Transcription
There are thousands of ways to use AI in business and finance. Everywhere you hear how AI can optimize everything, but almost nobody explains you how and when to use these tools in specific and real-world situations.
In this video, I didn't just collect 100 LEGO pieces that represent the best AI tips. Each block represents real use cases from my own work with AI in finance. To prevent your understanding of AI from collapsing, in this video, I've ordered every tip one after the other to build first foundations and then give you the best tips that can build on each other. And even better, some of the things that we'll be discussing can be copied and applied immediately after watching this video. My name is Nicholas. I have trained more than 10,000 finance professionals on how to implement AI in finance. Let's begin building this tower together.
Where does any construction begin? Many will say the foundations. Let's start with secret number one. Choosing the right license. My advice, if you're a Microsoft shop, go with Copilot. You trust Microsoft. You already have the license organized. So just get a Microsoft Copilot M365 license for all of your team and then you have no doubt about how to use data with AI. If you are a Google shop, same principles on top Gemini today is already included in your Google Workspace account. So you don't need to pay for it. If you are smaller like an individual fractional CFO, you can go with ChatGPT. It's a bit better in terms of functionality, but it's not really integrated with any of your data. So you have to connect everything.
Number two, AI is not deterministic but probabilistic. What does it mean? Generative AI is not calculating an answer, but generating it. So if you want to use AI for financial analysis, don't let AI generate the answer. Instead, ask for the formula or the code which you can audit. And for this, my tip is to ask to create an Excel file with formulas or to get the Python code. Both you can audit, both you can reuse it and productize your work.
I could not detail all of the 100 AI tips in this video. I mentioned them, but if you want the detail of all of them and how to implement them for your company, check this top 100 AI tips that you can get in the description below. There is a full guide that I have made for you and you can get it for free.
You need to use a reasoning model for real decisions and fast models to write. You might not know this, but in ChatGPT, in Gemini, or in Copilot, you can decide when to use a fast model. So something really good to write an email or to correct your grammar. But when you need to face a complex problem like financial analysis, creating an Excel file or code, use the thinking model or reasoning model from ChatGPT, Gemini, or Copilot. Personally, I use the thinking model 95% of the time.
You need to decide how you handle confidential data before you upload anything into your AI tool. For this, create in your company a tier one, tier two, tier three model. Tier one being public data where you don't really care in which tool it goes because it's public information. Tier two, this is internal information and on this you need to know which AI tool you can use because it's internal information that you don't want to share. And tier three is critical and confidential information where you really need to know which tool and also which people should have access to this information. This is why you need to understand if your AI tool has enterprise-grade security and do your proper analysis.
Number five, my favorite, it's the CSI and FBI framework that I created. I've told that to more than 10,000 people because it helps people prompt the right way. CSI and FBI stands for the context. You give the context about who you are, where you work, specific. You explain your problem. You instruct AI. You tell AI what to do. You get like quite a good prompt and I'll say a good input. But if you want to have an excellent value from AI, you add the FBI. Like this CSI and FBI, you are sure to solve your crime. You give the format. For example, you can ask for a table, for a PowerPoint, for code. The thing you have in your mind, the blueprint, you can express in your prompt. So for example, you want low-hanging fruit, you want to be challenged, you want quick actions, or you want just Excel formulas, nothing else. This is the blueprint. And I this is where you can give AI an identity and ask AI, for example, to be an expert in cost cutting, to be a teacher, to be your mentor, to be your coach. It will also influence the AI's response. So remember that CSI FBI.
Number six, you need to make your data AI-ready. AI likes to have structured data. It means you need to have your data structured in columns and with transactions in rows. If you use AI with those tables that are made for humans, you are not going to get a good output. So instead, prepare your data. Take also the raw data that you get from your ERP, from your systems. Those are the most ideal data if you want to let AI analyze those tables.
A lot of people don't know this, but AI cannot process that much information. Ask AI to write a book. It will just write you the curriculum and maybe the first pages. So this is why there is a technique called the chunking method where you have to chunk the work you ask to be done with AI, and that also works with a big problem. Don't ask AI to solve a big problem. Instead, cut the problem into smaller parts and then attack the first part first and then move on to the second one once you are happy with the result of the first task.
Really important. I had it again today when I gave a training. People think that they can automate everything by just throwing the problem at AI. No, it doesn't work like this. When you want to automate a task, probably if you work in finance, your task is deterministic. You have a problem. You have a consolidation of data. Well, every month you want this data to be consolidated the same way. And for this, you can ask AI to automate the script, but not do the consolidation for you. This is why you ask AI to create the script and then you use the script every month without having AI in the middle.
One of my favorite tips recently, you have been there, you have spent hours just typing into ChatGPT or Gemini or Copilot and you have like really, really long discussions and at the end, finally, you got the output you wanted from the beginning. But imagine if you need to do this every time you want to use AI, you're going to get crazy. You're going to lose a lot of time while you want to use AI to save time. This is why when you have spent half an hour, 1 hour, and you finally got this result, ask at the end, what is the system prompt that will have given me the same result from the first go? And AI will detail to you a really good prompt which after you can save in a document and reuse it later, or even better, you are going to save it and paste it into a custom GPT or a Gemini or a Copilot agent, and next time you use it, you get the answer straight away.
We have just talked about it. This function is underused. People, even if they think they are advanced at using AI, they don't know how to use custom GPTs, Copilot agents, or Gemini. Like I've seen all of that in my course that I give. People think that they can prompt, but they don't use this function. So if you want to be an advanced user, create your own custom GPT, your own Copilot agents, or your own Gemini. This will save you a lot of hours and will just increase the productivity when using AI.
I think many of you are wondering, Nicholas, we already knew most of this. So when will you get into the specifics of finance? I agree with you. And this is why, having finished the foundations of our tower, we're moving onto the next level: data and numbers. Crunching numbers and data analysis is what is taking us ages. But now with AI, we can accelerate this process.
First, start by using a thinking or reasoning model. Then you're going to prompt exactly this into the AI. Step one, ask AI to check the consistency of the data. This is all the habit I have as an ex-auditor. This helps you make sure that your analysis is always working on the right data. Number two, ask AI to work and think about the best analysis to do. But not only analysis in general, ask specifically to do descriptive analysis, diagnostic analysis, predictive analysis, and prescriptive analysis. This will increase the quality by 10x. Step three, ask AI now to calculate and show you the calculations. Step four, ask AI to also create graphs. This is the best way for us to consume data. And finally, last step, ask AI to create the commentaries like a good seasoned CFO.
You can also use AI to create financial models. For this, again, use the thinking model and then give your assumptions because how can AI know what are your assumptions about your business and also how you want to plan the future? Be specific enough, but don't write a textbook about your assumptions. Then, really important, ask AI to create a model that has formulas and also ask it to create it in Excel. Today, ChatGPT and Copilot can do that. Gemini, for now, at the date of this video, cannot do it, but I'm sure you can do it in the future. Once you have the model in Excel, review it, but then don't stay there. Continue the discussion inside ChatGPT or Gemini or Copilot and ask to create a tool that will show the scenario analysis, but also make it dynamic. And like this, when you go into your meeting in front of your boss, you can change the assumptions live and answer to all questions straight away in the meeting, agree on one scenario.
Now you have your analysis, you have your models, but how can you also create reporting or dashboards? There is a hidden tool which almost nobody knows about. This is called Canvas. You can activate it by clicking the plus button. And what you will do, once you are done with your analysis, ask AI to create a dashboard in HTML, and then it will use the analysis to create the dashboard. And of course, you need to review, but this will save you so much time. It's much more efficient than letting a BI team work on it for three months. And if you like Gemini, on top of creating the dashboard, you can create slides with Gemini. Again, going to Canvas to create slides for you, and it will create Google Slides with your information.
We did financial data analysis, we did models, we did dashboards. But actually, before this, how can you use AI to prepare your data? Well, you can use AI to create automations, and those automations, they will be scripts. And this is where you might be afraid of it, but AI is your best friend. I am also like you. I never learned how to code, but today I have hundreds of automations that are running that are helping me to clean the data, to consolidate files, to transform the data, to audit my files. And I do that with some Google Script, some Power Query, some VBA, some Python, just a combination of the best tools. And the best of it, you don't need to buy anything more because everything is already installed on your computer. It's just that today you are not using it because you don't know how to code. Well, now with AI, you let AI code, and you can automate all of this task in just a fraction of minutes. And this little investment that you perform once when you create the script will transform into hours of work saved every month.
Here are more advanced tips on data and numbers. Almost nobody knows about this function, and I did a video on this. You can with Excel agent mode create financial models within Excel. Create graphs, pivot tables, transform data, clean files, just within Excel. Other advanced functionalities that you should try: use AI to create the forecasting seasonality for your business, especially if you have a business where you know that in some months or some days of the week you have more or less demand. Something else you can also try when you are using Gemini is to extract the Python code from an analysis and ask to import it in Google Colab. Google Colab is the place where you can run code within Google. It's like Google Docs but for code, really easy, even for somebody like me who never learned how to use code. And even better, in Google Colab, you have Gemini also installed inside, and you can let a data agent working the full analysis on your file. And the best, when the analysis is done, you have the raw code which will help you to redo the analysis yourself next month once you have the new data and all of it following exactly the same steps. So you know exactly what happened. You don't let AI do the probabilistic job which we talked about earlier. You have a deterministic code that you can audit and that you can reuse in the future.
All the tips I mentioned in this level are great and they can save you days of work and thousands of dollars, but they will never work unless you know how to prompt these tasks. And this is why we move on on the next level called prompting mastery.
First one is the chain of thoughts. You need to understand that with AI, this is a good principle where you break down complex problems into step-by-step processes. The way it works, you ask a general problem. Let's imagine you want to reduce cost. You ask for a plan to reduce cost. And then when you identify one of the actions that you like that you want to investigate, drill down into this specific action and ask for more details. And once you have the details, ask how to implement it. This is a good way to go from general to specific thanks to the chain of thoughts principle.
Next technique, the chunking method. We talked about it earlier, but let's go back to it. Imagine you want to create a financial memo. The way to create a financial memo which is complete, which also follows your style, is first to ask the curriculum, the structure, and then to give the style chapter by chapter. So you say first how you want your chapter to be written. Once you have one chapter that is well written like you want, then you ask to apply it to the next chapter and then to the next one. And like this, in a few minutes, you get your financial memo written for you in your style.
Another technique which is really important for finance is the explicit reasoning. With explicit reasoning, you're going to let AI tell you how it achieved the result. Ask it to detail your calculations, to show the formulas, and if you are not sure, even ask it to review its own work.
Another of my favorite prompting technique is agent prompting. We are going to give AI a persona. We are going to say what type of identity it has, but also the traits, the characters, the output. So you're going to define this, and then you can start your prompt with the persona, with the agent, and all along the discussion, it will always act as this persona.
And once you have built your prompts with agents, you can actually build team prompting where you get several of these agents to talk to each other. And the best out of it is you create ideas that you have never thought about because you have these different personas interacting with each other. The way it works, at the beginning, you define each persona. So let's imagine you have somebody from sales, somebody from IT, somebody from finance, and you ask AI to think about the perspective for each of them. And after you build kind of a mini project building on each other's response to have something much more complete.
If you want to reduce the associations with AI and just to increase the quality, you can use the meta condition. With meta condition, you will ask AI to review its own work and ask to rate its own work. So, for example, after you have an output, ask AI on a scale from 1 to 10, how practical is this output? And then AI might say 7 out of 10. Then you ask AI to review its output and say, now make it a 10 out of 10. And you will get something much more practical.
Try also this technique. You are going to tell AI to ask you questions. And this is a good way to let AI ask you questions you have never thought about, but also let AI know more about a topic and go somewhere you will never have imagined. This is called Socratic prompting, from Socrates, because the goal here is to find the root cause, and this works only by asking you questions.
We talked a lot about prompting here in this video, but you know the best way to get the best prompt is to ask AI to improve your prompt with the prompt optimization technique. AI will start with your intention, and then you will ask, please improve this prompt, and you will see how well your prompt is written, and then you can reuse this prompt and ask AI to work on it.
The technique I'm going to reveal is something where a lot of people struggle with. You see, often we ask AI to give you some facts, but you might not know where does it come from. You might not know the source. And instead of searching yourself if the tax rate that was given to you by AI is the correct one, ask AI to give you the fact, to give you the source, to review online if it's exactly the same tax rate that it gave you. And you will save a lot of time and you can back up all of the work you do with AI if you do this fact-checking technique.
Then the last one, the last technique is something which will help you. The last technique is a technique that helped me get AI the right answer for me. You see, AI doesn't know anything about you. But by using the iterative inquiry and sequential questioning, we are going to let AI ask us a series of questions that also builds on each other to let AI know more about our problem, about us, to allow AI to answer better to ourselves. Try this especially if you want coaching. Try this if you have problems. Try this if you want to brainstorm. It will do wonders.
And on the note of personalizing AI, we slowly reach level number four, custom AI. First on design, how should you define your custom AI or custom GPT or Copilot agents or Gemini? You need to define them as one GPT, one mission. Each assistant has a focused purpose. For example, one which is there to create commentaries, another one to help you do reconciliation, another one that helps you categorizing expenses. Once you have the mission, then write the instructions inside. You will use the CSI FBI. You will give clear identity, clear instructions, clear output. What you can do also is give examples of the output you want, but also the output you don't want. The goal of it is to have GPTs that will help you do 80 to 90% of the work.
Then to make your GPTs or Copilot agents or Gemini much more powerful, add knowledge. You can add your policies, your bio if you want to do a digital twin, or you can even add examples of the output. But when you do this, chunk your documents. Don't upload 200 pages of documents. It will make your GPT much less valuable. Also, make sure you name the documents correctly. That has a huge impact in helping the GPT and the models searching for the right information in the right document. Also, always request your GPT or your Copilot agent or your Gemini to show the sources. Like this, you know what comes from the model or what comes from your documents. And don't forget, if your documents are not synchronized, because for now ChatGPT doesn't have synchronization with any document, only Copilot has this with SharePoint, remember to always refresh the documents if there is something new.
In terms of governance, you need to make sure that you know who has access to these GPTs and who should not have access. How do you optimize your GPTs? You will see over time that your custom GPTs, Copilot agents, or Gemini might work differently, might maybe give you not the answer that you always want, and it happens to me as well. So the way I do it is that I test it and I continuously improve it. I go into the instructions and I change and tweak some details. For example, today I made a custom GPT which only wanted to give me three options in a table. Well, I forced it by saying, you always have to give me at least five options. And now I have much more value because I have five options instead of three.
Lastly, make sure you audit and review your GPTs, but also have a list with all of your GPTs, what they do, who is responsible of it, and make sure that when somebody creates a GPT, they share it with everybody else, but also receive comments on how to improve them.
Our tower is now getting pretty tall. But so far, we have talked about tips that can apply to most of the people. And now let's stop the building up with tips for AI CFO leaders.
Let's start with the CFO strategy and AI operating model. You will need your AI roadmap, which you start by choosing the right license, then organizing an AI opener first session, and then doing a workshop. After that, you can even run a hackathon. This is what I advise to every team in 30 days: do this. Then you are going to define an AI maturity model. You will start with level one, which is more on an ad hoc basis. So the AI used only really slightly to level five AI native, where everything is done with AI. And you will do it for each of your processes: for closing, for FP&A, for treasury, for tax, for controls. You take every process and you see how you can do it today without AI and your target model, how you want to do it with AI. But pay attention. It's not because you have AI that you have to use everything with AI. Make sure you use AI at the right place, at the right moment, with the right people, and the right tool.
Then, with your team, you have to build a use case portfolio. What you will do is you will map AI ideas across all finance domains and then you will score them considering the impact and the effort. By doing this, you will have built your own metrics where you will know what is at the top right, meaning the AI use cases that have the highest impact with the minimum effort, and you will want, of course, to focus on these AI use cases.
Once you have your list of use cases, help your team with a standard business case template. It will force people to have the same structure and understand the problems, the options, the numbers, so how much it costs, but also how much it brings, the risks, and the recommendation. Once you have this business case template, look at having your transformation KPIs. Track adoption, cycle time, error rate, manual steps, power saved, and also decision quality.
Number three on forecasting. This is something where companies like Microsoft or Coca-Cola have done really well. They have revamped the way they are doing their forecasting. Instead of doing bottom-up forecasting and making all of the teams busy, they have centralized the forecasting with an AI model, and they will share the forecasting to all of the locations. And each team, instead of having to build from scratch a forecasting, they just have to review the forecasting baseline made by the model. And instead of working on the forecasting for one week, they will just have to review in a few hours and give their feedback to the central team. Imagine the time saved. And on top, Microsoft and Coca-Cola have reported that they have increased the accuracy of their forecasting against the actuals and saved a lot of time.
Another way you should use AI as a CFO is everything linked to your stakeholders, your board meetings, your investor relations. You can use AI as a sounding board to test yourself, to negotiate with stakeholders, to prepare your communications. But also, let's be honest, you are the last person in the chain, and all of the documentation is on you. So, if you want to save a lot of time on this documentation, use AI for all of these interactions with all of these stakeholders. And one good tip that somebody told me recently, they are using AI to review each of their meetings and ask AI how they could have done their meeting better, how they could negotiate better. And since they use AI for this, they had much more impact in their board meetings. And now, each time they go in front of investors or a bank to negotiate a new loan, they are just more ready and better at negotiating.
And finally, this would be nothing without the people. So use AI to help your organization to train them better. Make people skill assessment with AI. Prepare training plans. Prepare onboarding plans. Review also if you have hidden skills in your team that you have wrongly allocated to the wrong place. And one thing you can do is build the AI-native competency model. Basically, aligning your people with their own AI skills. See which ones already have the right AI skills, but also which ones need to build these AI skills. And like this, you'll have AI-native teams in a few months and a few weeks.
This tower from all the AI tips may feel overwhelming and scary considering how many small details you have to learn and know. In that case, you can consider joining the AI Friends Club where CFOs from all over the world share their minds and grow together.
Obviously, in this video, I couldn't go in detail through all of the 100 AI tips. This is why I created the Top 100 AI Tips guide where there you see all of these tips in detail and that you can go over and over with your teams to make you and your team AI-native in finance. To download the guide, just check the link in the description. Thank you guys for watching. I hope this video was helpful. Make sure you like and subscribe if you want to see more videos like this. If you think there is a tip that I've missed, let me know in the comments.