Name: Agicap Webinar: AI Roadmap for Finance Leaders with Nicolas Boucher
Uploaded: 2025-03-10T15:30:53.397Z
Duration: 1 h 13 min 20 s
Description: Agicap Webinar: AI Roadmap for Finance Leaders with Nicolas Boucher

Transcript for "Agicap Webinar: AI Roadmap for Finance Leaders with Nicolas Boucher": And we're live. Not sure how many people have joined yet, but, for those that have, I know that we have a particularly big audience today. So welcome to those who've already joined to the first of our series of webinars we're running with Nicolas Boucher, this week and next, for AI, road map for finance leaders. I'm gonna kick start the session with a quick poll on a very important question for, AI as it pertains to usage within businesses. You should, as of about now, see this poll. Do you have access to a corporate license for any l LLM, AKA ChatGPT, Copilot, Gemini, or or Claude? Any of these tools? Are you using them regularly? Just a simple yes, no. You should see top right, the poll on the, right hand side. Just fill it in as we go. I can see lots of people have already done so. Good. Numbers are racking up fast. This is nice. I'm sure Nichola will be pleased to see that there's more yeses than nos. So this should make for a great conversation. I'll just wait for people to keep voting, and then we will shut the poll and get started. As I know, we have a lot of content to get through in quite a short space of time. Okay. Keep them coming. Let's see if we can add a 50 votes. Okay. I'm gonna close the poll and share the results. Thank you everyone for participating already. So we see that 60%, sixty forty. I'll be very intrigued to hear whether Nikola, thought that was, kind of in line with what he's expecting, and what he's seen in, his webinars elsewhere. Yeah. That I think the last month, I've seen this number increased significantly. Especially a lot of company, what they do is they move with Copilot from Microsoft because most of companies have already Microsoft, And moving to Copilot is much easier than choosing any other tool because it's just like an additional option to take. And that's what I see most of the company doing. They just, like, added the Copilot option, and, I think that's where we saw a lot of differences compared to one month ago one year ago. One year ago, I think we were more at, like, 20% yes. And now seeing, like, two third of the people have already a corporate license shows that the problem of data confidentiality is going to be less and less a problem. Because if you have a corporate license, means you can already use the LNM bots for your work, and you have less fear of doing something wrong if you put data inside, especially in finance. Okay. Super interesting to hear. Also interested to hear if there's any kind of cultural element to this. I know this is probably one of the most international, webinars we'll have. I've already seen people from India, and The US on this, webinar, so people from all over the globe. If there's, a cultural element to people using different tools in different places, please add and share the names of any other tools you're using. Otherwise, without further ado, let's get to it. The AI roadmap for finance leaders. Just introducing, ourselves quickly. As you might guess, the person on the right is me. My name is Nat. I head up the partnerships team here at AGCap. And with me today is Nicolas himself, who you've just met, the number one AI and finance expert. If you do not already follow him on LinkedIn, definitely do that. And then we have Kelly from our own AI engineering team here at AGCap with me. In terms of what to expect today, we've got, as I said, quite a lot to get through, but upfront, we'll talk about kind of where you can get started. So we'll dig into that LLM piece, how you guys can master Gen AI. Then we'll dig into some of the the best use cases. So you we know already that AI is better at some stuff, than it is other things. Nicola will tell us where AI is particularly strong already. And then Kelly will really help us dig into this idea of embedded AI, so AI within products rather than AI as a complement to what you do in your own work, where you use the LLM. And then what I think is particularly interesting aspect of this discussion, which is AI being used for financial analysis. I think a lot of people think about AI as a means of automation, but can it support and to what extent with deep analysis? I think that's gonna be a really exciting part of the webinar. And then the bit, that I think is also really interesting is kind of that looking ahead in the two year time frame. Obviously, we're kind of feels we're at the beginning of this. There's still lots of tools for lots of different things specializing in different areas. How's that going to translate into how we work as finance teams in a couple of years time. And then please do drop your questions in the chat as we go super super important we definitely we'll have a Q and A at the end where we'll try and address as many of these questions as possible. So that's kind of really, what to expect from today. As I said at the beginning, this is part of a a series of webinars that we're running on this topic over the next few weeks in various different languages, and as part of a broader kind of thought leadership series that AchiCap runs about, finance topics, major finance topics from finance transformation to AI. So please do also follow AchiCap on LinkedIn, if and when you have time. I know we have a mix of kind of customers of AGICAP on this call. We also have, people who may never hurt have heard of AGICAP. So for those of you that don't know us, we are the next generation treasury management solution born in France, now a global solution with over 7,000 customers. We are the market leader according to g two for cash flow management based on customer satisfaction and market penetration, which in our case means ERP connectivity and banking connectivity. And in terms of trying to understand what that means from a practical point of view, in terms of finance tasks that HDCaps supports with, I kind of highlighted the main ones that we, cover in this slide where you can see, cash management, cash flow forecasting, treasury reconciliations. You can also make payments from the AGCAP platform and run automated collections as well as do much of your budgeting and financial reporting within the tool as well. For those of you who are interested in AGGCap, there is on the screen a a kind of request a demo button. So, you can click on that. It's an AGGCap demo, not a Goldcast demo. But, ultimately, AduCap is not the main topic here. Nicola, we would like to hear more about AI. So please, conscious of time, Nicola, could you tell us some of the cases you've seen? I mean, you've worked with hundreds of companies that start their journey with AI. Where would you recommend the finance teams start if they want to begin using AI? Yep. Sure. Thank you. And we said today we want to start with the AI road map because that's the question I get from a lot of CFOs. Where should I start? And that's why after, like, working with a lot of companies, I've defined a really clear AI roadmap that I want to share with all of you. And this is also a practical one because I think you can go to a lot of consulting companies like KPMG, PwC. And I worked at PwC, so I know, I really love PwC, but what will happen is that it will take so long before you have something concrete. And I know people want something concrete, so let me show you now. So if we want something concrete, I made the road map in three ways. The first way is for small companies. And we start with small companies because even if you work in a big company, you probably have a small team, or you are a small country, or you have a small department. So you need to start here even if you are in small comp even if you are in a big group. And the first thing is you need to have access to LLM bots, and we saw 60% from you already have that. So that's really good to know that here, a lot of you are advanced. If you are part of the 40% of the people that don't have a corporate license already, you need to work with your team, with your management to get one. Because once you have this, it's the fastest way to start with AI, and it's also the fastest way to get your team educated on how to use AI. So once you have this, then there is one process that here now in finance, all of us need to use AI. In this process, this is the purchase, so procure to pay, process, so or your accounts payables process. If today you are still doing that manually, you need to use there is, like, a lot of tools that do that. So but I need you need at least to use a tool that use AI for this to recognize invoice, but also to book the invoice without that the human is behind manually putting that into your ERP or accounting system. And then the last step, this is already something you can do if you have access to these LLMs. You can start automating your process using AI to code Python script. And I can show you an example later, but here is the fastest way for a small company and small organization to automate processes if, for example, every day you have to combine a lot of CSV files. Well, instead of doing that manually, you can write a script just in two minutes to let Python do that for you. And you don't need to know Python to yeah. You don't need to know Python. You just need to have the AI code that for you. Now let's move to the second type of companies, the bigger companies, so the middle size companies. And the first use case that you need to have in your roadmap is to identify if you have some processes that are really heavy. And if these processes are really heavy, then you need to look at which AI tool could cover these processes. For example, if you have to do the revenue recognition process over complex contract, I'm sure your accounting team is spending a lot of time on this. Well, you can use a tool like Trillion that is basically reviewing your contracts and proposing the booking for you, all of that done by AI. And the second part is to start forecasting with AI. This is not anymore where you need a team of data scientists to do your forecast with AI. You can do that with just, like, one laptop and the Internet connection and just having access to any LLM to code for you this mini code that will do for you the forecasting based on your historical data. And the third part, if you already have automations with RPA, so robotic process automations, normally, you should have the possibility either with, Power Automate from Microsoft, with UiPath, with Automation Anywhere, you have the possibility to integrate somewhere AI in these processes. So it could be that if you have an automated process for your accounts receivable where you send email, for reminders, you can use now AI to tailor these emails. And so the goal is just adding a bit of AI in already your automated processes. And then the last part for the, roadmap, and I'm not going to spend too much time because I don't think a lot of you are from big companies. But the big companies, what they do, they are working on building their own model where they add a lot of data, and they will train this data to have their own models. And based on this, they also build forecasting models. So for this, you need a data scientist team. But I've talked with a company like Coca Cola. That's what they already have in place since two, three years, and that's what big companies are doing. They are using AI with their own data to build forecasting models. And the last part, if you are a big company, you probably have a lot of interactions, a lot of departments. You can use AI chatbots to help you serve all of the other departments or even your clients or vendors. Super interesting, Nicola. Thanks so much for sharing that. I think what's really clear is, the range of use cases that we see already. Yeah. I mean, you started off there, talking about, the impact AI within the procure to paper process, and then you finish with, AI based forecasting. So, really, there's a lot a lot of use cases. For all the people on this webinar who maybe are doing either of those things, where would you recommend that they get started, or how do they go about prioritizing, the use cases of AI within their own business? Do you have a framework for them to get started in that way? So what I recommend is you need to educate people on how to use AI. You cannot start, like, a big project of AI if people don't even know how to use LLMs. So this, ChargeGPT or Copilot. Because once you have people that are really comfortable using Copilot or ChargeGPT, they will understand the power of AI, and they will have a tool next to them that will be able also to be better at, improving all of the other AI projects. And let me show you, like, just two or three use cases because we hear a lot of time departments like marketing using AI to create marketing, content. But people think that in finance, we cannot really use AI. I I want to show you that actually you can do a lot of, AI use cases with LLMs in finance. So let me show you a few use cases. For example, everything you need to write, here, I am, going to prepare a checklist to compare the HAGe base, so that's the German gaps, with IFRS. And I'm just using Copilot. For a lot of people thinking that Copilot is not good enough, that's really quite good. So you can use your LLMs to prepare a lot of what you usually do manually, like all of these checklist, internal control, standard operating procedures. You'll always LLMs to start from this draft instead of starting from, zero. What you can also do is everything that has to do with brainstorming. For example, here, I had a client that needed to identify KPI for their ESG report, and it's really new concept for a lot of teams. Well, you can use your LLM to have already much more meat, much more ideas, than if you didn't have this before. And one of the part that I love the most is to use, JGPT or Copilot to get 10 times better at at Excel. I remember, when I was a finance manager that my accounting staff didn't know how to use Excel to do the aging of my overuse. And I was always, like, asking somebody to help them or I was behind to make the formula for them. Now if you see here, I just ask, k. Give me the formula to do the aging of my overuse in Excel, and you get exactly the formulas you need, and you just need to apply that in your own Excel file. And that will help a lot of people. And last point, I want to show you that you can now also write commentaries. So imagine that you are a CFO and you have an Excel file with your p and l, and you have here a professional license for ChargeGPT. So I have ChargeGPT teams. I can upload my data. You can ask JGPT to write the commands for you. And look at what is happening when I ask JGPT to write the commands for me. First, it's using, Python to understand the Excel file, but after, it starts to write the commentaries using a new functionality called canvas. You can see that it's a bit different than the normal, char g p t. And here in canvas, we have on the left the discussion that we usually have with question and answer from ChargeGPT. But on the right side, I have actually a document that is being written by JGPT for me. And what is good with this document, I can, let me show you. I can go and select one part and change just this part by asking here, for example, more details. So it's like a document where you work together and you collaborate with AI to write these commentaries together. And here, I'm just asking, can you link this explanation to, the countries? And now AI is going to rewrite that for me. And together, we are rewriting this document because I can also delete manually. I can bold. I can, shorten it. And that's how you make AI help you to write your commentaries. So those were, like, sonic use cases in a few minutes that are already helpful for finance teams. Yeah. Again, I think, Nikola, that's a great range of of use cases we see there, of how everyone on this call can begin to apply AI in their day to day. And it all starts I think what you said at the beginning, it's most important for for all of us, whether refinance people, salespeople, marketing people, as you said, to really get up to speed on how to use an LLM. Cool. So, Kelly, I think, you are already an AI expert unlike, myself and potentially many people on this call. And you have, with the other AI engineers, AGCap, kind of been building, functionalities AI functionalities within the AduCap product. Can you talk us through how GenAI, so LLMs, are being developed inside of finance software tools like AduCap, to provide additional capability to many of the processes that finance teams already go through? Yeah. Sure. So the very first, use case I want to show you is the is the very first one we've implemented into AGCAT is the AI assistant. So the AI assistant is a module in the AGCAT product, and it is really interesting because it has access to all of your data in AGCAT. So instead of copy pasting the wall your wall financial context into CheckGPT and then using the response into AGCap, the AI assistant directly has access to all your data. So you can ask it, for example, here, I asked about a payment I wanted to do on a specific account. So the AI assistant will fetch my datas and retrieve the bank account balance, And here, it tells me, okay. For this payment, your test account will turn negative if you do it right now. So then I will ask him for advice. How can I do my payment? Because I really need to do it. And from what I've told him, it will give me some suggestions. So the first thing he's gonna do, it it is going to give me some generic suggestion, like delaying the payment, etcetera. But I want to use the power of, an AI connected to my account to have more personalized suggestion. So here, it would retrieve the cash balance of every of my accounts and tell me, okay. This test account too has the necessary fund to make your payments. So you should probably transfer some funds so that you can make your payments without having your first accounts being negative. So this is super powerful to have all your context within your assistant. And you can also ask some other things, to the assistance. For example, we have a super cool superb team, but sometime when you start using the tool, you have some questions about, for example, how to make a payment. And the assistant perfectly knows Educap, so it can guide you using a tutorial so that you can, go step by step and make your payment. Another really interesting feature we have, integrated into Educap, and I want to talk about it because it is not a chat. In we have a module called, account receivable where you can write and craft templates, email or related templates that will be sent through a workflow to the clients that are late paying the invoice. So you will spend many time crafting the perfect email for your clients, but you will probably do it in your language. And then you may have clients all over the world speaking many different language, and you would probably want to copy paste all this template into DeepL, then paste it again into a asicap, fitting all the placeholders to replace them, and this will be really time consuming. This is why we integrated this little translate button that will call an LLM to translate in as many languages as you want. And it is done in just a few seconds because the the model used is really small. So here with one click, you have done all your translation. This is a really good example, I think, of using Gen AI, in a context different from a chat. Thank you, Kelly. Super, super helpful, example of embedded products, or embedded AI, within a product rather. I think, another thing that would be really interesting for our audience to see, Kelly, is kind of how AI can automate many of the mundane. I think the examples you showed there show, how, you know, translating emails, at high speed like that. But much of, the challenge or the things that can really help us with in the short term are around, automating other mundane tasks. Do you have examples to show of what EdgeCap has built for those kind of mundane tasks to automate as well? Yeah. Sure. So one of the latest thing we have released in beta test for our user concerns payments. With the accounts payable module, our user are dealing with huge amounts of payments every day. And, you know, when you're dealing with so many data, mistakes can happen. Sometimes you have many users dealing with payments, so miscommunications can happen too, and sometimes fraud could happen too. So we decided to leverage AI to spot, the anomalies in this payment. For example, a payment with an amount that seems, unusual regarding the beneficiary or maybe a currency that is unusual or duplicate payment. So in the video here, I show you how we deal with the the case of a duplicate payment. So here, let's say I make a payment to DuPont beneficiary. It could be me, but it could be also someone else from the compatibility team. And I'm not aware that this payment has been done. So I create it, and then I sign it. I go through the the workflow. So here, I could already have done some mistake, like misplacing a comma or, I don't know, switching to payment. Those are human mistakes. But here, I go through my payment, I sign it, and then I don't know what why, let's say, I forgot that I've made this payment and I make it again. And here, the AI will just compare this payment to all the history of payments you've done. And then at the moment, you will try to add it to the payment batch. It will warn you, oh, this payment seems unusual. You're not supposed to do twice the same payment in such a little time. You should probably look at it a second time and enter it to the legit. And there is another another feature, about automation, it is categorization. In AGCAP, we are dealing with transactions that are representing cash flows, inflows, and outflows. And those transaction, we have to categorize them into predefined categories. This is really important so that we have a great overview in our treasury management. But the this categorization task can make many time if done by hand. So we decided to leverage AI to automate the rules creation. So for example here, you can see that the AI will check all the labels of the transactions and spot the keywords in those labels. By keywords, I mean the words that are holding the most semantic sense. So here, for example, it spotted that there were many transaction containing the word client. And so the AI suggested that we create a rule so that every transaction, past transaction, but also future common transaction containing the word client in the title will be, automatically categorized as one of services. So this is the suggestion of AI, but you could also, change it. And then you can create the rule and move on to the next one. So here, once you have created all your rules, it will be really automatic to categorize your transaction, and you won't have to do anything by end anymore. This is a really, really, big time saving, for our customer in ADJCAP. Yeah. Incredible, Kelly. I think, that see the depth of capability, of, AI when it comes to, data intensive tasks like categorization like that. I can see there's already a few questions cropping up about comparison between different tools as well. So maybe it's a good time to ask you, Nicolas, with no bias, of course. You will have seen hundreds of of of AI tools. What are the best ones you've seen that you'd recommend to finance teams? So what is clear is right now, you don't have the magic tool that will do everything. Just like to to challenge, and to manage expectations. But you have tools that are really good at one thing. And that's why when you choose to use AI finance tools, you need to identify where you have the best use cases. Let me show you, we did with my team a market map of the top AI finance tools. And, this market map, like, was after reviewing all these tools, talking to the founders. And let me show you two examples that are really visual on how AI can help us. For example, here, concourse, it's a new tool that has access, for example, to NetSuite or to QuickBooks. And from this, can create your monthly review report. All of this has been created by AI. You can see that's Michael with Matthieu, the the founder. And here, AI is writing all of the commentaries. AI is coding all of the graphs based just on the general ledger from QuickBooks or from NetSuite or I think they have also Microsoft Dynamics, that they were planning to do. And once that you have the draft of your monthly report, you can even change manually inside to remove some sentences to add, more information, or if you want, you can even generate more graphs. So here, for example, you can change the type of graph you want. So that's one tool, which is called concourse.io. And they are just new, so, you need, it's it's really new. If you want to check with them, you can check. I'm not partner with them, but that's an example of AI and how you can have a powerful help right now if you use the right tool, for your FP and A, for example. Another tool is Zebra AI. For Zebra dot ai, you upload an Excel file. And from this Excel file, Zebra dot ai will first identify what is inside. You can read that, it starts understanding what is the data. And then we'll create a reporting for you with some, graph that are adapted to your data, but also comments on this. And that's, again, something that before you needed a lot of time to create these graphs. And here, you have AI tool that will do that for you and will help you save a lot of time, but also make your reporting more valuable. So those are, some example. But like I showed you, like, you have hundreds of example. Katie also showed example of how a treasury tool can can help. And that's where where I think people need to expect that more and more, you'll have AI integrated in the tool you already use. And if you see that your vendor is not really doing that, then you need to look also, what are the alternatives. Because if we think in two, three years, if a tool doesn't have a AI functionality, I don't know if there there is still a market for this type of tool. Yeah. I definitely agree with that. I think, you can see in all the examples we've seen, how quickly the speed of development is, for all of these tools. One area where it would be good to get more insight on the speed of development is, obviously, AI with numerical analysis. I think the examples, Kelly shows show how good AI is with large datasets where you're picking out specific data points that have certain patterns within them. But how can our users on this call begin to perform financial analysis with AI? What where does it reach its limits there today, and where do you think it will it will get to, Nicolas? Yeah. Sure. So let me show you two examples that I've prepared. Let me show first this one where if you work for a SaaS company, you probably have to do cohort analysis. So cohort analysis, what it is is basically showing, on the left, you have the month. So that's when your clients started to be your client. And here on the the right axis, you can see how long they are with you. And based on this, you can see that in January 2022, you can see how long people stay with you. So after twenty three months, forty two percent of your client that join in January 2022 are staying with you. This type of analysis, if you want to do that in Excel, it requires a lot of work. Like, probably one afternoon if you are good with Excel, even even more if you are not really, Excel pro because you have formulas first, and then you have after, you have conditional formatting because you can see here that the dark are 100% and the yellow are 42%. So you need to have a lot of conditional formatting rules. So that takes a lot of time. Now let's see if we can do that with AI in just a few minutes. So here I have my data, which is a simple data with dates, customer ID, product, and invoice. And I will upload this data into ChargeGPT, and I will ask to create a cohort analysis, but also to create a visual analysis of it because that's the best way to visualize, cohort analysis. And if you ask the right question, so you can see here how I formulate it, especially writing visually, then you will have your AI tool that will start working on that. And especially, you can see the analyzing part. It starts to create a Python code that will replace all of the Excel work you have to do in one afternoon, and this Python code will help create analysis. And here is real time. Like, I'm not accelerating it. You have really the graph that is created just, like, probably thirty seconds, And you get exactly the same graph than before. But now instead of spending one afternoon, you had it in one minute. So what do you think about that, Matt? I think it's pretty extraordinary, and quite scary, to be honest, but very exciting at the same time. And for later, if you have other questions, I have another example with scenario modeling with AI, but, we can keep that for later maybe. I do have a a question around, best practice. How do you keep all of your inputs that you're putting into the machine? Because I guess it's a very different process when you if you would build this model in Excel yourself. You would show all of the work that you've done, it will be there. Whereas if you just ask the AI, to do it, how do users transport the AI's methods into into their work, I guess, at the end to say, okay. This is what actually happened. This is what we did. So here, I will recommend to use this only to prepare an analysis for yourself. But if you want to document your work to have your work auditable, I will reuse the code that was done by the machine, but do that in my own work. And there basically is where you can see the process where you have the input with the raw data, so the Excel file. And then the Python code is your process because the Python code says, okay. I'm taking the data. I am grouping by months, and I am, showing the graph that is a cohort like a heat map, based on these variables. And even if you don't know to how to code, Python is just like Excel, but just like a bit more advanced. So it's, it's not rocket science. Okay. Super helpful, and super, important. Obviously, it's that all financial analysis is is auditable. Kelly, just gonna come back to you now. I think to be very clear, both from Niklas' examples and your examples, the effective use of LLMs depends often on a lot of data, and having accurate data. But often, we see as a challenge kind of, that the AI tools don't have access to the company's data, maybe for security concerns or whatever it be. But you've actually built something inside EdgeCap that makes, financial analysis based on real company data. Do you wanna show us quickly how that works? Yeah. Sure. So the first feature I will show you is another feature of the AI assistant I introduced, previously. So the AI assistant I showed you, usually, you ask question to an assistant. And here is a more proactive feature when you can ask the assistant to scan your company and give you insights and advice. So for example, here, it has found, some my day sales outstanding. It will comparing to the industry standard, suggesting to reduce it, and maybe also it will suggest some investments. And it is a really game changer to have an assistant suggesting things like a really real proactive assistant. There is also something else really close to what Nicola, presented, and this is the last thing we are building. It is about the custom dashboard. So we have ways to build custom dashboards on AGCAP, but you have to it's about drop down menus, selecting your datas, and it is it can sometimes be a little limited, and you don't know how to set up exactly what you want. So we built an assistant connected to your data that will be able to create any graph or table you have in mind. For example, here, I'm asking for a waterfall chart. The waterfall chart is something that is not currently possible in the AGCAP classic dashboard, but here with the assistant, we will be able to do so. This is because the assistant has access to your data, so we it will first fetch any data, you need, and then it will execute Python code. And by exec by writing and executing Python code, it will it will be able to do any calculation you have in mind. So it's give us a lot of freedom regarding both the data and calculation and also the form of the graphs. We also provide some logs, about, within the codes so that you can see what's happening behind the scenes and also check that the data you see on the screen are the one you requested and you were waiting for. I also have another example of this feature because what is also really interesting is that you can come to the assistant with no precise idea of what you want. And I find it really powerful because, for example, here, I said, okay. I want to visualize, the evolution of my expenses, but I have no idea of the form of the graph I want, maybe a table. Well, I don't know. I just want the assistant to come up with something. So here, he start with a really classic simple bar chart, but then this would inspire me and we I can keep collaborating with the assistant to improve this graph and really get to the vision I have. So here I ask to add to change the color of the expenses bar, set them negative, and add the inflows. And here, step by step, I can keep on collaborating, with my assistant. And this is really super powerful to have this, human machine interaction and collaboration, about the the graph and visualizing my AGCap data without having to import, export, or copy pasting any, files or context in another external tool. Awesome, Kelly. Really, really cool examples of how, AI can support your analysis there. Conscious of time now, guys. We've, been going through some really, really cool examples. Kelly, as somebody kind of you're working who's working with AI every day, who's building AI into these products, do you wanna just share quickly with audience your view on where you think AI is going to help finance professionals going forwards? I believe AI will have a transformative role in the finance professionals' daily work. At AGCap, we see AI evolving rapidly, and we want to be part of this new era. Our mission is to test, iterate, and integrate only the most valuable features to our users. So we multiply projects, ambitious explorations. We run experiments. You have seen that every beta test, beta project I showed you. So we give those to our clients. We gather user feedbacks, and the objective is really to refine our integrations to ensure real impact. Okay. Very, very clear. K. Thank you. And, Nicola, big question for you as well, definitely, as the person who's seen full range of different tools, has got the most knowledge, obviously, on on the depth of where I can go. I guess the question for our audience that would be most relevant is what does it mean for their own jobs? You know, which skills should they develop to secure their careers? Where would you recommend people, target their self development? So let me tell you a story. In the AI finance club that I built, I have so it's a community of, CFOs that are interested to learn about AI and to use AI. And in one master class, we are talk talking about that, like the future of finance. And the ex CFO of Oracle is actually in our committee, Jeff Epstein. And he said, you know that the role of your team is not only to do the task. You should have in all of the job descriptions two roles. Do your task, but also work on improving how to do this task. And if we look at this one, basically, if you have tools, techniques, processes that are available to improve the way you do your task, meaning you are going to do it faster, better, then you need to do it. It should be part of your responsibilities, not only working the same way like two the last ten years, but actually working on getting better. And one example I can share is something that is getting now, like, it's exploding. And when it's exploding, you will hear probably that in the next, months. Those are the agents that you can code yourself. So it's not like you need IT team. You you don't need a tool like, JGPT. You just need to have access to an agent that you can code yourself and where basically you say what you do, but now you are coding it and it's not coding with real code. It's just like drag and drop button. Let me share with you an example visually because I can talk for years about that, but it's better to talk visually. So this is something we did during our, AI finance club master class where, one of our expert, Tobias Swinman, showed how you can automate the process of expense categorization. So imagine you have normally your general ledger, so or all of your transactions in a Google Sheet or Excel file. And then you have another tab, normally, that is just your categorization where you map everything. Well, you can use this agent here where you are going to put your data inside OpenAI. So inside, it's like ChargeGPT, but just like another way of, sending your data to OpenAI. And then you are going to give this prompt where you see your role is to classify the credit card transactions. And, here are the types of tasks that you have to do. And once you have done this, you your output will be in a Google Sheet or Excel file. And when you give that to OpenAI, OpenAI will give that back to you in a Google Sheet that is finished with everything mapped for you, so all the transactions. And that takes maybe, like, five to ten seconds to be processed. So this is really something you can do today already. It's not like a futuristic. You can already do that. I have myself, So that has nothing to do with finance, but myself, I have bots. I can show you that are for example, this one is mapping all of my emails and saying if it's an email to read, if it's an email to, to answer, if it's an email for my team. So all of this is doing that by having AI read my emails and then categorizing the emails. And then I have other agents that also prepare my meetings. I have agents that draft some emails. This is available already today. Wow. I think that, you know, is potentially the most, impressive thing we've seen today, simply for how easy it looks to use, to get started. Cool. Well, thanks very much for those two answers, both Kelly and Nicolas. We are coming very shortly to the q and a, but I'm just gonna share a couple of exciting, AGI Cap updates that we have coming up. If, I can reshare my screen. So for those of you that want to learn more about AduCap after this call, you can, this is sort of shameless self promotion I've become accustomed to. You can book your own call, with me, or connect with any of us on LinkedIn. Nikola has over a million followers, who you can connect with. You can follow him, Kelly, and also myself on LinkedIn. But much more excitingly, for, AduCap days. So AduCap days is AduCap's own in person event where you can see many of our AI initiatives and the rest of the platform in person as we come together with communities in five German cities, six French cities, four Italian ones, and then also London, where I'm based, to talk about key treasury topics, and show AGICAP being used in real use cases along with many of our partners, sponsoring all of these events at the bottom. It's a big in person event. We'd love to pick up with many of the people. If any of, you on this webinar recognize any of these cities to be close to home, please do sign up via the QR code. We'd love to meet you in person, discuss the biggest challenges that you face as finance teams, how you're solving them, and dig, deeper into treasury and AI topics with you, in person. But now important to come to our q and a as we've had a host of questions build up over the course of this session. I see one of them's even assembled four votes, so I'm just I think it might be good five votes. So it might be good to start with that one, Nicolas. Executive management is afraid of using sensitive company data to LLM to train LLM models. What tools exist to securely protect data as we train internal models? So number one, it's important to know what are your security requirements. If you are a company that is producing bread in just a small town, you don't have the same risk than if you are a big hospital handling medical records or even worse, especially now with all of the political topics. And I had to work for this type of company. If you are a defense company handling nuclear weapons, like, all of this needs to be really secured. And I remember we had a room where nobody could enter, and there was, like, magnetic walls everywhere. And so that's not the same type of security you need for this type of companies. So one is understand that. If you understand that, then you can assess if the security standards that are proposed by the LLM, if they are in line with your standards. And this is not hard to find out. This is actually everywhere written on the website of, for example, Microsoft or OpenAI. And Microsoft says clearly that the Copilot has the same security standard than all of the other Microsoft Office tools. So if you already use Microsoft for your emails, then you are already kind of aware that this is enough for also other topics, and that you can use that for, also for Copilot. And I'm not linked with Microsoft, but this is really clear and easy to understand. If you look at OpenAI, if you take something like JGPT teams or enterprise and only those two because the pro, they don't have the same type of, commitments. But the team and enterprise, they have SOC two standards, and a lot of, data cloud solutions don't even have SOC two. So those are already good standards, but I remember my previous company, the defense company, they will never work with, these standards for the nuclear topics because they it's too risky. But, again, not every company has to deal with nuclear topics or with defense topics. Most of the companies, for them, is enough to use these standards. And even if you are not comfortable with this, so if you really have a problem, with giving your data away, you have now with Meta, with Mistral, with, even now, like, the the Chinese one, but I will not personally, I will not trust that yet. But I will trust if we are in Europe more like something like Mistral or if you're in The US, Meta, you can download the model and run it on your server. The output is at 90% the same than any other models because, it's it's just about how you prompt and what you ask. So you have a lot of solutions right now. So it's good to be afraid, but, like, really transform this fear into actions. Don't stay at the fear, position. Go go into actions and choose what is the best solution for you. Yeah. I think, definitely, worth, transforming that fear into action given some of the the power of some of the examples we've seen today. Another really good question, I thought was put through was, obviously, AdiCap's not the only company out there embedding AI into its products. Lots of products doing this, and, indeed, AdiCap will use HubSpot. They're embedding Copilot into the product. What is the right approach for businesses to take? Is it to try and build some stuff in house for different processes, or is it to, really assess tools through how AI friendly they are, or is it to start from scratch? What would you kind of recommend for businesses when it comes to embedding AI into their processes? Should they try to build some of the stuff in house? Should they, outsource it all to big LLMs? How can they, come up with the right strategy to bring AI into their processes and softwares? So, there was also a question, like, about SAP. I think we can answer with this. You see, for example, that SAP is bringing AI into, its ERP, and the AI is called Jewel, so g o u l e. That is available in the cloud, but not yet on premises, and you'll see that more and more. So you don't need to build something that is going to be there really soon. But Jewel maybe will not have access to your mailbox. Maybe Jewel will not have access, to some of the processes you have built with automations. So that's where, like, what I showed you with, like, local automation could be a good alternative. And you will have unfortunately, like, there is not, like, one good answer, but I will first make sure that everybody has LLM access for everything that is like writing, brainstorming, prepare analysis. Then once you are more with a repetitive task, I will build this kind of mini agent, like I did. And then if you are so so if it's for you and for a mini team, like a small team, and then if it's more at enterprise level, I will look at one tool that does it. Or if it's really so tailored that you need, to have a consulting company, then bring a consulting company and make it build for you. But that will be the last solution because, that's the most costly, and also you need to maintain it. Okay. But it sounds like companies should always experiment themselves a bit, as well as looking to embed solutions or web. Big software providers like SAP are already bringing in, AI. They should look at what it's doing effectively. Another question that came up was comparison between these different, models. I think there's a question around Copilot versus ChatChiPT Pro. You've also mentioned the name of some other tools, and we mentioned Claude at the beginning. Do you see a big advantage for using any particular LLM out there, or do you think the outputs are more or less, similar? I don't know if you, heard the news, but the last days, there was an interview from, the CEO of Microsoft who basically, like, launched a bomb because, normally, they have a big partnership with OpenAI. And he said that LLMs are going to be a community. And we see it, like, so many new models, like the price going down and, all of these models come coming from everywhere. And I can see that now, like, it doesn't matter where I go when I prompt it the right way, I get always a good result. And that's really important for everybody to understand that knowing how to prompt is like knowing how to drive. If you don't know how to drive and you have a Ferrari, you are not going to beat a formula one, driver who has a really bad car. He's going to or she's going to kill you all the time and would be much faster. But if you know how to drive, then you with a ferry, you might compete. And so first know how to drive, know how to prompt, and then most of those LLMs will, help you bring to the right, to the right point. But what is going to become more and more important is how these tools are integrated. And if you see Copilot compared to ChargeDpT, for a company, if you already use Excel, Outlook, PowerPoint, SharePoint, Azure, Then if you have Copilot, it's much easier to get benefits because when you prompt, you get access to also your internal information. And if you use ChargeGPT, ChargeGPT is connected to nothing, so you are more and more Copilot is going to be more valuable because you added the company layer, of information. And that's where, like, yeah, OpenAI needs to to see how they are going to keep their edge, but LLM will be a community. Okay. Interesting. So you don't think, if I I've I've got Microsoft across my products, so Copilot is advantageous because it has access to my internal company data. You don't think there's a significant advantage to bringing in a different kind of model, say, Claude, and using it on my internal company data? No. It's too much work to connect it with it. Okay. Very interesting. Very interesting. Have we got time? We've got forty four seconds left. Have we got time for one more question? Is there any question that really stood out to you in the q and a, Nicola, that you would like to answer? I mean, we if we can stay another five minutes, I can show a use case. Let me show there was something about how to do, functional analysis and scenario modeling because I prepared this, actually. So, yeah, can we make a costing scenario estimation based on existing plans? So I have something similar that I've prepared. That was for somebody who is in my AI finance club who asked how he can show to his client dynamic business models because this, this member from the AI finance club is a fractional CFO, and he wants to use AI to build scenario model for a company who is a retail store, so selling clothes. So I went into cloud, and here, that's where you see, like, that's a functionality that you have in cloud, but you could have that, and you have that now more and more on JPGPT and all of the others. So cloud still have a small edge here because this functionality is very nice. So I just ask, can you create a dynamic model? And here, what is really cool, Claude is writing a code that takes in account all of this parameter. So a model that takes in account sales per store, the margin, and fixed cost, and then build three different scenarios. And as we know with financial storytelling, the best way to compare scenarios is actually not to have tables, but to have a graph. And here we can see the three different scenarios. I can open that in a web page, and this web page, everybody can use it if I just share the link. And now I can go in this web page and just change the parameters. So, for example, I can change the margin, and automatically, the scenario in the bottom will change. And now I can compare the scenario really quickly, having AI built the model for you and then me after playing with the scenario in a way where if you want to do that in Excel or in a web like this, you will need one day to program all of this. Wow. Again, I think really, really incredible example. Very nice interface, I think, Claude has, there as well, which makes a big difference. We have many more questions to keep answering. Another one, I guess, this is all kind of on the self development point. What can finance professionals do to stay updated as well as not being overwhelmed with the upcoming tools? I think that's a fair point. You know, there's so much information out there. There's so many different areas where AI feels like it's gonna have big impact starting with those repetitive tasks we talked about at the beginning through to the kind of analysis you just showed us. If there's one thing, or two things that you would say to finance professionals to really focus on other than becoming LLM experts, what would they be? This is where you need to force a bit the nature for yourself, because it doesn't come from, even from the from from the sky. First, we have enough people on LinkedIn or YouTube or Twitter that are talking a lot about it. I have a YouTube channel as well. So you can start there where you can follow the right people. That means on your time, you need to allocate time for this. You need to find the right information. So that could take you time, but that's free. What I recommend as well is either within your company. So check with your manager to organize a training on these tools. So that's what I also do. I give trainings on these tools. And the third part the third option is what I built, is the AI finance club, where basically every week, we have a content from experts where we explain how to use AI for finance. So this what I just showed you was one lesson. We have this week somebody explaining about how to use AI agents. We have a master class, about how to build AI agents. We had last, months, something about how to make data analysis with AI. And so this is a great place where you can, learn, and that's basically a bit of discipline to decide for yourself. K. I want to learn it, then you need to commit on something. And once you commit, then you will be on the top 1% because a lot of people stop at committing, time and money, for this. Okay. So make sure you you do the self learning on top of on top of the day job. One question I think, sort of stood out here is, at the moment, a lot of companies use budgeting tools for the kind of financial analysis, you did. Do you see AI as a big risk to those softwares? Because, you know, if you can just get Copilot to do it also, why do you need a a a budgeting tool? So the big advantage of a budgeting tool is one, if you need collaboration. Because usually in finance, you're not the only one budgeting. You need all of the other departments. And a lot of FP and A tools, they bring that. Second, if you need to have the data pulled from the system. So if you, if Excel is too big, if you have too many, too much data, then all of these tools normally are connected with your systems. And then the third is to document actually your budgeting process. Because here, you can go in the chat and play, but you cannot document a budget process where you have all of your head counts, where you have your different scenarios, where you have, all of your business rules. All of this then is documented in a FP and A tool, and then you can always come back, tweak some changes. You can audit what happened. And those are the three reason why if you are big enough and if you have these constraints, then you need a FPM tool or budgeting tool. But most of the company, they don't need collaboration because the finance is doing that with three peep other people, like maybe head of sales and the head of h HR and then head of production. So just like one Excel file with SharePoint is enough. Most company don't need to audit because they don't have, like, ten months of budget with 10 different, variations. They only have, like, maybe two or three, which Excel is again enough. And foremost of company, they don't have so much data, so Excel will be also okay. But if you have these three or two of these three, I will consider AFPneto. And by the way, they start integrating AI as well. Yeah. Of course. Okay. Very clear. So once your brother's in size, then it makes sense. And your teams, as they grow, the value of tools goes up, or specialized tools goes up with that. We have a few more questions. Have have you still got time, Nicola, to to keep running? I'm conscious we're already running over seven minutes. Here we have one around PDF documents that contain both qualitative and quantitative data. So I think Kelly showed a few examples of how you can easily extract, qualitative data via text. Which AI tools maybe we can reframe this as our best at consolidating quantitative and qualitative data into Excel files and then running the analysis. Are there any leaders in that space? Really hard to say which one is the best because it's changing all the time. What what I like from, OpenAI from ChargeGPT is the code, interpreter. So, basically, where when you upload a a file, you have Python that reads the file. And Python has some program to read PDF files, to read Excel files, to create PowerPoint document, to create PDF. So this is, powerful, but it's still limited because if you have a huge PDF and the data inside is, like, not well done, it will not be as good as a clean CSV file with, headers and proper lines. So we always teach that you need to have clean data and to start an analysis with AI. And if you don't have clean data, what you can do you should not say, oh, I cannot use AI. What you can do is you can use AI to have a mini script that is doing for you the clean cleaning of data. So it could be you have a PDF, you extract the data using a a Python script. So with tyceract, you can do that. It's a a Python script working with PDF. And you make just you ask your LLM, JGPT, or Copilot to give you the script, and then you can do that. And then after, you can say, okay. Now, like, remove this column, combine these files. And once you have the data finished, then you can go and start your analysis on it. Super clear. Thank you. And then I think I know kind of we already covered security, but I think, this one might be worth just clarifying a bit more detail for audience. When you can you just reassure our audiences our audience that when they're putting internal company data into any model, it's not being shared with the Internet world, effectively. So OpenAI doesn't own the data, and then take that back and feed it into the rest of its models. How do how can businesses who might be sharing, you know, their bank data or other very sensitive data into these models, even if it's internally owned, feel reassured that that data doesn't get shared with the Internet world as it's framed phrased here. Yeah. You need to look at the plan because if you take the Chargebee T free, for example, you cannot really control that. Like, you the control of our data is not that good. If you take Chargebee T pro, you need yourself to go and deactivate the data training on your data. And if you're on pro, it also doesn't say that, it's, like, enterprise level because it doesn't say SOC two and all of this. Just say we don't train on your data. If you go on Chargebee teams and enterprise, then there it says you have, we don't train on your data, and all of our system are covered with SOC two. It doesn't mean and, like, the same way then if you use Google, Outlook, or AWS, or any tools, it doesn't mean that it will never be hacked. But they have processes inside that have this standard, of security. And then that's where depending on your company, you need to know what is your standard. Right? Again, if you are working for a hospital, you have medical records, maybe, the standard of security is so high that is that Google or JGPT or Microsoft is not good enough. Maybe you need, like, a specific encryption in between. But then that doesn't mean you cannot do it. You just need to know what to do in between. And that's where a lot of people because it starts to be complex in their head, they prefer not to do anything. But it's only like you look at the program, you say, okay. What is my company requirements? And then based on this, I compare if the LLM that I want to use respect these requirements. Yes or no? And once you do that, and you can do that with your IT team, then you know where you can go. But don't stay at the at the step that you just have a security problem and you don't know how to solve it because it's just to compare. The the effort is just to compare, and that's not hard. Like, you can you read a lot of people, have done studies. You can go to your IT. You can go to a lot of companies that I know that are big companies. They are using all of these LLMs. So if big companies are doing it, without specific, I would say, like, encryption, that means that, for them, it's already good enough. Clear. And, I guess the key takeaway is, the level of security corresponds very closely to your plan, that you take out. Okay. Thanks so much, Nicolas, for joining us on this webinar today. Then we'll wrap it up there. Kelly, thank you for the examples you showed of how AI is being embedded into Adjacap's product. For those of us who for those of you who want to learn more about AI, follow Nicolas on LinkedIn. His posts, are frequent and very, very useful. I often, save them myself, and apply those examples. And for those of you who want to learn more about AGCap, feel free to get in touch with me by requesting a demo in the top right hand corner, of this page. Thanks so much. We will see you again for one of our next webinars very shortly. If not, at HD Cap days. Thank you, everyone. Thank you, everybody.