Learn
How AI tools like Claude are reshaping the financial modeling process.
The crucial role of human oversight and strong modeling skills, even with advanced AI.
Why AI in financial modeling is seen as an augmentation tool rather than a replacement.
How the role of financial modelers is evolving in the age of AI and automation.
What the future holds for Excel, and how AI integration is transforming it.
Here are a few quotes from the episode:
“We all trade in trust. Whether you use AI or not, it’s your name behind it. That’s the most valuable thing in our profession.” - Ian Bennett
“Financial modeling is so much more than just building a calculator in a spreadsheet. It’s a profession, it’s a discipline, it’s a career.” - Ian Bennett
Claude and other AI tools are significantly enhancing the way financial modeling is done, but as Ian Bennett highlights, trust and deep foundational knowledge remain critical. The future of financial modeling isn’t about AI replacing human modelers; it’s about how professionals can use AI to work smarter, faster, and more accurately.
Follow Ian:
LinkedIn - https://www.linkedin.com/in/ianschnoor/
Follow Giles:
LinkedIn - https://www.linkedin.com/in/giles-male-30643b15/
Follow Ian Bennett:
LinkedIn - https://www.linkedin.com/in/ianrbennett/
Website - https://www.pwc.com.au/deals/modelling.html
In today’s episode:
[02:10] – Introducing Ian Bennett
[07:00] – Testing Claude in Excel
[10:10] – Focusing Beyond the Build Phase
[14:00] – How AI is Enhancing Modeling Phases
[17:30] – The Importance of Reviewing AI-Generated Models
[19:45] – Trusting AI: Augmentation vs. Replacement
[25:00] – Rethinking the Role of Financial Modelers
[30:00] – The Future of Excel and AI Agent Mode
[33:00] – Reinventing the Financial Modeling Profession
[36:00] – The Future of Excel in the AI Era
[39:30] – Final Thoughts
Full Show Transcript
Host: Paul Barnhurst (01:10):
Welcome to another episode of the Mod Squad. We're super excited this week to have a featured guest with us that will introduce here in just one minute. But before we do that, I have back with me for this episode. My co-hosts Giles Male and Ian Giles. Want to take a minute and just introduce yourself? Yeah,
Co-Host 1: Giles Male (01:30):
Sure. Hi, it is Giles Male co-founder of Full Stack Modeller and new, I would say almost hype enthusiast for AI and Claude. I think it's fantastic. I think I've always said that, haven't I? From the start?
Yes. You've always loved the hype and then very enthusiastic. Right? Thank very much over to in for a quick introduction.
Co-host 2: Ian Schnoor (01:51):
You're always about the hype Giles, but No, just kidding. We're all starting to drink the juice now. Yeah. I'm Ian Schnoor. Great to be back. Paul and Giles with the two of you and our very special guest today, Ian. But I currently heading up executive director of Financial modeling Institute, world's only financial modeling accreditation body career in modeling and excited to be here to talk with Ian as our guest to get his perspectives on what he's seeing in the world of modeling these days.
Host: Paul Barnhurst (02:18):
And then of course, we have our guest Ian Bennett. I'll share a brief bio about him, let him introduce himself here. And then I know Ian Schnoor wanted to say a minute, so I'm going to do Ian and the other Ian. I'm kidding. Ian Bennett is a PwC partner who leads deal modeling in Australia and globally. He's a professional financial modeler with over 25 years of experience. He has built and reviewed complex models for transactions, infrastructure, finance transformations, and reporting across Australia and the UK. He was one of the inaugural MFMs with the Financial Modeling Institute. He has helped shape industry practice through advisory roles as lead author of PwC’s Global Financial Modeling Guidelines and the publication of a library of articles. He was a founding member of FMI’s Advisory Council and is now a member of FMI’s Financial Modeling Global Leaders Council. So, Ian, welcome back to the show. It's been a while.
Guest: Ian Bennett (03:12):
Hi guys. It is fantastic to be back.
Host: Paul Barnhurst (03:14):
We're really excited to have you. There's been a lot going on over the last year, I think in modeling, so it'll be a great conversation. Ian, I know you've known Ian Bennett for a long time. Anything you'd like to add before we jump into the questions? Yeah,
Co-host 2: Ian Schnoor (03:28):
What I will just quickly jump in and chime in and sing Ian's praises a little bit. He and I first met, gosh, it's been quite a while now, Ian, hasn't it? It's possible. It's been more like six or seven years at this point. And when I say the other Ian, I feel like we are almost very similar to Ian's in the modeling world that are on opposite sides of the world. And yet where I've always felt a strong connection is I feel like a lot of the things I have always talked about in a training career for 20 years. Ian speaks the same language. He has always, I feel like as I've tried to reinforce, modeling is so much more than just a spreadsheet discipline. It's always so much more than just building a calculator in a spreadsheet. He's always been impressed me with his view that it's a profession, it's a discipline, it's a career.
(04:17):
He's always talked a lot and you'll hear today I'm sure about modeling so much more. That includes process and planning and organisation and structure and communication and it's a whole multifaceted skillset and I've always felt very, very aligned on that. I think probably Ian, the only thing that we've ever maybe mildly disagreed on is that you would probably fight to the death about the fact that if anyone ever put a circular reference in a model, you'd probably come close to knock their heads off. And I think you've sometimes accused me of feeling. I don't actually feel differently than that. I just believe that everyone doing modeling is going to encounter a circular reference. You better understand it and I want people to understand it. If you happen to work in an organisation where they encourage you or it, that's fine, but mostly I want people to understand it so they can at least make the right decisions. But it's thrilled to have you on Ian. I've always felt a close alignment and I'm really impressed with what you've done to build financial modeling as a discipline around the world and I wanted to share that with our audience before we get started.
Guest: Ian Bennett (05:16):
So thanks Ian, and thanks to all of you. I'll add one thing to that incredibly warm and kind introduction is that I'm also identify as a fan of the Mod Squad where I've watched all your videos and your content and I just want to thank you if I'm able to on behalf of the financial modeling community and indeed as you say, the profession in for the work that you have done to drive that profession and to give it credibility. Incredibly grateful for all of the hard work that you do. So thank you for that and thank you for inviting me back. It's such a lovely group of people to be hanging around with. And on that circular reference point, it is I think a testament to the quality of a relationship that you can agree on so much that you so quickly get to a point of great disagreement that we had when we very first met, but only ever in fun. And I think the way that FMI has taken that point and many others and help people around the world to understand them and to have a common language around them, we should probably come back to that question of common language as we chat today has been an amazing difference ever since you kicked that off as a big project all those years ago. So thank you for that too.
Co-host 2: Ian Schnoor (06:27):
Well thank you. Alright, Paul, why don't you do your magic.
Host: Paul Barnhurst (06:30):
I was going to say, I think I might find one other area we could have split agreement here is modeler one L or two Ls
Co-Host 1: Giles Male (06:37):
Oh L two. Thanks.
Guest: Ian Bennett (06:41):
We've all got energy but we don't have energy for that debate.
Host: Paul Barnhurst (06:43):
Exactly. We're not going to settle that today, but I could resist. All right, so why don't we jump in with I think what's been on everybody's mind. As you know, our last episode we dropped this week has been most popular episode yet. We tested Claude. There's been a lot of buzz around Claude in Excel, the whole beta. So two part question for you. One, curious if you've had the opportunity to test it and then we'd love to just get your thoughts, what you're hearing, what your experience was like, a little bit of that.
Guest: Ian Bennett (07:14):
Yeah, so we are coming into this conversation at a really interesting point. I dunno what the delay will be between recording and publishing, but something's going to happen between then and between now and then as well, right? So I'm
Host: Paul Barnhurst (07:25):
Sure
Guest: Ian Bennett (07:25):
Acknowledge we're moving pretty quickly here, but yes, you are right. There's been a lot of great content including the video that you guys have just done on CLA and Excel. I think there was a consensus probably just a few weeks ago that Claude was running ahead a little bit in this area and I think we've seen that proven out. I haven't personally tested it yet. I haven't had that specific need to do so, but primarily because there's been so much great content around produced by people who have used it and I've been looking quite carefully at that to look for some of the edge cases, the types of use cases that people are using and how it's responded to certain types of prompts and we are forming a view. But if I'm honest, I don't feel the need to immediately get that one because although this does feel like a big shift and there's genuine excitement about it, each one of these shifts has been significant and each one of them is very much on the pathway that we had predicted.
(08:22):
I don't think anybody thought we would never get to this point. So I think we're at that point. Now, I also am not overly surprised by the speed, but I can understand why people that have been working as many years and decades as I have would be surprised by the speed because nothing has moved that this quick in my lifetime. And so that feels fast. I don't know, I haven't asked someone who's just come out of university whether they feel like this is fast, I have feeling they might have a slightly different view, but certainly the speed feels fast. If I think about the long-term trajectory that we predicted, I think we're on the path.
Host: Paul Barnhurst (09:01):
Interesting. I think it's great. You kind of mentioned the path for me, it feels like it's gone a little faster than I thought when we started testing six months ago. I think I made the comment, look, we're where coding vibe coding was 18 months ago and so I figured it would kind of take us 18 months, at least a year to get where we're at now. I think Claude surprised me and I think in and Giles would probably agree with that when they tested it that we didn't expect that big of a leap just because it felt like a quantum leap forward to what we had tested just two months ago.
Co-Host 1: Giles Male (09:32):
Yeah, maybe the one thing I would add to that is I mentioned this on the last episode, I no longer feel like I'm a tester with Claude. I'm genuinely just using it. Everything else we've done, we've tested it because we wanted to. I'm just getting huge personal value from Claude every day. I use it for multiple things. So that's been quite a shift personally.
Host: Paul Barnhurst (09:54):
So the next area I wanted to talk about and Giles, I think I'll send this over to you. I know you've been really big on the whole different phases of modeling and how there's been such a focus on the build phase. So I know you had a question there particularly.
Co-Host 1: Giles Male (10:07):
Yeah, it is probably quite a broad question in, but as I think all of us have seen, there's been so much attention on the build, especially on LinkedIn. Wow, look at this. You click a button, grab a coffee and the model's built, which to me felt very reckless last year and I know we're getting close to things working a lot more reliably, but there's all the other phases. There's the scoping, the the planning, the post model build and I'd be really intrigued to see what experiences you've had of value coming from AI anywhere but the build basically, right?
Guest: Ian Bennett (10:44):
Big question guys. So let's do the foundation. So publicly I've been telling people 20 odd years scope specify, design, build, test and then handover or use and then model running, which effectively is many iterations of the same phases I've just talked about. So if you think about that first stage five, 10% scoping, understanding the problem, what's the questions you're trying to answer? What's the story you're trying to tell specification 25% of total build time. So if you've got two weeks to build a model and you're not working the weekends, then you're kind of looking up to Thursday or week one for that. Then you move into build phase, similar length of time to the spec phase, then test phase similar length of time and then a small amount of time at the end for that real quality handover. So the question is why has the focus only been on the build phase?
(11:36):
And the answer is that the focus has always been on the build phase. If I go back right the way through my career, that is really the only place that the volume of conversation and effort has been. Why is that? Was a couple of reasons. First one is it's the fun sexy bit. So everybody loves getting in there, grabbing some stuff and building something, right? There's a real joy in creating something and for financial modelers there's a real joy in problem solving and problem solving with an elegant solution we get, I think you're all nodding, we get a really visceral joy from going from nothing. We hate editing other people's models, we hate it. What we really love is creating something from scratch that is embodied of the way that we think and is elegant and it's beautiful and it solves a really complex commercial problem in a simple way.
(12:25):
There's joy in that bit. Also, if I look at all the literature and the training that's out there, if I was to graph that volume, take FMI for example, but and add FMI to everything else, it is almost all, I dunno what percentage on the bill facts, it's about the skills you need in order to be able to code spreadsheet to build the model. We don't spend two day training course on, actually we do inside our firm, but most of the world doesn't spend two days training how to do a specification and it probably doesn't spend two days training how to do the review either. I'm not overly surprised by that shift, right? That was exactly the same when people were looking at offshoring large modeling teams. It was exactly the same when Excel released new versions, it was exactly the same when power career power pivot started getting us to question how we were going to be doing financial modeling.
(13:17):
These questions have come many times over the years and the focus has always been on that phase. For me, if I'm going to do deliver a significant project, I'm going to be adhering to those phases. For now, let's just remind ourselves that these things are always in question as new technology comes in, I need to just question again, is that the right phasing? Is that the right methodology? But let's say it is for a while, I need to think about how AI is going to impact on everything the way across that spectrum. So that's why I think we focused mainly on those things by testing part of it. It's not people's most favourite part of the role. So some people love it, don't get me wrong, but again, I can't see that there's a sort of a reason why we don't go well, let's talk about how we can do testing. So I think that's why we focused on build.
Co-Host 1: Giles Male (14:09):
Are you seeing particular, for example in the last episode, I had this idea in my head for ages that you could take a picture of the whiteboard model maps and it could turn it into a document that says here's the plan and it kind of does do that really well. Are there any kind of other specifics you've seen beyond the build where you're like, okay, yeah, this is a change to how we've always operated. It adds huge value.
Guest: Ian Bennett (14:34):
Yeah, so I think this is a question about how is it going to impact on us, right? The way from end to end I think, and that's a really big question. So lemme just very quickly give you a little bit of a framework about how we're thinking about it. And I've touched on some of this in my LinkedIn articles I've produced to try and give framing to it. But in summary, there's the things we do every day, how do we make them better? How do we make them faster, quicker, lower risk, higher quality, et cetera, just change the way we're doing things. The next thing for us is how do we improve the value proposition? How do I make it more attractive? How do I win more work? How do I attract premium pricing maybe or maybe just reduce my costs so that I'm more attractive to work with lots of different things.
(15:13):
I can think about how that value proposition might change the experience that the client has. And finally I need to think about new products and services. So when I'm thinking about those sorts of questions, it feels quite specific, but I need to think about can I turn that into a brand new product and sell that as a new service to our clients because suddenly there's an opportunity to do that that didn't exist before. So if I'm going to go through those three things, I then need to think about where can AI impact on each one of those and what we've done across our team from build and review and also into what we call modern finance transformation and data modeling is looked at every task in the team, like the most junior person to the most senior person through every phase of that model build.
(15:55):
And through every phase of the model audit, what are we actually doing in those steps and where can we augment and where can we replace? And we are testing lots of different things around that continually. And we've launched some what we call more custom GPTs in order to be able to have that embedded in the team and they're using standard GPTs to support 'em in their work. You mentioned specification, I think that's the most ripe area for this kind of work and I've said that publicly and we're seeing that come to life. We have a couple of tools that are designed specifically for the specification phase and targeting the risks around it. So yes, there's creation of the content. So you're right, driver trees on a whiteboard, a hundred percent notes for taken from the meeting, transcription of a meeting. All that content can be assimilated into a specification document if you are really clear about what a good specification document looks like.
(16:51):
And if you want to build a prototype model or an input template or an output template, you need to be clear with it what you wanted to create and ask it to start creating those things that you'll use then to confirm you've got the specification correct. And then we have another one which tests that content and tests it for things like ambiguity, relevant aspects from an industry which are missing inconsistency within the document where you've contradicted yourself. And then simple things like using defined terms and such the like. So we're really enhancing that particular phase as well as all the others.
Host: Paul Barnhurst (17:25):
Thank you, that's really helpful to think about it inch nor any question here you'd like to ask as we're going through this, I'm sure there's some things you're thinking about, maybe
Co-host 2: Ian Schnoor (17:33):
I'll touch on the future of work and the future of skills and teams. And this is not specific to your employer pwc, but just obviously you have colleagues, you have friends, you are in the space. Just curious to get your thoughts. I mean of course there is a camp of people in the world who are of the belief that many if not all white college jobs are going to come to an end that there will be an end to the work that we do in light of ai. I don't feel that way, I feel differently, but I want to hear your thoughts but clearly we can all agree that the way we work is undeniably changing, has changed, there's no doubt about that. So what do you think, as you think about your sector, as you think about your industry, your group, as you think about other people you that work in quantitative jobs and corporate jobs, what are the skills that you think sort of a junior and mid-level person needs to focus on maybe differently today? Let's assume that there will be some jobs out there. What are the skills that in our world, in the modeling world people should be focusing on because it's got to change, it's not going to be exactly the same. And I guess what would your advice be to people in or getting in or in their careers right now? What would you say?
Guest: Ian Bennett (18:53):
So that's a really important question because you're right, it's way broader than our sector. And so I have a real benefit of being involved in that broader conversation, global firm level. These are things that we are considering. But equally when I speak to CFOs, leaders in finance around my client base, that's one of the first questions they're asking me as well. Not necessarily about financial modeling, but what does this all mean for my finance team and what's the new structure of my finance team? And there's a lot of concern about that. There's also a lot of energy and excitement and not just from the bosses who are looking to transform, but for the individuals in those businesses who are thinking about how they can take advantage and use it as an accelerator. And there's lots of different perspectives. So what is my advice for financial modelers and what are the skills that are needed?
(19:45):
So it has always been the case that a great financial modeler is curious and they have innovation in their core and they want to solve problems to do that, they need excellent modeling skills. I think those things will still hold as foundations now to really innovate, to really test and try and fail, you need a culture that supports that. If those individuals are in a team, then you need to look for a team that is going to support that kind of innovative style where you can learn from each other, you can try something, it doesn't go so well. You celebrate what you've learned from that and you move forward. We have to have that culture wrapped around those skills in order for able to that to move forward after that. How is it going to change on a day-to-day basis? Well, we're going to be critiquing and considering content much more than we are creating and as I mentioned at the start that we have to call out.
(20:49):
I think we haven't been honest enough about that because there is a visceral joy with coding and creating in Excel, at least for those of us who get that pleasure that I know I appreciate. Not everybody feels like that many of the people watching this podcast will get that pleasure and there will be less of that I think. But that said, it has always been in the past harder to do the things that we do than it is today. If I go back 20 odd years when I started my career, it was slower and harder to build the models and it was much slower and harder to review them and check that they worked. We haven't complained each time a new tool has come in, which has helped us to do that faster. We are having that conversation today because of the rate of change and the likely more significant impact.
(21:43):
But we have always been adopting better ways. And I'm not going to go back and suggest that we were creating spreadsheets with paper and pen, but we kind of were. So there's been an evolution right the way through all those sorts of things where we have adopted new technology and I'm sure there were people who were like, hang on, if you don't do it with a paper and pen, how can you possibly understand what you're doing if you give it to the computer? I'm sure there was arguments along those sorts of lines. And when you come to review, there were people doing things manually that then add-ins that we all now take for granted would do. And they were probably concerned about that there'll be less thought, less imagination, less interrogation if you outsourced it to those tools. So I think we've worked out how to bring these things in and make the very best of them.
(22:26):
So we will now become more reviewers because we're asking the thing to build things and then we need to look at it. But I think that review thing is going to require some specific skills in don't mind. I think we need to be clear. So yes, you need to be a good model reviewer, you need to be really good at that. And that's the case for most financial modelers. You can hunt things down, you can look for them, you understand the formally that you're looking at and you can intuitively tell if it's right or wrong. You need to be ready for the fact that for some time. But again, this will probably go eventually when AI is coding your formula, it's detecting algorithms, it's using the full Excel library to do so. And as you guys have seen many times, it then uses some extraordinary combination of algorithms that are in the Excel library, which you have never seen in a financial model to do something which intuitively we all know you do it with a sum if guys obviously that's how you do it and it's something and it takes us a while to reorientate and I'm sure that will gradually go, but you have to accept the fact that the machine is coding in a very different way to the way that we think.
(23:31):
So we need to review for things which are more complex, we need to review things which have been created very, very quickly. We need to review things that have been created by a thing that we don't know. That's important because when I train a graduate over a period of years, I watch them develop and know their strengths and weaknesses. I know when they're having a good day and when they're having a bad day, if they roll in the morning and their kind of eyes are kind of blurry and they're taking a couple of pair of seat more, I know that's not going to be a productive day versus the day before where they just look like they just jumped out of bed with a fire in their soul. I dunno that about the AI machine. I dunno whether it's having a good day or a bad day.
(24:08):
And the main reason, I dunno that is because of this plausible quality that is produced every single time. And I've talked about this plausible garbage where there's things which look on first glance absolutely brilliant, but then as you start to, oh hang on a minute, you've made five points and four of them are the same and the fifth one doesn't mean anything. Like hang on. And suddenly your confidence has dropped. And I was using the X Excel agent yesterday and playing around with that, not CLA but the other one. And in a first glance this model was absolutely fantastic that it built and I got right to the financial statements and right the way down before I press and I pressed F two to look at the formula and the net income was not the sum of the two rows above, it was the sum of the subtotal and the row above that it was just a row out.
(24:53):
And it had done that four times in the financial statements. Again, it'll get better, but it took time to find that and to realise that it was wrong. And then of course I'm like how does this balance sheet balance something else has been fudged somewhere in order for that to work if that error is there. So then you have to go and do your review in a way that we've never had to do before because we don't have those signals and we don't have those ways of testing. So long answer, but the answer is review. The answer is we need to get really, really good at testing what's given to us. And maybe if I just give you one last perspective, you've got a judgement to make every time it creates it. So it's just built a house for you. You've literally said build me a house and it's taken however longing it's taken done
Co-host 2: Ian Schnoor (25:38):
One. Just quick follow up then on that, I think I know the three of us and Mod Squad have a view on this. I have a view on this and I was recently interviewing someone who's been using Claude in the beta mode for a long time, number of months since the fall. And he was feeling very confident and I asked him if he was feeling confident using it largely because he already had strong modeling skills and he said a hundred percent he said, I would be scared out of my mind to let someone build a model with AI that does not understand modeling. So my question is, does that scare you? Are you still feeling that to do all the things you just talked about, the reviews and other things, do you still believe that it's critical to have strong modeling skills? I mean I'll tell you I do, but I'm wondering if you do because it's hard to review and find errors and to push back and challenge and argue with it if you don't really know how to do it yourself. But I want to know what your thought on that particular point,
Guest: Ian Bennett (26:33):
I couldn't agree more, but I am terrified in this being on the public record and appearing complacent or looking like I've got my head in the sand and just hoping that this all just goes away. We have been through all of the steps, we beat through all the things we need to do to deliver a service of trust because that's what I trade in. So whether it's the build or whether it's the review, whether it's data modeling and modern finance, whether it's the project finance model audit, I trade in trust and my client has to be able to trust what I do and therefore I have to be able to trust what we've produced. I cannot get there without humans with really good commercial and financial modeling skills. I just cannot see a way that I can get to that point unless trust is needed less. But as you've made the point here, many times on this podcast, you are going to be in that boardroom at some point. Someone's got to carry the can. That trust needs to sit with somebody. And so if I'm not giving it to the stakeholders, where are they getting it from? So I think when you think back from trust, I think you have to have those sorts of stuff.
Co-host 2: Ian Schnoor (27:38):
I love love that analogy and I mean you're right, I've said it kind of differently, but I love that analogy. It's a trust between humans and it's pretty darn hard to look someone in the eye and deliver that trust if you or the team or the person you've done it doesn't have the skills you need. Love that angle and that analogy Paul, that's great. Thanks.
Host: Paul Barnhurst (27:58):
Thank you Ann. And I think it gets to something you've said before, Ann, like the two uses of ai, one is augmentation and I think we're all on the same board that it should be augmenting our knowledge. Is it going to know formulas? We don't know. Of course it is. Is it going to be able to build some things that we may not be as familiar with sure or very familiar with, but if we understand modeling, we understand the accounting, the financial statements, the commercial, we can still validate it. And so I really, I think of AI at this point as augmentation versus you hear this idea of AI's just going to do everything and it's like yeah, it might build our models but that doesn't mean it's doing everything.
Co-Host 1: Giles Male (28:41):
Can I add one point to that as well, an observation? But yeah,
Host: Paul Barnhurst (28:44):
Go for it.
Co-Host 1: Giles Male (28:44):
Keen to get all your thoughts afterwards. The fact that these are stochastic models or the fact that there is no real intelligence in the sense that it's patent recognition, very, very clever patent recognition and the point you were just talking about, Ian, where you might be working on a project finance audit or very large transaction. And if you step back from that saying, alright, it's probably this model that called has produced is probably 95% likely to be correct, but for a lot of industries that's nowhere near enough. You can't go in at 95%. So there's this huge area of modeling that I guess we're all involved in where maybe there's always going to be a problem because it may never be a hundred percent. And really the reason I guess why large companies involved in this spend so much money and train their team so hard is that you've got to get as close to a hundred percent certain that the model is right or free from material error as possible. So the example I gave in the last episode, Paul was like maybe if you're a young entrepreneur and you don't know modelers and you've got to pitch something to try and get angel investor funding, it's like, yeah, I could see where Claude, if the other option is nothing or you trying to learn Excel in your own through Google, I can see where Claw could add massive value and 95% is a very good number for that situation. But then for a project finance deal, 95% is nowhere near enough. That's probably my observation, Charles.
Guest: Ian Bennett (30:13):
I'll play that back from a model audit perspective. You mentioned it. So typical project finance model has between three and a half thousand four and a half thousand unique formula. And I can usually rely on the fact that a unique formula is one formula per row. That'd be a best practise.
Intro (30:27):
I'm
Guest: Ian Bennett (30:27):
Assuming that's what the AI would've done. I don't know for certain, but if it's a 60 time period model and they haven't, then it's three and a half thousand multiplied by 60 or mini in that model. But I need to check every single one of them. And that takes time to go through each of those things, get comfort with that thing. There is no such thing as 95% materiality in a model audit because the concept of materiality struggles around switches and redundancies. In a model I have a switch over here or a MinMax, which means that if I adjust my leverage just a little bit, cashflow start to flow into this section of the model which isn't currently used. And in that section of the model, the interest rate is materially higher. The current base case model looks fine, adjust one assumption model flows in calculation is fundamentally incorrect.
(31:20):
That becomes very material. So that's quite binary as things move sections of model which currently have zeros in them suddenly become materially incorrect. And to do that, you have to understand almost every single formula in the model. But maybe going back to what Paul said, now if a model audit is, and I dunno how other people perform model audits in the world, but if a model audit is checking every single formula in the model, will there be a future where I can still provide the same comfort, the same letter and report to a financier to a client? But I haven't checked all the formula in the model or at least a human hasn't checked every single formula in the model. Almost certainly because if I think about the complexity of most of the formula, commercial complexity and technical complexity, a very large portion of those formally are actually really simple. In fact, maybe even trivial. So maybe I can outsource some of that checking and then review it. I still got to review their work, but that will be an accelerant that will help me move faster to a deadline or meet project finances deadlines quicker and maybe even lower my cost to serve, et cetera. So there is some replacement there. There's a possibility of replacement in that model audit, there's no doubt about it. But on lots of other things it's augmentation.
Host: Paul Barnhurst (32:34):
Thank you for that makes a lot of sense. We just have a few more minutes here. I want to touch on something. You wrote an article, you've written several on LinkedIn, I know, but you did one recently about modern finance transformation and you said something I thought that was really interesting and I'd love to just drill into a little farther. You said AI agent mode was the single most transformative moment in the history of Excel. I'm curious, why did you say that? What is it about the agent mode that is transforming Excel?
Guest: Ian Bennett (33:04):
I presume I'm not allowed to say it's just clickbait. No, it should be some thought. There's probably needed to be some thought behind it. So you're
Host: Paul Barnhurst (33:11):
Getting good on LinkedIn then Welcome to the club in, I'm just kidding.
Guest: Ian Bennett (33:16):
So I said Excel right now. Let's think about the Excel user base for just two seconds. So anybody in a corporate setting and many outside of it with a laptop has Excel and 95% of them have used it. I don't know what that number is. You guys will probably know how many hundreds of millions of people, maybe billions of people that is, it's a very large number of people say it's a billion,
Co-Host 1: Giles Male (33:42):
Right? I say that's like the rough top end number, isn't it?
Guest: Ian Bennett (33:45):
So let's take the right nice brand number. It's a billion. So how many of those people will be able to use Excel mode to change the way they do things? So there's definitely a portion down the bottom that won't and there's definitely a portion of the top and the curve is pretty steep when it comes to skills and knowledge. Our conversation is limited to this incredibly narrow bit on the far side here. That's where we've been hanging around a lot of debate, a lot of discussion, a lot of hype as you guys are frequently saying in that very narrow bit on the end, I think the opportunity is this sort of other half, right? Maybe it's 45, 50% of Excel users this can help. And that doesn't worry me because I never have any interaction as a professional with any of that, right? That's just going to improve people's days.
(34:33):
It's going to improve the speed that a finance function works at. It's going to mean financial close happens a little bit quicker. It means you get your reports to the board a little bit faster. It means that an organisation that's desperately trying to raise funds for its charitable purposes do so in a way that looks professional and looks slick, which I've never been able to do before. We will be able to achieve things that we've never been able to do before, but I don't think that impacts on my world hardly at all until it goes wrong. And then maybe then I'll start to see some of the things that come out of that. But that's why I think it's transformative is because it's the impact it's going to have on everybody, not just the impact it has on us professional financial models. That said, if we didn't see this as a moment of change, a material change, if we didn't see this as a greatest opportunity to reinvent the financial modeling profession, to understand what we do, to understand the value we give, what is our purpose?
(35:30):
Then we have missed a massive opportunity. So when I think I mentioned to you, I'm going to bring the global financial modeling leaders from PWC together in London in a few weeks time, our question will be how do we reinvent our profession? We have an opportunity here to redefine it. And if anybody in that room thinks, well we don't have a value anymore because I'm valued for the quality of my vlookup, they won't. By the way, these are all extraordinary leaders, but I'm sure there are people in the financial, but not vivo kidding. But there are people in the world that think they're valued on the quality of the formula they write, right? In our world, in our profession. And I think you just have to understand that's not what we valued for, right? We're valued for something much greater than that and we can take that and we can enhance it. And
Co-Host 1: Giles Male (36:11):
Just to say I completely agree with that. Again, I think we mentioned this last episode, we are in this bubble of the top end, whatever it is. But for everyone that doesn't wake up thinking God can't wait to get better at Excel today, the fact that you could potentially go to an agent and do some pretty impressive stuff is incredible. I love that idea. I really do.
Host: Paul Barnhurst (36:34):
So before we wrap up here, I want to ask one more question. We've all seen it, the hype and the clickbait. Excel is dead, we now have AI and all this stuff. I find it not surprising at all, the Claude didn't build its own spreadsheet, it released a tool inside of Excel. So how do you think about the future of Excel with all this AI and Microsoft? I mean obviously they're putting a lot of money behind it, but would love your thoughts of just how you think of the future of the spreadsheet as our world seems to keep changing every time we turn around.
Guest: Ian Bennett (37:04):
Yeah. Here again, I could be end danger of being complacent, right? Because my love of Excel is pretty well known as it is for all of us on this call. And I clearly want to believe that it has a future. There have been times where many people, the overwhelming sense was that it was dead and I was defending it and I'm on the public record as talking about where I felt like it wasn't loved anymore, that it was ignored, that it was just something that existed in the world of technology. Everybody had it and the conversation was somewhere else. Two things have changed. One's a bit technical and another one is strategic. The first one is that we are shifting to a world where licencing cost activity is paid for on a consumption basis. That means every time you do something, every time you interact, every time you fire up some intelligence, et cetera.
(37:53):
That's kind of the cost model that people are starting to face. It's not, I bought a licence for a year and now I get unlimited use. So that's important. Maybe I frame mine a second. And then the second thing is that Microsoft have realised that for the finance world and many other worlds, there's nowhere we feel happier. I'm more comfortable and more safe than inside Excel. If we start our journey there, we're reassured, we're happy. Don't say to me, oh, I've got this new system. It's a black box. But if you look at the outputs, you'll understand them. I'm not going to get that. And Power BI dashboards have always struggled for that very same reason. I want to be able to look at inside it and interrogate it. I want to start from a blank canvas and create it. Excel gives us that and we are super comfortable there.
(38:42):
There's no surprise that power create power pivot, power automate power apps, copilot are all accessible through Excel and Microsoft. No, that's our comforting home and we'll start there. Yes, we'll move on into these other tools from there, but that's where we'll start. And so Microsoft has spent a lot of time making that home feel comfy. They want us to stay there and they want us to consume from there. So from that Excel place, we will consume ai. We will consume copilot or consume Claude as well. They actually don't mind that because it's still using that platform through which they're attracting us in and asking us to use it. And we'll consume power query, we'll consume power pivot, we'll consume it from that place that we feel most comfortable. So I think the whole perspective on the importance of that place that we've loved for so long has changed. Thank,
Host: Paul Barnhurst (39:29):
I think that's a great answer. I really appreciate that. Before we wrap up here, why don't we just go around everyone, any last thoughts or last question you might want to ask before we finalise Giles, anything you want to say before we wrap up here?
Co-Host 1: Giles Male (39:41):
I'm just reassured again, it's just really interesting to hear your views in and reassuring to hear that they're not completely different to what the three of us have broadly reached I think over the last few months. And I'm annoyed with all of us because, or you said that you'd be struggling with any sports analogies and not one of us threw a sports analogy and to confuse you to very disappointed
Host: Paul Barnhurst (40:02):
Football American, no kidding. Inno, anything you'd like to add?
Co-host 2: Ian Schnoor (40:06):
It's always great to catch up with you Ian, and to speak and to hear your thoughts and to hear your views and very, it's confirmatory, but it gives us a lens into your world and in your mind and your thinking around the future of humanity and the future of thinking in the future of delivering some value that is different than what can be created necessarily from a computer inside a box. And so who knows where we're all heading? Well, I agree with you though We are far and away. You and I guess similar vintage can think back to other disruptive technologies in the last 25 years. But absolutely, certainly nothing compares to the speed and the intensity and the potential disruption that we are dealing with right now. We're in the eye of the storm, so no one knows what it's going to look like on the out once we get through it, but we all know it's happening, it's here and change is in the air. But to hear your views that there will still be a need for what we bring to the table collectively is encouraging and validating. So thank you for all your continued leadership and support and look forward to continuing our journey here together
Host: Paul Barnhurst (41:17):
In Bennett. Last words go to you.
Guest: Ian Bennett (41:19):
Oh, thanks guys. It's been a real pleasure. What's my last words? We all trade in trust. Like every single one of us has a stakeholder. Every single one of us has a boss. So if we're going to ask these things to take advantage of these things, to do them, I think I said they build us a house, you need to very quickly have all the skills, all the financial modeling skills and passion and excel skills better tell, is this a house that needs to quick polish? Is it the house that needs a full referred or do I have to move the entire house six inches to the left? Because if you have, you should have done it yourself from the start. And we need to be able to know that really, really quickly so that we can offer that trust up to others. And I think we mostly have all of those skills and I think we're innovative, we embrace this moment and we've got a fantastic future ahead of us as a career, as a profession.
Host: Paul Barnhurst (42:05):
Thank you so much for that and
Co-host 2: Ian Schnoor (42:07):
A very positive way to end. Love that. Thanks. Thanks Paul.
Host: Paul Barnhurst (42:10):
Thank you everybody. Thank you Giles. In and in. This is another episode of the Mod Squad and we're really excited that you joined us for this. And I think I'll just end on the note. I think in made a great point here. Remember, we trade in trust. Regardless of what tool you're using, regardless of how you do the build, it's your name behind it. At the end of the day, whether you use Claude, whether you use copilot, whether you built it yourself. So remember the most valuable thing you have in this profession is trust. People want to know you do a good job. I think that's the key thing here. No matter how much you use technology to save yourselves time, always remember that trust factor. Alright, thanks guys.