Critical AI Prompting Techniques for CFOs

Megan - 0:51: Today, my guest is Nicolas Boucher. Nicolas is a finance thought leader specializing in AI who's trained more than 5,000 people and is on a mission to help more than 1,000,000 finance professionals. He teaches finance people how to use AI. Nicolas, thank you very much for being my guest on today's episode of CFO Weekly.

Nicolas - 1:12: Hi, Megan. How are you doing?

Megan - 1:14: I'm great. And, again, thank you so much for taking the time to be here with us today. Our discussion focuses on AI for finance, which is probably the hottest topic in all of finance and accounting at the moment. And these are truly exciting times to be in finance, and I'm looking forward to learning about you and this topic. So let's jump right in.

Nicolas - 1:36: Sure.

Megan - 1:38: So Nicolas, just to start, and so the audience can have some bit of knowledge as to what you've done and where you've come from throughout your career, can you just give us a brief overview of your accomplishments to date?

Nicolas - 1:51: I'm a finance guy first. I worked fifteen years in finance. The first half was at PwC as an auditor. I got the chance to work in Luxembourg, but also in Singapore, in Asia as a settlement. And then the second half of my career, I worked on the corporate side for a company called Thales, which is like Boeing for those who are more in North America, but European. So imagine a mini Boeing in Europe. So that's what Thales is doing. And there I was a finance leader in controlling and accounting, in FP&A. And basically, in all of this experience, I always loved to use technology and to teach people, but I felt that in this role, I could not help as many people as I wanted. I decided to try to help people on LinkedIn first, to voice it. And when AI arrived, that's where I really felt the calling to help finance teams, to teach them how to use AI for finance. And right now, it's crazy to say that, but I have a social media following of more than 1,000,000 people, mainly on LinkedIn, and then also Twitter and a newsletter. And I help more than 5,000 people through my courses to learn how to use AI for finance.

Megan - 3:08: That is amazing. So as you look back, when was it that you got started? When was it that you first saw AI make an impact? And do you remember how that made you feel and your reaction to that?

Nicolas - 3:23: Yes. So, like a lot of us who became a new AI adopter, it was when ChatGPT got published in November 2022. I was one of the first to get the license and straight away, I was playing like at Christmas when you get a gift and you cannot stop using it at night and you're just so excited about it. So I tried it for so many use cases first for myself, because I was publishing a lot of content on LinkedIn. So my brainstorming for new content just got crazy. And then quickly, I thought this can also help us in finance by drafting a lot of the documents we used to spend a lot of time on. For example, reminder letters or a revenue recognition memo. But then also I remember when I had to work in my previous experience to find cost savings, I would have loved to have this tool next to me just as a free consultant. That's where I thought, okay, I know how to use technology. I know how to use finance, but I know a lot of people don't see that yet. I know a lot of people don't see the power of ChatGPT for finance or AI for finance. And so I decided in February 2023, so about three or four months after, I published a guide on how to use ChatGPT for finance. And this guide more than three to 4,000 people downloaded it. And that's where I started to help more and more people on it.

Megan - 4:52: So at this point you've trained thousands of finance professionals on AI. And what is it that you see as the biggest knowledge gap that finance leaders face when trying to use AI?

Nicolas - 5:04: The biggest mistake is to use those LLM chatbots like Google. Google was something we are used to having for fifteen years. So we got into the habit of just searching with small keywords, but the LLM chatbot like ChatGPT or Copilot is not really a Google tool or a search tool. It's more like imagine having an intern working for you. And this intern is really, really smart. They read all of Wikipedia. They read all the books. On top of that, they can write perfect English. On top of that, they know how to code. On top of that, they know everything about Excel, about SAP. And so this assistant next to you is smart, but without your instruction, this assistant is useless. And when I teach people, I give them a framework called the CSI and FBI framework where basically CSI stands for Context, Specific, and Instructions. So imagine you need to create a reminder letter, you are not just going to type "reminder letter." Instead, you are going to say, okay, I am an accountant. So that's the context. Then your problem, so specific: I have an invoice that has not been paid for two months. And then with the 'I' of CSI, you give the instructions saying, "Please draft for me a reminder letter." And by giving that, you already get a good output. But then if you add the FBI team, that's where you have much more chance to solve your crime. Because with CSI and FBI together, normally the crime should be solved. And so with FBI on top, you add the format. So you're saying, is your reminder letter an email or maybe a footnote because you might want to call your client. Or for other contexts, it could be that you want a presentation, a table, you want code, you want Excel formulas. So the format is really important. Then 'B' is actually where you'll have the most value when you use the 'B' correctly, which is the Blueprint. So in FBI, the blueprint here in my reminder letter, I might say, "I want words like legal actions." I want the tone to be hard. I want you to have like exact legal actions that could happen. So that's the blueprint that you give to your assistant. And then the last 'I' of FBI is for Identity. Because you can give your AI chatbot an identity and here you could ask, "Write this like you are the best lawyer." And so you get a reminder letter really quickly that is of really high quality, that most probably you would not have been able to write it by yourself. And this will help you in your job because what you want at the end is to recover the money.

Megan - 7:57: I love that framework, and thank you for sharing that. And I'm curious, so for our listeners out there, and I talk to CFOs every week and many of them are still struggling to figure out where and how to learn more about what it is you offer, where can they find information?

Nicolas - 8:16: Normally they can find me easily on LinkedIn with my name, Nicolas Boucher, or what I've started to, what I've built now for one year, is the AI Finance Club, which is a community for CFOs that want to become AI CFOs. And an AI CFO is really a CFO that is already in the future, I would say, because I can see all of the members, how fast they learn, how well they progress. It's a CFO that uses AI to automate some of the tasks, that is using AI or she is using AI because we have a lot of women as well. So a lot of CFOs that are using AI also for their clients, for their board, some of them are fractional CFOs. So they use it also to get more sales to automate processes at their client. And what we did, for example, the last weeks, is starting to create agents together. In the AI Finance Club, we had an expert that came and showed us how to create AI agents. Now we have some of our AI CFOs that are starting to have their own agents, saving a lot of time, showing also to the rest of their peers and colleagues and clients, if they are fractional CFOs, that they are expert matters on AI.

Megan - 9:34: Thank you. It sounds like you guys are doing amazing things. And so with that, we're gonna dig into some details as to how AI can actually transform some of the stuff that we do on a daily basis. So when you look at AI, how do you see it transforming financial forecasting, budgeting, and strategic decision-making?

Nicolas - 9:55: This is actually something where AI was already present in finance before ChatGPT arrived. Because AI is not only a chatbot. AI is actually mainly machine learning. With machine learning, this is something that us in finance, we are doing, but let's say it's like human learning. When you prepare a forecast, most of the time you will look at the past, so historical data. And if there is seasonality in your business, you will want to reflect that also in your forecast. If in March it's a good month, because the month of March, nobody is on holidays, there are thirty-one days, then you want to show that you have higher sales. But then maybe August is a lower month because you have people on holidays, it's summer, so you have less sales, then you want to show that in your sales. Until now, as humans, we do that with some formulas or we do some plugs in our forecast, but it's not really accurate and we just do it with a lot of effort. If you give that to a machine, the machine can process much faster all of this historical data to identify the real seasonality and not only month by month but it could be also week by week, it could be also hours of the day if you are a business that is selling twenty-four hours a day and maybe with some opening hours or a website with some traffic. And then you can get this historical data and put that in an algorithm. And that's where a lot of us in finance think that is a black box, the algorithm, but it's not. Because you have a lot of algorithms that are just, I would say like something that we do in Excel anyway with some formulas. And one of them has been created by Facebook, the algorithm is called Prophet, and it's accessible to everybody. So you don't need a data scientist team, you just ask ChatGPT, "Give me the code for this algorithm," and you only put inside two values. You put your dates, so imagine you have three hundred and sixty-five days or the last three years with all of the days of each year. And then another column with just the value, could be your volume, could be your total sales, could be the hours if you are looking at productivity. And then the algorithm will show you the seasonality over the years, over the months, over the weeks. It will show you first the past to show you that with your eyes, you can already learn what the seasonality is, but also if you ask the algorithm, you can say, "Please now draft for me or forecast for me the next six months." And then you will see based on the seasonality, the forecast drafted for you. And this is something that is machine learning and where you start breaking the fear of using machine learning, you can see that it's actually something easy to use and where we don't need any more to give that to other people. And in finance, we can start doing it. So maybe as an example to make that concrete, I had in the AI Finance Club, an expert from Coca-Cola. We had an expert from Coca-Cola that built with his team the forecasting model of all of Coca-Cola locations. And what they did is just with two data scientists and two people in finance for the whole group, they created an algorithm that takes into account historical data. And with Coca-Cola, they are lucky because it's a simple product. It's basically just everybody knows what Coca-Cola is and you don't have that much complexity in the product. And then on top of this, they have external data like weather, macroeconomical data, maybe the time of the year. And just based on this, they are able to create a forecast which they call the baseline forecast. And this forecast follows the same model for all of the locations. And every month this forecast is distributed to all finance teams and it helps them make the forecast in one day and before they were doing it manually in one week and now on top of having it done in one day they have a better accuracy because the baseline is actually better than the humans. I think they increased their accuracy from 50% to 80%, something like this. On top of earning or saving a lot of time, they are also getting better at forecasting.

Megan - 14:22: That's amazing. What a powerful tool AI is. To think how far we've come and where we're heading in the next few years is very exciting. So let's take a look at risk assessment and how AI can enhance things like fraud detection, compliance, and overall risk management for an organization?

Nicolas - 14:43: I was a head of finance. I just got nominated. And they told me already before, and I accepted the challenge that three people would go into retirement. So I started with a team of eleven people and within three months, three people left. And the challenge is that we could not really face all of the workload with the team that was there without changing a lot of processes and it takes time to change processes. So one of the processes or the tasks that we left a bit on the side were all of the travel expenses, all of the reimbursement for restaurant spend. So, basically, everything that the employees needed to be reimbursed, I thought it's first important to pay our vendors and to get the money from our clients. And then for the employees, we can be a bit late. But then, of course, after one month, some of the employees started to complain, and we had no other choice than spending time on processing all of these travel expenses. And for those who already went through that, I think they can figure out and picture what we went through. So that was my first time, but I looked all of my team inside and also the FP&A people, everybody who never did that. And I said, we have all of this pile of papers and we have to process that all of us as a team and me included. And seriously, that was the worst day of my life because the worst day in finance, because it was really not a nice job to do. The thing is, when we were doing that, because none of us were already experts in that and all of us just wanted it done, yes, we processed all of the paper, but we were not that careful if everything was a legitimate expense, if everything was booked, like by the cent correctly. And if now you use the solutions and I hope that I had the solutions where, you know, your employees, instead of getting their paper receipt, they just have an app on their phone, they scan it, and then it's sent already immediately to the tool. This tool will scan the receipt and will already identify if you are allowed to spend that much money on a restaurant. Or if you went to China and you went to a gentleman's club, the tool is going to translate it through AI, the Chinese into English. And if it's written "gentleman's club," then it will be flagged as a receipt that cannot be reimbursed because you are outside of your company policy. And it's not a lot of money when we are talking about that, but it's a lot of work that finance people have to do. And if you replace that by AI, first you remove this mundane task, which nobody wants to do. And on top, you get better compliance because AI is not lazy. AI doesn't want to go home at 6:00 PM every evening because there is something better to do and spend time with your kids. AI will flag if there's a brand, will understand it, will also basically follow the rules. And this is where you can save a lot of time for the teams, but also as a company, you can avoid a lot of problems. You can save also a lot of money. And this is where if you use the strength of AI, which is it can process a lot of data, it can follow an algorithm, and it can understand text now, thanks to NLP, so natural language processing. Then you have a win-win because finance has more time to do human tasks, and then we can delegate all of the robotic tasks to the robots.

Megan - 18:15: So that covers how AI can help with risks. Do you think that AI introduces any risks, or is there anything that you would point out that CFOs should be concerned about when implementing AI?

Nicolas - 18:32: Yes. That's something that happened to me actually two weeks ago. I wanted to recruit somebody to help me with data analysis. And I just gave a file with revenue data. And I asked, "Can you do an analysis by country and by product?" A few hours after, I just got a really nice Google Doc with a perfect analysis done with graphs. And I was, wow, like I didn't think this person could do that. And that was somebody I wanted to hire. So I thought, okay, "Show me how you did that." And for me, it's okay to use AI for this, but I want to understand if the person understands what they are doing. And when I asked them to show me what they did, it was basically just uploading the file into ChatGPT and making a copy and paste of my instructions without anything else. And then after just continuing the discussion saying, "Yes, yes, continue, continue," and then made a copy and paste into a Google Doc. That means this person didn't verify if AI, so here, ChatGPT, if the results were correct. The person didn't guide the AI tool, meaning the person himself, because there was a guy that did that, didn't understand what was needed to do the analysis. So when you have this, it's really dangerous because it looks smart, but if the person is not understanding what they do, there is a big risk that they are going to deliver a work that looks very smart, but behind there are a lot of mistakes. And so I want to really emphasize that as managers, you need to encourage the use of AI, but you need to ask your team to be transparent with you and ask, "Okay, how did you use it? Show me. How did you make sure that the calculations are correct? How did you also verify that it was the right approach?" And if the person understands that and you can learn it also with AI, you can ask AI or ChatGPT, for example, "What is a sensitivity analysis? Explain it to me." And if you understand it because you learned it like reading a book, then you have more of a chance to be critical and to use a sensitivity analysis the right way. So this is the risk and reward, and that's something to be really mentioned because I think it's like at school where more and more teachers have difficulties giving homework to students because they always receive basically just the homework done with ChatGPT. So we need to rethink how we are delegating, to rethink how we are also putting people into a position where they still have a lot of critical thinking, because now it doesn't matter anymore if you are able to do formulas in Excel, if you know how to process data, because AI can do that. What matters is, are you asking the right questions? Are you asking about topics and to do the right analysis? And if yes, you can do that, then you will be super valuable. But if you are not having this behavior, you put yourself and the company at a big risk. And it's one day or the other, somebody will notice and then will just say, "Okay. I don't need you because I can myself go to AI, and I don't need just a messenger between the AI tool and myself."

Megan - 21:49: And I'm just curious, but, like, thinking about kids using this at school, I mean, do you think there'll ever be a point in time where humans lose their ability to think critically? Kind of a futuristic question, but...

Nicolas - 22:04: I think we had phases in our history where first there was the book. When books were written, everything that was written in the book was the truth. Then you had newspapers, then you had TV, then you had the internet, then now we have the AI chatbots. With each wave, we have kind of a new source of truth and the same principles that were available when people were searching for the truth through Google search. They still apply now with AI, or the same principle where somebody was watching the news on TV and believed that this is the truth. Again, like I was always taught at school and by my parents and by also my wife who is also really critical. You need to make your own research. You need to be critical yourself and to understand it yourself. And this will become more and more important because the information is processed faster and it sounds more and more intelligent. And the only people who keep their critical thinking, they will really take an edge compared to people that are keeping this critical thinking. So that's becoming more and more important. It's not new. It's just accelerating.

Megan - 23:15: Yeah. And you mentioned there being human tasks versus AI tasks. So where do you see the balance between human expertise and AI-driven insights?

Nicolas - 23:26: If you are in a business where the human contact is still really important, I can take the example of my last business where one call with one client could change our forecast by 20%. And as a finance person, your Excel expertise was not as important as just having a really good relationship with the sales team and basically making sure that the sales team, each time they had the latest news, that just ten minutes after they would give you the news like this. You had the most accurate forecast. And for me, that's what I call the human skill. What you can have also is that AI through a company where you have much more clients, a company where you have much more volume of interactions with the client and the sales team. Well, if you are a head of finance and you call the head of sales, well, the head of sales is not going to be aware of all of the interactions with all of the clients. But what you will have is in the CRM, in all of the emails, you will have a lot of touch points, which AI can analyze and give a trend and explain that, okay, over the last week, we had an increasing number of touch points from our client. We had an increase of positive reactions or positive requests, and this is translated in an increase of a forecast compared to our baseline. So that's something to, I will say, like it's an arbitrage between the type of business you have, the amount of data you have, and for some business, for some projects, for some types of information, you will need the human contact. For others, you will need AI to be there to process it.

Megan - 25:06: And we've talked about ChatGPT, but what are some other impactful AI tools that you currently see in the marketplace for finance professionals?

Nicolas - 25:16: First, the question I always get in my trainings is, "Which tool is the best?" Is it ChatGPT? Is it Claude? Is it Gemini? Is it Copilot? Is it now DeepSeek from China? Is it Mistral from Europe? And my answer to that is not to respond to any of the models because imagine if you need to drive from New York to Boston. Basically, you have the choice between a cabriolet. You have the choice between a truck, a Ferrari, or a Tesla or an old car. If you never learn how to drive, it doesn't matter which car you take, you will never reach your point. If you are a new driver and you have an experienced driver, again, the experienced driver will drive faster than you and arrive at the point faster than you. Now, if you have the two same drivers, but one driver needs to go with his family of five people and the other one is just with his fiancée, then you need two different cars. And so based on this, it doesn't really matter the tool. What matters first is, can you drive the tool? So learn how to drive, learn how to prompt, because then you can do 90% of any task with any of the LLMs. Second, based on your task, you might have some LLMs that are a bit better than others, but those are really marginal. So those are my recommendations. I would say that the type of model will be just a commodity, because now we see so many models come. What will be important is their integration in all of your company's environment. And we see that with Copilot. We're now with Copilot from Microsoft. When you do a query in Copilot, it's not only that it has access to the model that is behind, but it also has access to your email, to your Excel, to your SharePoint, to your PowerPoints. And so it can answer to you with a qualified answer because it has more information about you than any other tool that is not connected to your environment. And that I think will play more and more of a role because that's how you are going to get value as this is where the information or edge is in your company data.

Megan - 27:30: And I'm just curious, what do you see as some of the biggest hurdles that finance teams face when integrating AI? I'm sure like young people coming into the workforce are probably ready to embrace this, but maybe there's some people out there that are stuck in the way things were done. How do you encourage change?

Nicolas - 27:49: So first, that's what I advise in the roadmap, you need to give an LLM chatbot license. So Copilot, ChatGPT, or Gemini, you need to give that to all of your employees. The day you give that to your employees, they feel allowed to use AI. Because before that, a lot of people were using it on their own with their own email address, and maybe they were doing things that they should not do because they are using their account, which is not authorized. But if you give a license, then you allow everybody to do it. And so people were afraid to use it. Now they are authorized to use it. So the acceptance will just increase immediately. And when they start using this chatbot, then their acceptance will increase, but also their appetite. Because once they will see what they can do, they will start to learn more and more about this AI chatbot. And then they will want to do more because they will want to see what happens if now it's connected with my system. And now they are going to look at dedicated AI finance tools because that's what we have right now. The LLM chatbots are not really connected with your ERP or with your accounting systems, but what you have is AI finance tools. One example that was released really recently is concourse.io. This tool is connected to your QuickBooks and from the general ledger, this tool can in two minutes generate a monthly review with a perfect commentary, like a CFO would write. And I say CFO, I don't say junior, really like a CFO would write plus graphs plus tables. And this is done really like in two to five minutes. And on top, you can customize it after. You can change it. You can add a graph, you can make it shorter. And again, AI can do it for you, like make it shorter. So that is already available today and tomorrow, but it's also available today, but I will say tomorrow because I think we are just starting, you will start to have people that are using agents to replace some parts of their tasks. One thing we did recently in the AI Finance Club, we had an expert that showed us a tool made with N8N. It's a low-code automation tool. It's a bit like Zapier or Make, but it's more adapted to AI. And so with N8N, what we created is an agent that is categorizing credit card transactions. And the way it worked, you just give the agent access to a Google Sheet where inside you have your credit card transactions and then you have also the categories of expense that you use in your business. So it could be travel, could be office supply, maybe software, and then you combine both and you give that to the OpenAI API with a prompt explaining to OpenAI, "You are my assistant and you will have to categorize all the introductions based on these categories that I've defined for you. And on top, if you see the city in the transaction, please add a location based on where the city is." For example, if you are in the U.S. And really like in thirty seconds, this assistant processes all the transactions, will map all of the transactions to a category and on top adds a location when they see a city. And then you get your document categorized, which I know it because I have people in the AI Finance Club, they cut their time they spend on categorizing expenses from three days per week to just half an hour per week.

Megan - 31:35: That's crazy. So for people out there listening that are looking to upskill in AI or stay ahead of the AI curve, where should they start? Like what resources would you recommend for them?

Nicolas - 31:49: So I think you have two choices. If you love to learn by yourself, if like me, you prefer to watch a YouTube video about an AI agent than to watch a soap on TV, you can look on YouTube. I have my own YouTube channel where I explain a lot. You can follow great people on LinkedIn. You can follow also on social media. So that's your own path. If you don't have time and you don't know where to start, what I did is I created a course for ChatGPT. So it's a ChatGPT video course. It's called "ChatGPT for Finance." That's a great start or the best is actually what I've built is the AI Finance Club community because AI is changing every day and you have always new information. And the advantage of being in a community is that you get every week new content and every week you can also ask your questions. You can participate in live events where with other CFOs or other finance leaders, you basically bring your program and you ask questions and you'll have an expert that will be there to help you and to guide you to use AI in your own problematics. On top, if somebody comes and has already solved the problem, then you can learn from them and just save a lot of time.

Megan - 33:05: And last question, but looking ahead, where do you see this evolving over the next three to five years and what new skills or mindsets are gonna be critical?

Nicolas - 33:15: What I think is we are going to be more and more a manager of humans plus robots. And when I say that means you have to recruit the human first. Here's the thing, a robot, you need to recruit either you get a tool from outside or either you start maybe to train the robot. Like you also get a new intern that you have to train. Then once the people start working for you, you have to give instructions. But with robots, it's the same. If your instructions are already clear, if you know yourself the work, then you can help your teams to perform better. But then if your team is making mistakes, you need to be there to give feedback, to correct actions. Same with robots. You need to be there to change, to change the parameters, to verify that their work is doing well. And this is not a process that you do once, it's a continuous process where you need to be there. You need to coach, you need to review, you need maybe to hire new people, maybe to fire some people and with robots, it will be the same. And the people who are already doing that and taking the humans plus the robot as just a team together, I think they have more of a chance to succeed and you don't need to leave that to other teams. You don't need to leave that to IT. It's like, imagine if you leave the management of your team to HR. I love HR and I've partnered with a lot of HR people, but HR is not there to manage your team. They are just there to help you with the resource to manage your team. The same way IT is there to help you give you the resource to manage your robots, but you have to manage your robots yourself and you have to learn how to do that. And this requires, like we discussed before, to learn and to be critical. And I think it also requires that yourself, you need to spend more time understanding what needs to be done, because how can you coach a robot how to do a revenue recognition? If yourself, you don't understand the ASC six zero six. So how can you review that? It's impossible. So here, same. So you will have also more time because you will have more time to learn that, but it means you need to be a better knowledge worker and you need to manage these robots. And at the end, I think it drives us to always reflect. Am I doing something really human or am I doing something that a robot should do? And if you think that the robot should do it, then delegate it.

Megan - 35:44: Nicolas, thank you so much for being my guest today.

Nicolas - 35:47: Thank you, Megan. And I hope it inspired a lot of people. I think where you should start is start small, but use these LLMs first for you, then also encourage your team, give them licenses, learn by yourself if you can. If not, if you need any help, I'm here. Feel free to connect with me. I am Nicolas Boucher on LinkedIn.

Megan - 36:06: Thank you so much for your time today. This has been really inspirational. I'm excited about the future.

Nicolas - 36:12: Thank you, Megan.