#289 The AI Skills Gap Every Finance Leader Needs to Close with Guy Weaver GrowCFO Facilitator

Kevin Appleby: The fear of everybody as AI started to take over was oh, this is going to take our jobs away. We're now talking about very much different jobs that never existed before AI. What are the skills that people should be looking to gain?

Guy Weaver: It's more around the mindset shift in employees. The best uh, AI automations and benefits within an organization will come from the employees. Accountants jobs are dead. CFO jobs are dead because Claude can build you a uh, financial model. But then if you ask Claude can build uh, an integrated financial model, it'll say no.

Kevin Appleby: Grow CFO is where finance leaders grow together. Join thousands of like minded professionals using Grow CFO to access the combined knowledge and experience of the finance leader community. You can join us today growcfo.net hello and welcome to the Grow CFO Show. I'm your host Kevin Appleby. And uh, today we've got one of our training team in Grow CFO with us, Guy Weaver, who, who specializes in training in A.I. a uh, very, very hot topic at the moment. So Guy, welcome to the Grow CFO show.

Guy Weaver: Great to be here Guy.

Kevin Appleby: Tell us a little bit about you before we get into training in AI.

Guy Weaver: Okay, so my background is I uh, am a chartered accountant. I trained at KPMG quite a long time ago now. I spent 19 years at KPMG in various functions. So probably like everyone start off in audit and then work out what I want to do and then transitioned into corporate finance doing M and A through to valuations and then through to debt advisory. And then it's probably just over 7 years ago I was offered a job to join Protora to launch Prora Ventures which is at the time there was no venture capital firm based in Manchester. So joined the team. There was a bit of a baptism of fire because there was probably six of us in a uh, what used to be a pool room with wires everywhere, dark room working out how you build a venture capital firm. And I suppose my interest in AI came from probably at the start of 2025 I put in Copilot into Pro Ventures. I was looking at Copilot at the time because I had that many teams channels, that many teams messages coming through and emails because had a portfolio of 60 businesses to manage as I was a uh, portfolio director at the time. And I was thinking, oh, Copilot can help me do this, help me identify where I'm getting good feedback, where I'm not getting so good, not so good feedback. And I remember I just couldn't get Copilot agent to do what I wanted it to do. And I remember speaking to the IT service desk at the time and they said no one's asking about this, no one's really doing Copilot agents, we don't have any training material. So I kind of continued to play around, learn on YouTube. And then I guess the turning point for me around AI was at uh, the start of June last year I was made redundant from Proctora. So I was put in the garden for six months and probably instead of spending my time learning how to garden I actually thought right, I'll learn how to use AI. Because at the time I thought it'd just be an added skill for my CV for when I look to get back into the workforce. And I kind of got sucked in. It turned into a few hours of training on courses through to probably a couple of hundred hours through to building things with agents, projects, automations, using the API keys, probably into agentic AI for some of the functions. And really my knowledge grew over that six month period really focused um, on using AI in a business sense. So I'd have to do a lot of training courses because they were very generic, going through all the various tools, not really focused to business and then apply it in a business way with what I was trying to build myself. And I guess I was lucky because I didn't have a job at the time. So I had the benefit of just having a Google account and linking up things quickly and tying things together and I guess my experience is built from there. And since I was coming to the end of gardening leave, what I've ended up doing is building up my own community around fundraising which without AI I would never have been able to build a uh, website. And I've learned that with CLAUDE particularly I can knock up new web pages within the matter of minutes, copy the code in and up its live on the website through to lots of different functionalities to really help that part of my business. But then spearing from that is I'm now going out and talking to businesses around how they can use Copilot, how they get the most from AI, from the foundations through to more advanced techniques through to architecting it, being aware of context drift and then also some of the free tools that are out there which are actually better than Copilot or claude. And actually that then plays into what is the data security policy of companies because a lot of companies will often have everything has to go through Copilot. But actually employees are doing shadow AI, uh anyway on a ChatGPT or a Claude on their phone. So is the better Approach actually anything confidential has to go within the enterprise system. But if it's research where it's not linked back to your company, is it then better to give the employees the freedom to use other free tools that are out there, uh, which could give you a far better response. So kind of in a nutshell that's what I'm doing at the moment. Kind of have my own business but also going out teaching companies about AI. And what I love is just seeing the benefit instantly from at the end of the sessions, companies spinning up copilot agents to automate various things. In the last session I did in HR and finance before I've even got home and after I got home I've opened my inbox and I've got an email from the CFO saying we've automated that task, we're moving on to the next one. And for me it's just really rewarding talking through the knowledge what I've learned and actually seeing the real power it can deliver to businesses.

Kevin Appleby: That is fantastic. And I've been involved in finance transformation for many years and I'm seeing something that is truly game changing at the moment. You can streamline processes overnight, processes that you used to take months and months and months to think through and change, speed of changes. Definitely one of the big impacts I'm seeing. But no, this is needing a whole new set of skills right across the finance team. What are the skills that people should be looking to gain? Guy?

Guy Weaver: Well, for me I'll start with the basics. I mean my big thing that I tell companies when I'm um, going in and agreeing to do training is I think it's more around the mindset, mindset shift in employees because if you can show them um, how to use AI and the art of the possible and its practical sessions where you're going through building agents or projects or architecting tasks. Actually suddenly you can see a change in the employees in terms of the mindset. Start to think about what are the potential possibilities. And I think that's probably the biggest thing with AI, it's um, unlocking that potential because I do believe the best AI automations and benefits, certainly the low hanging fruit within an organization will come from the employees, the ideas from the employees. Once they understand what AI can do, you'll have those solutions and opportunities presented to you from your employees.

Kevin Appleby: As interesting as that's always been my approach around finance transformation across the years, the people that will tell you what is wrong with a process are the people that are doing it every day. So get them together in a big darkened room and brainstorm with them and prioritize what's wrong. The useful thing now is if we start telling AI what's wrong, AI, uh, can start helping us fix it.

Guy Weaver: It's absolutely. And I think the slight difference with AI, I think to other tech transformations that have come along is if you learn how to use AI in a business sense, it also helps you in a personal sense with your AI at home, which actually then if you're using AI at home and getting better at it, helps you in a business sense. But also, AI is moving so fast. I mean, before Christmas, ChatGPT was my go to platform, then Claude, then Claude's reduced the token limits and then ChatGPT's released the 55 and Codex, and I kind of now moving more onto Codex. And actually what I'm seeing is the platforms and functionalities are, uh, continuing to evolve, but actually the way you interact with them is exactly the same across all the platforms apart, uh, from maybe image generation, like midjourney, where you need a slightly different prompt and input, but they're all basically the same. You need to know the foundations of Persona, good, uh, prompt structure, how to architect things, not overload it. And you'll get the same result from, or really improve result from, no matter which platform you use. And I always tell companies I'm going into, you've just got to think about AI as an employee. So if you hire a new employee, you'd sit down with them, um, explain what their role is, what your role is, give them the context around the organization. Even when you look to give a task to an employee, you spend time briefing them, and you'd probably want to see part of the task done before you give them the next part of the project to do. And that's exactly, I think, how you should treat AI. Treat it like an employee. Pick the right employee for the task, which is the right AI tool for the task.

Kevin Appleby: So the biggest single skill is learning how to prompt AI, I think so

Guy Weaver: I think it's understanding, prompting, understanding architecture, understanding how you link things together, which will play into agentic, but then it morphs into context. So I think if you get the basics right, you'll get an uplift in your skills. And then it all starts to come into context, having good context files, which then plays into how organizations, I think, will have to change so to keep context files up to date. Because if they go stale, your AI tools won't be as good as what they were.

Kevin Appleby: That's an interesting thing, is I Know that AI will forget things. If you told it something 10 conversations ago, you can't necessarily expect it to remember that thing in the conversation you're currently having. And uh, I think learning all about context and how to create sort of ongoing documents in there that it always refers to, ah, are things that folk have to learn about because it can be incredibly powerful.

Guy Weaver: Absolutely. And I guess I'm continuing to learn all the time, going on the training courses and actually I'm learning from developers in terms of their techniques because I think AI, the coding tools like Codex, Claw Code, Claude Code are kind of were built for developers and they were using it in a slightly different way to business. But actually the two are coming together and when you talk to developers they've got tools like Obsidian, uh, which manage all the context files, which continually update on a daily basis. If you look at certain contexts and certain firms that have put AI throughout the whole firm probably are moving towards a native AI firm. Actually what they've got is context files at the top, almost the commandments for the company. And it's how you then manage those context files that drive the whole AI system throughout the organization. Keep it live, keep it relevant, but also who signs off on the updates. Because if you get the update wrong, that's going to be feeding through the organization and the implications, uh, could be quite stark. I think there's a change coming with companies. Once AI is adopted, embedded within businesses, it'll move more towards context file management to really get the most out of AI tools. But it's how you then manage that. Is that the job of a CTO or is it the job of a, uh, coo? Because actually the context files are both used for tech and used for people and agents. Yeah.

Kevin Appleby: Ah, and I wonder if the context files actually, uh, they're not the response. The CTO is creating the infrastructure, but the context files themselves. Is it the COO by themselves or is it the CFOs creating some, the COOs creating some, the uh, head of marketing's creating some. Because surely what you want is the context dropping down to a much more detailed level.

Guy Weaver: Yes, you're absolutely right, you will have that. But you'd have a. I think it's the hierarchy of how it's set up and how it's pulled together because the master contact will have to feed into the various tiers and it's then how you start to look at that within an organization of who's signing off what, who's maintaining what, as well as, I suppose the HR function, Are they going to be managing people as well as agents? Do agents need to go through quarterly appraisals every quarter to tell them what they're good at and then refine what they're doing to optimize them, just like employees. There's various things that I think are, uh, going to be coming through in the next couple of years.

Kevin Appleby: This is really interesting here, Guy, because the fear of everybody is that AI started take hold, was, oh, this is going to take our jobs away. We're now talking about very much different jobs that never existed before AI or certainly different responsibilities that the human in the loop is going to have to do.

Guy Weaver: Absolutely. And I think it's those that are adopting AI and getting the benefits early. What they're actually finding is not only are they getting the cost saving, but they're using AI to find other opportunities and grow revenue because it's adopted throughout the business. I mean, Boston Consulting Group, PwC, recent reports in April were, uh, flagging this around AI and the adoption and also flagging the risk of the capability gap between those that are adopting AI and really starting to motor around how they're using AI to the companies that I guess are yet to get started or still on that learning curve. And I think companies are on various stages of that learning curve. And I suppose the other challenge now coming out, whether it'll play out is actually the cost to serve of the AI tools. So we've already seen it this year in terms of claude, which I almost see it as they gave out free cigarettes and then started charging for them when they reduced the token limit for the subscriptions. And certainly if you look at the opus 4.7, if I run that, it completely wipes me out of tokens from that. Uh, and I need to wait for a reset. And you've got ChatGPT's new model, which is 20% more expensive than the latest CLAUDE model. And then the. You've got all the noise at the moment about Claude mythos. Is it really not being released because of data security fears, or is it because if they release it, they can't service it? In terms of the demand for tokens, there isn't enough COMPUTE power out there. And I think what we're going to see is the end of, I guess, AI subsidies, pricing going up, pricing being for usage of tokens, certain models and companies are going to have to grapple with that. And it could be, in some cases actually AI becomes the same cost as having an employee for certain use cases. And the next phase I think on once you go beyond training, understanding how you're deploying AI, it will move into optimization. Just like tech businesses we're doing with optimizing AWS costs and Azure hosting costs. It's going to be exactly the same with AI. How do you optimize the cost? So it's about understanding how you use less tokens when you put the prompt in, how you use less tokens with various attachments, context files, which model you use, how you specify the output. So actually you're using the right model with the right prompts with the right context files. So actually you're not burning through tokens which will be a cost implication for the company. And again I think that's going to be a challenge that companies and leaderships are going to have to get their heads around because I don't think they'll be able to get say a uh, Microsoft platform and give every employee the best possible model that is available on Microsoft because this costs will be significant in terms of opex.

Kevin Appleby: So Guy, we've started talking about some quite sophisticated stuff, some quite sophisticated new roles that are going to emerge. But let's just wind back to where we are today. You're going and training folk in this, you're going into companies and training the finance team and wider, uh, generally what's the level of experience out there that you're finding? Where do you generally find people are at when you first meet them?

Guy Weaver: I think it's fairly mixed. I've been in some interesting training sessions where people are just getting started through to others that are actually quite competent using it. And in some sessions I've had new recruits that ah, have actually done AI as a degree which has been quite interesting in doing introductions through the table. But I think you do the sessions, everyone learns something from it because particularly when you start talking about what the various free tools that are out there, uh, or tools that the company might want to consider as a subscription. But I need to kind of be a bit light on my feet around kind of what I present, what I demonstrate. So actually everyone in the sessions gets the most they can from it. So what I try to do is the sessions where it's practical look at probably 8 to 10 people max and then really I've got a slide deck I talk through. Well, uh, mainly just to keep me on the agenda because I get excited about AI as you can probably tell, to make sure I cover everything and all the tips that I've learned. For example one that a lot of people miss is within the AI tool, you've got temporary chat, so you'll have it within copilot a chatgpt, it's called Incognito in Claude. And a lot of my time in AI, I didn't really understand what was the point of that. Yeah, you can do a chat and it, it disappears. But actually one of the things that I find it really useful for and I recommend everyone to use this is the AI tool that you're using is building up your profile, it's understanding all about you, what you like, et cetera. So it's kind of got a little bit of a bias because it's trained, it's learning and that's the power of AI. So if you're creating a report and you want someone to sense check it and maybe there isn't an employee round that you'd normally give it to review that report, why not put it into a temporary chat or incognito? Because what you're doing there is you're giving it to an AI chat tool that has no background information on you or history or bias and it's given a clean review or output. And it's quite interesting how even if you do it, you create the report in claude, but using cognito in Claude, actually you get some really good inputs about how you can make that document better that your normal CLAUDE chart just didn't give you.

Kevin Appleby: Interesting. Let's have a look at that. I've never used Incognito.

Guy Weaver: Yeah, it's one of them. There's so much in these tools that you just, they're adding all the time. Well, that one's been there up time and you do wonder why it's there as a, uh, like a temporary chat. You do a chat but then you lose it from your history. But that's one of the use cases I found for it.

Kevin Appleby: Yeah. So guy, you're teaching this stuff, you're there in the classroom, engaging these small groups now, how fast is this changing? And I uh, see that potentially as the biggest challenge in training folk in AI. Is that what you taught them last week is going to be out the date this week?

Guy Weaver: Yes. I mean I've got a week of training next week and this morning I've been going through updating the slides for the latest thinking or the latest thinking. I've learned what's happening in the market and actually for uh, some of the training, I've already given it to some of the employees at this company and I'm actually having to email them do top up training. So Actually they're still getting the up to date training whether that for example could be building skills within Claude because they're quite useful or what I'm finding now is I'm actually building Copilot agents with various instructions and files that I'm basically sending to the clients I'm working with to spin up within their enterprise system because I can then use that to help the team share examples of the training, build context files relevant for them and really speed up how quickly they get up the curb in terms of AI use AI training and get really get the benefit from it. So that's something that I've started doing is whilst I don't have a Copilot enterprise license, the way you build up agents within copilot is exactly the same as you build up custom GPTs within chatgpt so I can spin things up on um, my system and just take those instructions and just spin them up when in the clients I'm going to. It is evolving all the time Like I said what I'm liking at the moment is I like the new ChatGPT model I find it better for certain things than Claude but it does take you down rabbit holes because it always suggests at the end can I do this for you next? Can I do that for you next? But then what I am liking is Codex in terms of being able to plug in various systems, do automations and really crunch data really well which Claude can do in certain aspects but what I found is I'm m not burning through as Many tokens within ChatGPT or I don't think I've ever hit my limits within Chat GPT on a daily basis apart from using deep research.

Kevin Appleby: Now a couple of months ago there seemed to be a game changer from Claude of finance came out. Ah, is that still streets ahead for finance people that it seemed to be on the day it was launched?

Guy Weaver: I think it is. I think people are still adopting it. I think it's interesting a couple of the AI sessions I've done recently I've gone into companies and the finance teams have been there and we've talked through Microsoft Copilot and I've demonstrated Claude the capabilities Claude has. They've actually gone away the finance teams and bought an enterprise license for Claude particularly on one of the automations we were going through with the finance team M We built the automation within agents within copilot but actually it wouldn't give us the output in a PDF that we needed to put into another finance system But Claude could do that and Then you could schedule the tasks within Claude. But I think very much Claude is ahead. But I think the challenge we've got at the moment is ChatGPT or OpenAI have and what we've got coming that is in testing at the moment is Copilot Cowork alongside Anthropic. So I expect that'll be released shortly and probably be another game changer. So all the people that have moved to Claude might be then looking back and going, well Copilot suddenly leaped ahead and I think we're in this world where that's just going to continue to happen, where one products better at something than another. And what I'm saying to companies is just get good at using the tool that you've got at the moment. Move yourself from a 20, 25 or 30% usage across your employees to a uh, 60, 70% and once you're at that point then you can look at do you need to swap in Claude because you're already maximizing the benefit of say Copilot. Yeah.

Kevin Appleby: And in a way, okay, you might not at the moment be using the tool that's top of the leaderboard, but the next release is just around the corner so you might be next week.

Guy Weaver: That is the challenge. I mean Claude is very good certainly at Excel and you look at LinkedIn and there's lots of people going, accountants jobs are dead. CFO jobs are dead. Because CLAUDE can build you a uh, financial model. But then if you ask Claude can it build an integrated financial model, it'll say no. It can build the template framework but you need an accountant to really understand it and actually check it. Because I find that our uh, mistakes that do come through in claudexl and actually you need the knowledge of accounting, how things play in. And every financial set of accounts or models are slightly different because of the underlying business. And that knowledge you need to input into say CLAUDE within Excel to ensure you do get the model you want. Whereas I'm kind of looking at these LinkedIn posts at these people that are building financial models with no accounting expertise and I'm thinking, God, those models, they're not going to be the greatest. Certainly those companies go and go for investment, they're going to get ripped apart.

Kevin Appleby: Uh, I've always taken the view when you're building a uh, model that's looking at specific stuff, the model is only as good as the assumptions you've got. Now the assumptions are, I uh, remember years ago the Ministry of Defence was the client had to build a whole life costing model for a particular missile system they were introducing before it was adopted. And we had to put all sorts of assumptions into that model. We had to expected life of these missiles before maintenance, expect what happens if you change the storage conditions, how many are going to be able to fire off on the training range, things like that. All sorts of assumptions went in. And I just think it's great that if you can have AI taking care of the formulas within the model, you as the finance person can do what you really should be doing. And that's sitting with the rest of the business talking about the business assumptions and getting those right and spending much more time. If I was reviewing a model like that, I wouldn't take them through all of the calculations in the model and show them the numbers flowing through. I'd take them back to the list of assumptions and say we've assumed this, this, this and this are ah, these assumptions, right?

Guy Weaver: Absolutely. And I think that is the assumption. And I was also taught how to build a financial model at kpmg. I remember the model builder was someone who built the financial model for the Channel Tunnel and it was probably three days of going through how to structure it where you have one tab for uh, outputs, another tab for uh, assumptions, another tab for uh, calculations. You never mix the two and then you always try and separate everything as much as possible so you don't have complex formulas so you can quickly spot the mistakes because there's often mistakes in models. And actually what I did find when using Copilot in Excel, it would come up with these formulas that I've never even seen before suggesting to put into Excel. And I think the challenge with that is a lot of financial models, yes, are owned by the finance function, but if you then need to go and raise capital, someone else is reviewing them. Can they review them or can someone else in the business review them? M if the formulas are so complex rather than breaking out the assumptions. I think we've come a long way since that. I think that was probably Copilot last year when I was also playing around with Python within Copilot because you could spin out Python as an analysis tool within Copilot, which again I found quite clunky and found it actually it was far quicker to just build array tables and vlookups and hlookups myself to do the analysis analysis than actually using Python. But I think it's coming a long way and I think it'll continue to evolve. And the skills you need to keep up with it are uh, just the foundational skills to get the most out of AI today and the best way to continue to get fresh is, or continue to stay up to speed is just being quisitive and playing around with the new functionality, the new tools, the art of the possible, because that's where the really valuable training comes from.

Kevin Appleby: Yeah, I can see that the training should be there to say no, uh, don't let AI kind of take over and do everything. You've got to keep those traditional structures you'd have in financial models. You've got to keep all that financial modeling best practice and then apply AI within that danger is that somebody now can pick up AI with very little knowledge of that best practice and produce a financial model with AI. That's a bit of a Frankenstein.

Guy Weaver: Yes, I think that will be coming to a lot of venture capital firms in due course where certain companies, early stage businesses, they never want to hire finance uh, directors, head of finance, CFOs because they don't see the value of the role in there. I've seen it and we've often had the argument you need a finance person in a business and they're just going to do it themselves and they're going to mess it up. And most people in venture capital probably have an accountancy background and will rip the model to shreds. I think we're going to start to see that coming through and people are going to learn from it.

Kevin Appleby: So Guy, you've been very much self taught on this, the journey you've gone through, but you had the nice luxury of six months off work on gardening leave to do it. Now if you are an individual in a finance function, can you self teach yourself on this or do you need to go on courses and have experts around you so you can get up to speed at the right rate?

Guy Weaver: I think going on doing courses certainly helped me get started. It helped me understand how I should be using AI and AI tools. And there's a lot of online courses out there for different specialties. Probably the courses I did were very generic. So I'd recommend looking at finance AI courses to really upskill you because once you've got the basic skills around Persona and good prompting structure and learnt architecture, you can apply them to so much more. And then really where you get the learning from is just keeping track of what are the different possibilities that coming from the new models or what are the people using AI for and starting to play around. And can you use them within your business? Because as you're playing around, I use the term playing around with the AI tools and iterating them, that's where you learn. So some of the best tools I've built, actually the first prompts or structuring of an automation or a project or an agent I've built have not been that good. They've been okay. And it's actually taken weeks of effort to just continually refine them to get to the output I wanted. But, and that's really just iterating and playing around with am I using the right model, am I using the right tool? Have I got the right context files behind it is the prompt right? There's a lot to think about and it's really just experimenting, playing around with different structures and really that'll upskill you. I think that's the best way to learn is just diving in and playing around with all the different functionality that's coming from these AI tools.

Kevin Appleby: Brilliant. So we've covered a lot, Guy. I'm conscious the time's going on now. Uh, I dare say we've tweaked a lot of people's interest in wanting to upskill in this area. So look, going forward, come across to growcfo.net, have a look at the AI courses we've got on offer in Grow CFO. There's some self paced learning there in the Grow CFO Academy. There are, ah, live courses that we're running. Guy is one of the tutors on many of these courses. But we've got a team of experts now. If you're also looking to upscale your whole team, well, Guy is doing that himself a lot of the time. I know Guy, you're going directly with some clients and putting together big, decent training packages on this. Lots of ways that we can help you get up to speed in AI. It's the new absolutely essential skill in finance. And the faster we learn it, the better off we are. So Guy, thank you for being this week's guest on the Grow CFO show.

Guy Weaver: Thank you. Good to be here.