How Microsoft is Making Excel an AI-Powered Platform with Product Manager Brian Jones
In this episode of Financial Modeler’s Corner, host Paul Barnhurst sits down with Brian Jones, Vice-President: Microsoft Excel, Forms, Office Platforms, to discuss how Excel is evolving in the era of AI and large language models. They explore how Copilot, agent mode, and AI-driven features are reshaping how professionals build, analyse, and improve spreadsheets. Brian also shares insights into how AI can assist financial modelers, automate tedious spreadsheet tasks, and help users understand complex workbooks faster.
Brian Jones is the Vice-President: Microsoft Excel, Forms. He has worked at Microsoft since 1999 across several productivity platforms, including Word, Office Forms, and developer extensibility tools. In his current role, Brian leads the Excel product team as they integrate AI capabilities like Copilot and agent mode to expand what users can accomplish inside spreadsheets.
Expect to Learn
How AI and Copilot are changing the Excel user experience
What “agent mode” means for automating spreadsheet tasks
How AI can help analyse and improve financial models
Why Excel still requires strong modeling and formula knowledge
How skill sheets and templates can guide AI inside workbooks
Here are a few quotes from the episode:
“Excel really is like a developer tool. When you're writing formulas, you're essentially coding.” – Brian Jones
“One of the strengths of Excel is that the work is visible. You can see the formulas, the logic, and how the result was produced.” – Brian Jones
Brian explains how AI is transforming Excel from a traditional spreadsheet tool into a more collaborative environment where users can interact with their data through conversation. Copilot can analyse spreadsheets, suggest formulas, correct errors, and even rebuild broken references automatically.
Follow Brian:
LinkedIn: https://www.linkedin.com/in/brijones/
Website: https://learn.microsoft.com/en-us/archive/blogs/brian_jones/
Follow Financial Modeler's Corner:
LinkedIn Page- https://www.linkedin.com/company/financial-modeler-s-corner/
Newsletter - Subscribe on LinkedIn -https://www.linkedin.com/build-relation/newsletter-follow?entityUrn=7079020077076905984
Sign up for the Advanced Financial Modeler Accreditation Today and receive 15% off by using the special show code ‘Podcast’.
Visit https://bit.ly/497oAqW and use the code “Podcast” to save 15% when you register.
In today’s episode:
[00:00] – Trailer
[02:53] – Brian’s Background at Microsoft
[03:09] – Returning to the Excel Team
[04:24] – The Rise of AI and LLMs in Excel
[07:08] – Customer Questions About Copilot
[17:52] – What Skill Sheets Are and Why They Matter
[23:07] – Power Query and Excel Programming Languages
[30:51] – Do We Still Need to Learn Formulas
[36:07] – Getting the Most Out of Copilot
[38:40] – Favorite Excel Shortcut and Functions
[41:31] – Where to Follow Brian
Full Show Transcript:
Host: Paul Barnhurst (00:00):
Financial Modeler's Corner is the world's Premier modeling podcast. It is brought to you by Financial Modeling Institute, the world's leading financial modeling accreditation organisation. Welcome to Financial Modeler's Corner. I am your host, Paul Barnhurst, AKA, the FP&A guy. In this podcast, we talk all about the art and science of financial modeling with distinguished financial modelers from around the globe. The Financial modelers Corner podcast is brought to you by the Financial modeling Institute. FMI offers the most respected accreditations and financial modeling and that's why I completed the Advanced Financial modeler. This week I'm thrilled to have a special guest with us, Brian Jones. Brian Jones is the head of Microsoft Excel, the product team there and we're thrilled to have him. So Brian, welcome to the show. Thank
Guest: Brian Jones (00:53):
You. Thanks for having me. I'm excited. Yeah,
Host: Paul Barnhurst (00:54):
Excited to get to talk Excel. I think everybody's excited with all the announcements we seem to hear almost on a daily basis now about something with Excel.
Guest: Brian Jones (01:05):
Absolutely.
Host: Paul Barnhurst (01:06):
So how about we start with you giving us a little bit about your background. We'd love to know a little bit more about you.
Guest: Brian Jones (01:10):
I've been at Microsoft since 1999, so quite a while I joined right out of school and I've always been working in the general productivity space, so all the kinds of apps you're used to using Excel, Access. Started the office forms team at one point, but also did some work around more on the extensibility side, how developers go and build solutions on top of things like Excel. So I've been using Excel. This is my second go on the Excel team and I've been in this role currently for about nine or 10 months.
Host: Paul Barnhurst (01:44):
Got it. So you've been back in the role for a little under a year then back. Yeah,
Guest: Brian Jones (01:48):
That's right.
Host: Paul Barnhurst (01:48):
Given your background, I know you didn't spend a lot of time as modeling, so I modified a few of the questions. I'm not going to ask you about that horror story with a financial model, even though you might have one, yours are a little different in how you get to work. So where I'd love to start is I'm curious when you started back in Microsoft, if I got it right, I think it was 1999 with Word. That's
Guest: Brian Jones (02:08):
Right, yeah.
Host: Paul Barnhurst (02:09):
Did you ever imagine that 27 years later you'd still be at Microsoft working with the productivity teams?
Guest: Brian Jones (02:15):
No, not at all. I thought I'd be at Microsoft for a few years and then go try something else. The thing that's kind of nice about Microsoft and even just the office team within Microsoft, and when I say office I mean more like M 365, everything from SharePoint teams, outlook Exchange, all of those things. You can essentially get new jobs pretty regularly when you're ready to try something new and each team kind of has its own unique culture. Clearly there's a lot of things that are similar across, but there's a lot of things that are unique both in terms of the people, the product, and even a customer base. So whenever I started to get a little bit bored, there'd always be that next role to go and take on that I'd get fired up about again and excited and intimidated at the same time.
Host: Paul Barnhurst (02:59):
Definitely one of the benefits of working for a large company is you have many different products, different things you can do, and it sounds like you've done quite a few over your career.
Guest: Brian Jones (03:08):
Yeah, that's right.
Host: Paul Barnhurst (03:09):
So what brought you back to Excel? You were there from 2016 to 2022 and then you said you came back about 10 months ago or so. What decided to bring you back toward Excel?
Guest: Brian Jones (03:20):
I was almost always in more of a typical product role, end user facing products, but I went for about two years and focused on our SMB business still from the product engineering side, like building experiences around how people buy our products or how partners go and manage it. And that was fun. It gave me a lot more exposure to how people make the purchase decision, how they acquire and manage and all of those things. But I'd done that for about two years and during that time, all of a sudden the explosion around LLMs and AI really happened and I was focused a lot on how to make it easier for SMBs to acquire our products, like the AI investments that we had. But I was also just so excited to get back into being on one of the core apps and really think through how those should evolve. And so right around spring of last year, the leader for the overall office reached out to me and asked me if I wanted to come back and I jumped on it, I was pretty excited to come back and start thinking about Excel in this new world.
Host: Paul Barnhurst (04:24):
I definitely want to spend some time in this new world. That's where I'd like to talk quite a bit about. I think that's where our audience is excited. I mean, what were you excited about in this new world? What got you excited about working on the product and AI and LLMs? Was there something in particular or just the overall thinking about the strategy or what was it that was like, yeah, this totally makes sense? Yeah,
Guest: Brian Jones (04:46):
I'd say it came from a few different areas. One is I've always, the thing that always gets me excited about Excel, and this goes back to a lot of my experience more on the extensibility developer side, is that Excel really is like a developer tool. It's an IDE. People don't usually identify as a developer when they're using it, but you're coding, when you're writing formulas, you're building out logic. It's got this beautiful model with Instant Recalc where you don't have that right code to compile and see the results. You just make the change and you immediately see the results. So it's a really easy way to start to learn, but at the end of the day you're still programming and coding. So it's this really neat combination of a design time coding experience plus a runtime to go and run whatever app you've built. And a lot of the really neat early innovation on the LLM AI side was more around coding tools, like helping you go and build web apps and things like that.
(05:45):
And it had started to get to a point where it was almost easier to go build a complex web app and a coding tool than it was to build a complex spreadsheet, which is a complete flip in how it had been for the decades before. And so to me it was just really clear that we had a huge opportunity to bring those same concepts in a different way to Excel users and at a minimum, at least make it so it's faster so that you can build the thing you were wanting to build, but it can automate certain pieces or produce what you're intending, but ideally it takes you further than what you may have thought you could do. There's plenty of people that use Excel for very basic tasks, just managing lists, but as soon as they see things like pivot tables for summarization, learn more about how to write formulas, things like that, it's a big unlock to the type of what they can build, what type of analysis they can do. So I was just really excited to really rethink how people would experience Excel once you put these AI tools alongside it, really more deeply integrate them in the overall experience.
Host: Paul Barnhurst (06:57):
That makes a lot of sense. I'm curious, what are the questions you're getting today from customers? What are people asking you about Excel? I'm sure AI is big, but what are the general questions you're getting? Yeah,
Guest: Brian Jones (07:08):
It's really varied because everybody is at different stages of AI adoption. Some folks, it's kind of just really much more basic, what can I do, what can AI do for me? And really their people are pretty blown away immediately by what you can accomplish that they didn't really think that they could do before in Excel or they had to have a much higher level of mastery. But there are other companies that we talked to that are really deep already in their AI journey, and so for them there's more advanced things where they're kind of curious about more advanced concepts, things like skills and things like that. How do those work when they're using copilot and Excel? We also have it where we've had a few different waves of release for our copilot capabilities. So when you bring up the copilot task pane in Excel, we had a first wave, it was back in the spring when I was first showing up.
(07:56):
The team had already released some capabilities, but it was before the models were really great at deeper chain of thought tool calling orchestration. And so we almost had to guide the model, do specific tasks, and so the first wave you would just do something like you could say add a conditional formatted column or add a new column that uses formulas based off of these other columns. It was a lot of it really heavy on tables and working with tables. The problem was it's still had just a text box where you could ask any question. And so a lot of the questions people would ask, we'd have to say, sorry, I don't do that. These are the things I do, and that's not a great experience. Right. No,
Host: Paul Barnhurst (08:40):
I remember getting a few of those.
Guest: Brian Jones (08:42):
Okay. And so then we had the next wave, which where we said, okay, let's at least make it so that the things that you could do off in a different tool if you were just going to go and launch M 360 copilot on its own and maybe ask it how to write a formula or ask it to do some analysis for you, most of these tools use Python for doing analysis. Could we at least make it so that you don't have to leave Excel so you could go and it can look at your spreadsheet answer questions for you. And so we went and rolled that out towards the end of the summer, but of course the majority of the things people really wanted was to ask it to actually have agency and do something on the spreadsheet, change it for me, fix things, build something.
(09:28):
And so that's what we've now been rolling out over the past couple months. It currently shows up in the product as a toggle called agent mode where you bring up copilot and you turn it into agent mode. That toggle will go away. Really, you should think of it as that's the new version of what copilot and Excel will be. It's as we're rolling it out, we have it where you can still toggle between the two, but eventually it'll just be you bring up copilot and copilot acts that same way as what you get in agent mode. And so I think a lot of the questions we get is that back to your original question is that different folks have experienced it as different points in the journey. And so a lot of people also, depending on what their own internal company practises are around rolling out versions of office, a lot of them still have some of the older versions of copilot and so they'll go and see what we have and maybe read a post that I might do on LinkedIn or something and say, I don't see any of that. So that's another big question we get is just from people saying, Hey, how do I get this latest, what you're currently calling agent mode, what I've always expected Excel copilot to be able to do.
Host: Paul Barnhurst (10:39):
Yeah, that makes a lot of sense and I know even for myself just trying to follow is the changes of happen and I've done some training, I did some training recently with a company and some of 'em are like, I don't have that piece or it doesn't look that way on my screen and trying to manage all that. And so I know that that's a challenge and a frustration at times, and I know it's a big challenge for you guys is there's rollouts, annual rollouts, people on beta, and then there's people that don't even have 365 in some cases and you're managing a lot of different versions. So I could only imagine with how quick copilot works, it increases that challenge.
Guest: Brian Jones (11:19):
Yeah, there's certain pieces that we're able to push out pretty quickly just from the service side, but there's other things that actually require the client itself to be updated, like the rich clients with the Windows or Mac. And so yeah, sometimes there can be pretty significant delays depending on how quickly that company gets the latest builds, whether or not it's on semi-annual versus the monthly enterprise channel versus our current channel. I don't know how much the modeling community cares that much about the logistics of deploying bits for Excel, but you could imagine it could have its own model itself to go and analyse and see whether or not we have too many different options right now.
Host: Paul Barnhurst (11:57):
I'm sure somebody's built a model. Kind of speaking of the modeling community, what are you hearing from the modeling community about Excel agent?
Guest: Brian Jones (12:03):
The thing that I've kind of found is there's kind of a couple of different levels to what people are trying out. I think the first thing most people try out is just building something from scratch, and so they'll have the one that, the common one, everybody always is do a DCF on this company and build it from scratch. And so that's what a lot of folks go and try and we let you use not to overload the term model. That's funny for you guys. I'm curious because the actual LLMs, which we also refer to as the models, so not a financial model, but the language models, we have choice. So you can choose between the Anthropic models or the OpenAI models.
Host: Paul Barnhurst (12:45):
And
Guest: Brian Jones (12:45):
So right now we've got GPT five two or Opus four five, and then of course, we'll updates of Opus four six and GPT five four just came out that was just announced yesterday. And a lot of folks, you actually will see very different outputs depending on the model you pick, right? And that's because if you think about it, there's kind of for Excel copilot, there's the base level of capabilities. Does it know how to write a formula or enter conditional formatting or do borders in the right way? Just does it know how to use all the features of Excel and then if it does, does it know how to compose them in the right way? But you get to these higher level concepts, does it know how to build a model the right way or does it know the different instances of what type of a model you're using?
(13:34):
And so the open AI versus the claud models have different awareness of finance concepts, the overall workflows you would do. And so one of the things that I see a lot as folks just playing around with the differences between the two models and the differences in output. The other quick thing I'll say though is that clearly usually you don't start from scratch, and so I think that that's what people are trying. They're just kicking the tyres, but most work you have something you already have in you're working with, you're augmenting it, improving on it, changing it, and so that's another key thing for us is really making sure that copilot is really good at really being this companion you can work with. So it's not like just the one shot go build a thing, but it's more help me improve on this thing, help me analyse it, help me find errors in this workbook.
(14:21):
Is there something wrong in the logic that I used around how I built this up? Help me go and make a change. That's something I really enjoyed. Those things are those tedious tasks, right? Like, oh, I want to break this out now in this different way, but now man, that's going to be a pain to go and do all of that. It's great just to go and ask the model to do those where it's actually not that complex. It's just takes a lot of time. One of the demos I love doing right now is where you've got that massive set of pound ref errors in your workbook because of some copy and pasting that had happened at some point or something, right? And it does a really good job going and analysing that, seeing the surrounding tech so it can infer, oh, it meant to refer to this worksheet and that worksheet's not there anymore. You're now using this other worksheet, so I'm going to go and update all the formulas so that they go and point at that, which is again, something you could do. It just would be a pain
Host: Paul Barnhurst (15:07):
Finding and replacing broken links so to speak can be a real pain and that's something that's great at. And even if it gets one of 'em wrong, it's not like you can't reprompt it and say, Hey, you should have used this link instead of that.
Guest: Brian Jones (15:22):
Exactly. Yeah, and that's the thing. It'll tell you what it did and then you can say, no, I don't. That's not what you should have done. I had rather you do it this way. I get that a lot where it'll go and create a chart for me automatically when I ask for a dashboard and it won't be the chart I want it, right? I'll say, no, I wanted this to be a line chart or I want it to be formatted in this way. Can you go and update that and it'll go and fix that for you? The next step of course is for us to start to build in memory so that we'll remember those preferences for you, so over time you don't have to do that a couple of times, then it'll learn.
Host: Paul Barnhurst (15:55):
Yeah, definitely starting to see more and more of that with some of the third party and different things where they're trying to find instructions or memory, and I know it's coming there too, but we're starting to see more of that. The one that I've always found for whatever reason it struggles with and at first I could never get one. I think I finally got one that looked right is to do the totals correctly for a waterfall chart. I just want the first and last one to be set total. I tried that with six different tools and not when they first came out and not a single one did it right, and I kept prompting 'em and I finally just said, forget it. It was pretty funny. I don't know why, but that's one where rarely ever the first time I ask it to do a waterfall chart, does it set both of 'em to a toll level at the end no matter what I,
Guest: Brian Jones (16:40):
I'll have to go try that now and see
Host: Paul Barnhurst (16:43):
Where I think I did get one recently that did it, but it was just a pain. I'm like, all right, I'm just going to do the chart on my own. I'm done playing with this.
Guest: Brian Jones (16:52):
Yeah, I think that's the other key piece, just that point Paul, is these things just keep getting better and better. The models keep getting better and we keep 'em updated with the latest models, but also just the set of capabilities, things like skills, memory, things like that. And so what you tried a month and a half ago that didn't work might work now, and it's a weird pattern to get into where we're not used to that. We're used to trying software. You frame it in your head and you're like, okay, I now know what this thing can do. And so the idea of continuing to come back to things that might've failed before is kind of a weird experience. It's something we're used to with humans. You're like, oh, maybe you learned that, so I'm going to now try and talk to you about it again. But we're not really used to that with software.
Host: Paul Barnhurst (17:36):
I agree. And so I want to go a little deeper. I know something you shared recently on LinkedIn we've talked a little bit about is the use of skill sheets with copilot. Maybe can you tell our audience what skill sheets are and why should people be using them to get more out of ai?
Guest: Brian Jones (17:52):
I mean skills, it's a pretty common pattern right now in the industry around tools. You see it a lot, especially in the coding tools, but the way to think of skills at a high level is it's just think of it almost like a guide for the LLM for the AI to go and use where out of the box it's trained on things. It's got a bunch of knowledge, but just like all of us, we don't all have everything in our heads all the time. And so you want to go and in a certain case you might want to say, Hey, I know you kind of have a basic concept of like DCF analysis, but here's some even more depth in terms of rules and patterns. I want you to go unfollow. I want you to go and reference these data sources anytime you're doing a lookup or I always want to make sure that you're laying things out in this way.
(18:38):
For us out of the box, when we refer some of the initial waves of our agent, we still were very preferred to go to column major tables like the actual table feature. And some of that's because I gave you the story of the evolution we went on with copilot. Some of the initial features we had were around helping you more of those tracking list type scenarios and things like that. And obviously financial models don't look like that in any way, and so we'd still have it at times where we were, sometimes we were falling into listing quarters vertically rather than horizontally as an example, and we continue to get better and guide it, but that's a place where skills also plays a role where if you have a set of preferences for how you want to do something, even if our model is doing something in a good approach from an industry perspective, but there's still some special way that you work in your company, the way you lay things out, the kind of key things you want to analyse, you can just create.
(19:34):
Right now we just call it a skill sheet. It's just a worksheet. It could be called skills. You just put it inside your workbook and you could lay out those additional rules. You could do formatting. So if you say, Hey, I really want totals to look like this, you could just then format that cell and then the model will automatically see that and it'll take all of those instructions into account as it's building up and working on your scenario. Eventually we'll have this as a thing that's more centrally managed where just you can have a place where your skills are stored that you can go and refer to, but we figured it was easy enough just to go and copy worksheets around that. That was a good place to start because then we could start to work with customers, see what additional skills would they want to go and write and put into their workbooks, and then we can start looking at those and if any of them are more broadly horizontally applicable, we might say let's just build that into the agent out of the box so you don't have to give that kind of guidance and you can really get into the things that really are more unique for you.
Host: Paul Barnhurst (20:33):
While my background is in fact, I am also passionate about financial modeling. Like many financial modelers, I was self-taught. Then I discovered the Financial Modeling Institute, the organization that offers the Advanced Financial Modeler program. I am a proud holder of the AFM. Preparing for the AFM exam made me a better modeler. If you want to improve your modeling skills, I recommend the AFM program podcast listeners save 15% on the AFM program. Just use Code Podcast.
Guest: Brian Jones (21:28):
Yeah, that was
Host: Paul Barnhurst (21:29):
The
Guest: Brian Jones (21:29):
Example that I gave.
Host: Paul Barnhurst (21:30):
Financial modeling or whatever it might be for different types of things we're doing. There's kind of at least a template you can start with maybe then you go in and adjust it. Yeah,
Guest: Brian Jones (21:42):
That's kind the idea. And there'll probably be a broad set that we'll have out of the box. Like I said, this is a lot of, Claude has a set that they've gone and built around finance for instance, and so there'll probably be a set that we'll have that are the base level set of skills. Is this a project management scenario, so here's the best practises versus is this a financial modeling scenario? But in general, we see that every company does things a bit different. They have their own guidelines around even things like formula authoring and things like that. And so we think that we'll have a base set and then we want to make it so that anybody can then also go and augment with their own additional ones. And I think the example of the term you used template is kind of the way I like to refer to it as well. I mean, as you know, it undersells it a little bit because skills really can almost be like a programme where you talk about scripts that should be run or these are the data sources. So it's not template really in the traditional office sense of just a formatted spreadsheet, but I think conceptually that's kind of how templates had been used. It's just more that this is a much more advanced version of a template,
Host: Paul Barnhurst (22:50):
A template on steroids. It's what I call the power query. Join feature a look up on steroids when people always argue over what your favourite lookup is. I'm like, I like Power Query myself. That gets some looks like, wait, but we're discussing V lookup versus X lookup. I know, I don't care.
Guest: Brian Jones (23:07):
And is it something where folks actually aren't even aware of it or they're aware they just haven't built the habit around it?
Host: Paul Barnhurst (23:12):
I find mostly the audience that follows me are usually pretty educated in Excel. So many are familiar with Power Query and joins, but you definitely get some like, oh wait, what? And you're just like, okay, you need to go learn this tool. It still amazes me more when I train. Train teams are like What Power Query? It's amazing how few in finance still use it. When I bring it up on LinkedIn, I usually have audiences that are following me that are fairly advanced, a lot of financial modelers and people that are more advanced in Excel. So there I don't get it as much.
Guest: Brian Jones (23:44):
Okay. Yeah, I mean I love Power Query. It's pretty amazing what you can do at that transform layer. For folks that don't know, it also has its own programming language, right? M where there's a whole language behind it. We try to abstract that from you. So there's the UI for the transforms, but yeah, underlying that is just M,
Host: Paul Barnhurst (24:07):
And you'll probably laugh at this, I'll share this real quick. As I one time tried to list all the different languages in Excel, so custom number formatting kind of saying loosely like M script and D and VVA and formulas and custom formatting, and I think you could do MDX in a few places with the old cubes and different things. And I came up with a list with 12 and then everybody's like, no, you can also do this with the ID and this. And I think by the time we were done it was up to 20. Depending on how you defined it,
Guest: Brian Jones (24:37):
It gets massive. And then if you go back, there's some older ones too. I always forget, there's the thing that we had before VBA shoot, I don't remember. I should know that was my job for a while, but maybe I've put it out of my head. But yeah, there's quite a few, especially when you go back to some of the old legacy stuff that were more kind of targeted, purposeful. Well, some of the examples you gave are again, not something that people would typically think of as a programming language, but really are.
Host: Paul Barnhurst (25:06):
Yeah, and it I'm like programming or scripting depending on how you want to define it. I get they're not if you want to say touring complete per se, but they're still basically programming languages. Oh,
Guest: Brian Jones (25:17):
Python
Host: Paul Barnhurst (25:19):
Was Python ones.
Guest: Brian Jones (25:20):
Okay.
Host: Paul Barnhurst (25:20):
That was enough office scripts, so yeah, it was long list. Do
Guest: Brian Jones (25:23):
You see that at all in the modeling community? Any use of the Python function Much?
Host: Paul Barnhurst (25:28):
Most people I talk to, I don't see using it much with modeling. I think it's much more analysis right now. Yeah, some people might be using it with data or charts, but I don't see people using it for writing for building models. I haven't seen that. I'd love to know somebody out there. I'd love to interview them if they're building a model. Mostly using Python Joe,
Guest: Brian Jones (25:50):
We have a guy on our team. Joe kind of leads a lot of
Host: Paul Barnhurst (25:54):
Our strategy. Yeah, I know Joe. I met Joe few times at the modeling World Cup. He's been there last few
Guest: Brian Jones (25:59):
Years. Oh great. Yeah, he's our thought leader in terms of how we're evolving the Excel formula language. He's the one that helped bring things like dynamic arrays and lambdas and all of that stuff. Sometimes when he and I would be talking through like, okay, where are the limitations of the formula, language formula, I always wondered like, Hey, this is when we were working on the Python stuff, if we were going to get to a point where people would discover that and then just have those things where the formulas just it's too hard of a formula. They're getting stuck and lemme just switch over for this one straight python. I can just go and do a web search and find a quick Python script and put it in. But I haven't seen it a tonne yet. I see it more like you're saying as more for the, I'm analysing a large set of data. I'm passing in a data frame, then I'm getting a chart out or things like that. But it is something that if you're proficient in both, you could easily flip back and forth when you're building out your spreadsheet.
Host: Paul Barnhurst (26:53):
I think there's a couple things that probably have limited how far people have gone with Python. Actually, I say one big thing, it's just a whole AI world because now so many people are focusing on trying to figure out what they can do with AI and where they should do it that they may not have spent as much time on Python. If you look at the last couple years, Excel has just continued to develop and change quicker than I would say anytime in its history.
Guest: Brian Jones (27:16):
Yeah, it is hard to keep up for sure.
Host: Paul Barnhurst (27:18):
I mean I'm a Microsoft MVPI do training and I still often feel like I'm behind like, oh yeah, you can do that. Oh yeah, there's this, there's that, and nobody knows everything as you well know, there's just way too much in Excel. Everybody has to focus on a few things. You can learn well or yeah, you could try to learn everything a little bit, but you're going an inch deep.
Guest: Brian Jones (27:40):
Yeah, I mean that's a challenge we have just within our team is how much do we break things out by areas and let people just focus as opposed to try and make sure people are aware of the broader thing and there are certain parts of the product where we are moving much faster. And then how do you make sure that the rest of the team is aware of those advances? Clearly, like the calc engine, we're going to move a bit slower on and keep super. That's not the place where we just try random new things and if it breaks, it breaks. Right?
Host: Paul Barnhurst (28:07):
Bad financial models can lead to bad decisions or worse. So how do you minimise the risk of a bad model? You make sure the models you build are great Financial modeling Institute developed the Advanced Financial Modeler 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 a FMI accredited podcast listeners, save 15% on the AFM programme. Just use code podcast at http://www.fminstitute.com/podcast.
(29:27):
The group by function reverted to having to use a lambda for every other function. It it was only there for a day, but some would not work if I just typed some the Lambda ETA, I don't know what caused it. I didn't report it and a day or two later I was testing it fine, but it was of course, it was the day I was trying to prepare a file to teach everybody on group buy and I couldn't use some for any of my examples. Nobody's going to understand why am I using a lambda here and not on every other one. So I kept, well,
Guest: Brian Jones (30:00):
I'm sorry about that.
Host: Paul Barnhurst (30:00):
Kept average is by example. It was pretty funny.
Guest: Brian Jones (30:03):
I did not realise that that happened.
Host: Paul Barnhurst (30:06):
Like I said, it was fixed, I think within 24, 48 hours. It was really quickly corrected because I'm going to my partner that we're training on, I'm like, this has to be a mistake. And he's like, yeah, it's not, and I had the beta version that I was working on, so it was kind of a fun moment. So I get it. You have to be really careful there. You don't want to release something like that. Everybody's like, what's going on?
Guest: Brian Jones (30:26):
Yeah, no, absolutely. Like you're saying, workbooks need to recalc the same way every time, right? That's pretty fundamental. There's other places in terms of experiences over the top where you could say maybe that one we can move a bit faster on, but yeah, there's core parts of the platform that just need to be super, super robust.
Host: Paul Barnhurst (30:43):
No, I totally agree. So I see this argument a lot. Hey, we have AI now we don't need to learn formulas. What's your take on that?
Guest: Brian Jones (30:51):
It's interesting. It's like, it's funny how much the spreadsheet world is very similar to the coding world on some of these conversations because the same thing is happening in the coding world right now where people are saying, wow, do I need to look at the underlying code? And there's all these debates, some folks are like, Hey, it's still good for you to have a deep understanding just so that you can understand, even if you're not going to read all the code, you actually can understand what the models have built. Especially when it comes to the overall architecture, you kind of want to know what the constraints are, but there's other folks that think, oh, I don't need to look at it again. So I think we'll probably see the same thing. I think there's times that folks want to be able to audit and understand the formula.
(31:33):
What does it do? I do find that the descriptions that copilot gives you when you ask what does this formula do are pretty good. There's times that I'm kind of struggling, especially those bigger, more complex formulas. Now of course there's a lot of folks that best practise, you never write of a complex formula, break it out once you get to that point, which again makes it much easier to go and inspect and understand, but when I do those cases where I do have a big complex formula, I find the AI does a real good job helping out. So I think that there'll be the set of folks that still just want to go and understand, but I think there will be a lot of people that assuming the agent can provide the right explanations and maybe even debug ability where you can go and walk through and see, well, okay, if these numbers change, what would the results be? Okay, I can see how it's flowing. I buy the way that this thing was built, so I dunno, what do you think?
Host: Paul Barnhurst (32:25):
I come at a lot from the modeling side and I tell people, look, you still need to know how to model. I think you still, you need to know the functions, especially in finance, outside of finance, more general. I get where you're coming from, but I tell most finance people, you need to know Excel and modeling better, not less because your boss is going to expect you to be able to explain everything, to know all of the assumptions. I find that AI sometimes uses more complex formulas, not less. It does sometimes.
Guest: Brian Jones (32:53):
Yeah,
Host: Paul Barnhurst (32:54):
I've seen that a lot. And so you need to be able to understand what's doing If you need to edit it, what if you're in the room and you don't have copilot or your boss is expecting you to update something on the fly and you don't know what you're doing. So I say to people, you should spend more time really learning Excel and modeling not less, and then use AI to augment that and accomplish even more. My big saying I say is AI is a magnifier. If you know what you're doing, it's going to magnify that. If you don't, it's going to magnify that as well. Just a question away,
Guest: Brian Jones (33:26):
The thing I like, I do this just all the time for work. I always have a few agent sessions going outside of Excel just to help me as a thought partner. I'm sure a lot of people do this right? Hey, I want to write a strategy on this. Here's what I'm currently thinking. Go start asking me a bunch of questions to help me kind of think through it better, those type of experiences. And I think the same would be the case in Excel, right? Where it really does, back to the thing around the formula explanation, I find we all have this where somebody sends you a spreadsheet and you're like, what in the world is this thing? Where am I? It doesn't map to my mental way. I would've thought through logically not quite as bad as if somebody shares their notes with you what their handwritten notes, but sometimes it's kind of close to that and so I found that just being able to go and have that conversation with copilot like, Hey, what's this workbook about?
(34:19):
How does it flow and start? And I can start exploring more and then I can ask questions, but it's almost like that coworker that, sorry to use the same term that is there helping you understand maybe even start to model things out like, hey, if I changed this, what would happen? I have that a lot where it's, I mean again, back to Excel being like a development platform by that analogy you'd say every spreadsheet is an app. A lot of apps when you first boot them, they have this really nice getting started experience to tell you how to use the app and where to go and all of that stuff. Excel doesn't have that. You should probably build it. I wanted to build that for a while, but right now you just are thrown right into it and you're like figure it out. I don't quite, hopefully they did formatting in a decent enough way know where the assumptions are and inputs and all of that stuff, but it might not have been done that way and now I have to spend time figuring that out and I find Copilot does a really good job helping you with that.
Host: Paul Barnhurst (35:15):
Yeah, I've used several agents now as they've got better. I do a deferred revenue schedule and I said, Hey, give me instructions and build out a sheet and tell me what all the formulas are doing and it does a good job there. Outside of a few disciplines, I'd say most people are not good with documentation in Excel and so that's an area I've been saying since the beginning. Use AI to help you with your documentation, simplify the project. You might have to clean it up some. It may not be a hundred percent right, but it's a lot better than nothing.
Guest: Brian Jones (35:40):
Yeah, absolutely. I do that now sometimes where I'll even ask it just to put comments in certain places and so then I can go over and see the comments that it left and then once I agree I can just resolve those.
Host: Paul Barnhurst (35:51):
Hadn't thought of that, but that's a good idea as well. I have a few kind of questions I want to ask a little broader than just Excel, but kind of some standard questions we ask. Before I do that, what's the advice you'd give to our audience to be getting the most out of copilot? These are mostly modelers. What advice would you give?
Guest: Brian Jones (36:07):
I mean the biggest thing, which is my fault and I'm sorry for this, is just make sure that you actually have the latest, right? So make sure you've got what we're currently calling agent mode. The easiest way to know that is just ask it to go and start working on the workbook and start making changes. And if it does some analysis and shows you things but says, sorry, I can't put those into the workbook, that means you're not on the latest. That means you're on that kind of version two that I talked about. So I mean that's really the biggest thing and sometimes if your IT department hasn't rolled out the latest in terms of the desktop app, sometimes you can just go and use the web application that'll usually have the latest and greatest and our web app is quite capable at this point.
(36:48):
Obviously for advanced financial modelers, we've done a lot of work, for instance, to try and get keyboard shortcuts the same. Not a hundred percent though, right? Still it's pretty decent, but there's a lot of things where if you're super used to the desktop app then you might still have some things wanting in a web app, but it's pretty decent, especially it you want to just start seeing what copilot is capable of. So I say that's the first and then the next one is just explore, try things in different ways. It's not like the old school, I click this button, it does a thing. As we know with all of these tools, very different approaches to how you ask a question can have pretty incredibly different results, right?
Host: Paul Barnhurst (37:29):
Yeah. It's still something we're all getting used to, right? If you've worked in Excel, you're used to two plus two always equals four, right? Or this function always gives, if I put these inputs, it's going to give the exact same output every time. That's no longer the case for certain things.
Guest: Brian Jones (37:44):
One of the many reasons why I'm really excited about the role Excel plays in all of this is that even if the model itself, sorry, the model meaning the language model kind of does unpredictable things. You ask it to do some analysis and it gives you an answer. The thing that's great about Excel is it will go on show its work in the spreadsheet, so it's audible verifiable. Whereas if you're just in a standalone chat app and you ask to do analysis and it gives you an answer, you're not really quite sure exactly how do they get to that and is that something you can trust? Maybe there was some assumption that they made that you weren't aware of, but at least when it does it in Excel it's using formulas, it's using the grid, right? It's using Excel to go and do that analysis. And then it's also using Excel then as a communication tool with you so you can see how did it arrive at that.
Host: Paul Barnhurst (38:33):
That's a great point. Alright, so I'm going to ask a couple fun questions here. First one, what's your favourite Excel shortcut?
Guest: Brian Jones (38:40):
Oh my goodness, this is kind of embarrassing, but it's just control T to turn data into a table
Host: Paul Barnhurst (38:47):
That's high on my list. It's one of my favourite. Yeah,
Guest: Brian Jones (38:50):
I think it's because of my role, the way I use Excel, it's a lot for managing the team and managing projects and things like that. So a lot of my Excel use cases are still more like tracking scenarios. So I use tables quite a bit.
Host: Paul Barnhurst (39:04):
Not surprised. Favourite function,
Guest: Brian Jones (39:06):
I think it's probably unique equals unique. I mean all of the array functions I can't imagine living without, but there's a lot of things where I used to have to use pivot tables and it wasn't really what I wanted. I wasn't trying to do exploration, I was really just trying to lay out summary information. And so I love using unique for back to my scenarios where it's tracking stuff. Let me really quickly get a list of what are all the projects that the team has and so everybody's filled out what the projects, the work item and the project it lines up to and then I can really quickly just say equals unique and then I got the list of what all the projects are and somehow a new project showed up that I wasn't aware of. Now I see that right? It I love using it. So
Host: Paul Barnhurst (39:47):
When do we get a unique ifs?
Guest: Brian Jones (39:51):
Can you compose that? Maybe you could do it with a combination of filter.
Host: Paul Barnhurst (39:54):
We can do it with filter. That's what most people do. You do a combination of unique and filter.
Guest: Brian Jones (39:58):
I think that was the thing where I used to have people would ask for new ifs like Max if or things like that. And it was like as soon as we did filter, then you had the out where you're like, I just do a filter first and then wrap that.
Host: Paul Barnhurst (40:09):
Yeah, for sure. Alright, so totally nothing to do with Excel, but I know we just had the Super Bowl a month ago. When you're a Seahawks fan, what was it like watching him win?
Guest: Brian Jones (40:20):
It was amazing. I went to it with my son, so this is the fourth Super Bowl, but I'd never gone in person and I always liked when we went, the second wave went two years in a row. It kind of felt like, oh, we're going to do this a lot. The team seemed pretty great and so I was like, oh, I'll just go to another one. Then there's this long dry spell and so this time I was like, oh man, I got to go. My son. I don't know, by the next time my son might be a full grown adult so I got to take advantage of now the game. I'd say that the game before was probably the more challenging game against the Rams for us. That was a lot more tense, but it was nice just being able to enjoy it and not having being too tense.
Host: Paul Barnhurst (40:55):
Well awesome. I'm glad you're able to take your son and when you said a long time in between, I'm a Titans fan so don't even give me long, but, and a jazz fan, so I don't have many titles.
Guest: Brian Jones (41:05):
Were you an Oilers fan originally?
Host: Paul Barnhurst (41:07):
I was, yes. I was a Warren Moon fan as a kid.
Guest: Brian Jones (41:09):
Oh, I went to the University of Washington, so there
Host: Paul Barnhurst (41:12):
You go. Big
Guest: Brian Jones (41:12):
Warren Moon fan. Yeah, he even came back. He played quarterback for the Seahawks at the end of his career.
Host: Paul Barnhurst (41:17):
Yeah, I vaguely remember that. He was there for a little while. Alright, well I appreciate you joining me. I don't want to keep you any longer, but last question. If people maybe want to get in touch or see the things you do, what's the best way for them to follow you?
Guest: Brian Jones (41:31):
I'm on LinkedIn. I don't know what the URL would be, but Brian Jones Excel would probably find it. So you can always just message me through the outer follow me there. I've also off and on posted on X. I'm trying to build a little bit more of a habit around posting there too. I think out there I'm just like Jones 2 0 6 Seattle area code. So either one of those.
Host: Paul Barnhurst (41:51):
Right. Perfect. Well thank you for taking some time today, Brian. I really enjoyed chatting with you and I'm excited to see how Excel continues to develop in the AI era. And I will say anyone who says Excel is dead is not paying attention. So thank you for the work you're doing. We really appreciate it. Yeah,
Guest: Brian Jones (42:07):
Thanks. Paul
Host: Paul Barnhurst (42:08):
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 enroll in one of the accreditations today.