# 3 Claude Agents Every Finance Analyst Should Steal (Full Prompts Included) Source: https://qvantx.com/insights/financial-ai-agents-real-world-use-cases-prompts/ Author: Ani Björkström — QvantX Sweden AB, Stockholm Published: 2026-08-13 Updated: 2026-09-20 Video: https://www.youtube.com/watch?v=7KC1gV2BWdQ (17 min) Topic: AI in Finance License: free to quote and cite with attribution to https://qvantx.com --- _AI IN FINANCE_ ## How can AI agents automate real finance workflows in an enterprise setting? #### Key takeaways - Three AI agents built for actual finance clients — a due diligence analyst, a scheduled news brief, and an Excel bug-finder — run in real enterprise environments. - A scheduled agent compiles the top five sustainability news items every weekday at 7:00 as a Gmail draft, replacing a client's 5 a.m. office routine before her 9:00 briefing. - Claude inside Excel traced a NAV discrepancy to a formula summing B11:B13 instead of B11:B15, fixed it on approval, and cut the difference to 0.00001 — within the 0.01 tolerance. Most AI-agent tutorials on YouTube are built on private computers and would never survive a real enterprise environment. Here, Ani Björkström, a tech consultant based in Stockholm, shares three finance agents actually delivered to clients, including the prompts behind them. ### What does an AI due diligence agent actually do? A due diligence agent reads fund documents the way a senior analyst would and answers investor questions using only those documents. The first demo builds one for a Sweden-registered equity fund classified as Article 9 under SFDR, with a prompt structured in four parts: who you are, what I need, how to do it, and rules and constraints. The agent lives in a Claude project rather than a chat. A project has a larger context window and persists, so users can return the next day without losing anything; once the answers satisfy the team, the knowledge is saved as a skill and reused on other funds. ### How do you put a morning market brief on autopilot? Create a project with a four-part prompt, connect the Gmail connector, and schedule the run for every weekday at 7:00. The client had to present 24 hours of sustainability news at a 9:00 daily brief, which forced her into the office at 5 or 6 in the morning; the agent now finds, summarizes, and drafts the news for her commute. Precision comes from iteration with the actual user, not from the first prompt. The brief was scoped to the Nordic market at the client's request, and instructions were updated after each review round until the agent answered exactly the way she wanted. ### Can AI find a real bug in an Excel model? Yes — Claude in Excel located the source of a NAV discrepancy between the front-office model and the fund administrator's figure. NAV is calculated daily by both front office and back office, and because it is published to investors, any per-share discrepancy must be resolved fast. Claude reads every formula in every tab, so one prompt asking it to trace the model found that total liabilities summed only rows B11 to B13 while rows 14 and 15 had been added without updating the formula. After a manual approval, it corrected the sum to B11:B15, leaving a difference of 0.00001 against a tolerance of 0.01. | Agent | Setup | Output | |---|---|---| | Due diligence analyst | Claude project referencing a local folder of fund documents | Answers on objectives, risk category, and annual costs from the documents only | | Morning brief | Project with web search, Gmail connector, weekday 7:00 schedule | Top five Nordic sustainability news items as a Gmail draft with sources | | Excel bug-finder | Claude opened inside the Excel workbook | Root cause of a NAV discrepancy plus the corrected formula | #### FAQ **Which model should a document-extraction agent use?** A cheaper model is often enough — extracting data from given documents is not the most advanced thinking task, so a lower-tier model performs well at lower cost. **Should agents approve their own actions?** Start with manual approval, where Claude pauses before each action, and shift to automatic approval once you are comfortable with how the agent behaves. **Does Claude in Excel require the desktop app?** No. The demo runs on Microsoft 365 in the browser, and the same setup works on Excel running locally on your computer. ## Full video transcript Chapters: 0:00 Intro: The Problem with AI Demos · 0:49 Use Case 1: Read a Fund Like an Analyst (Due Diligence AI Agent) · 2:48 Use Case 2: Daily Brief AI Agent · 4:59 Use Case 3: NAV Reconciliation AI Agent · 7:15 Key Takeaways & Conclusion [0:00] [music] [music] Hi guys, my name is Annie. I'm a tech consultant based in Stockholm. In this tutorial, I will show you three actual user cases where IIE is used for creating financial agents in financial industry and those solutions are delivered for actual clients. There are a lot of amazing information in YouTube but most of those are created by individuals from their own private computers and most of the times when I watch those YouTube context I understand that those will never work in actual enterprise environments. So that is why I decided to show and share three actual user cases that I created for my clients. [0:46] Let's get started. For the first example, we will create a DD agent that reads a fund like a senior analyst. Now let's try to understand what is a DD agent. The due diligence process is a process that every investor or investor company does before understanding if they want to invest in a company or not. If we go to the cheat sheet, here is the prompt that we will use for creating this agent. Uh you can see that the prompt is divided in four parts. We are answering two questions. Who you are? You are a senior investment analyst supporting due diligence on the hundles bank and hold energy fund. A Sweden registrate equity fund classified as article 9 under SFDR. What I need rounded due diligence analyst of this fund. [1:34] Then how to do it? Apply an equity fund due diligence lens throughout. Next rules and constraints. Reference only the uploaded documents. to not bring in outside information unless I explicitly ask. So when creating this prompt, I normally sit with business people who will use this solution and we are defining their specific needs so that the agent can answer and reply as the business person wants. So here let's try to copy this prompt and go to cloud co-work. Now let's open cloud co-work. Then we will go to project. So the difference between a project and a chat is that the in the chat you don't have as big context window as you have in the project and also if you are creating something that you will come back to for example in this case when we create a due diligence agent maybe you will come back to it and ask more questions tomorrow or the day after and that is more difficult to do with the chat because at some point the chat will be limited and you have to create a new chat and all the information and knowledge will be forgotten when you open a new chart. [2:43] So that is why it is important to use project for creating a due diligent agent or for any project where you will come back to. And here we have start from scratch import a project and use an existing folder because we already created a due diligence folder. Then it makes sense to use an existing folder. And from here we will choose the folder. The name of the folder was due diligence and it was in the desktop. So let's click on the desktop and DD ag and here is the folder. Okay. Now you can see that the folder is chosen and it is referenced by the project which means that the project is referencing the local path where all our fold and under that exist instructions. Here we will copy paste the information that we used from cheat sheet and then we will create the project. [3:32] Here under context you can see that it shows all the documents and then we have the memory instructions and here are no scheduled which means that we haven't scheduled this agent. So now once this is created let's go ahead and ask the question. So right now what we did we gave cloud all the documents that we want to use for due diligence and we gave it instructions which means how to think. And now let's go ahead and ask the questions. questions are also in the cheat sheet. The first question is for example this one. According to the key information document which is one of the documents that we give to the agent, what is this phone's investment objective? Now let's copy this question and ask it while cloud starts to work on the question. [4:20] You can see that here I'm using Optus 4.8 and of course you can use Optus 5 or Fable. The reason that I use this is because I feel like this agent just needs to extract data from the given documents which is not the most advanced thinking process and that is why I use Optus 4.8 which is still a great model but is cheaper than the model that comes after Optus 4.8 like F2 like 4.9 or five. So here you can see that it answered the risk category ongoing annual costs and so on. And here if we go back and ask the other question it will answer the other questions as well. And the whole idea here is that once this is done then the business people will sit here and will ask all their questions and based on the replies they will define if they still want to invest in the investment object or not. [5:13] And the investment object can be fund or a private company or publicly traded company or any other investment object. And also you can see that here we have this little button that says cloud pauses. So you can approve each action. So right now if there is something that needs to be approved, cloud will ask us. So we will manually approve it. You can even choose automatically approve when cloud runs on its own and pauses to ask if anything looks unsafe or you can just skip all approvals which means that cloud never pauses even for unsafe actions. So I will advise to use manual approve to start with. Next when you feel like you are more convenient with cloud co-work then you can shift to automatically approve. [5:59] Here we have an answer and you can see that it gave us very when we are happy with all the answers we ask cloud to create a skill on this knowledge and once we do that then we can use this due diligent agent skill on different phones. So if you remember we are using this on a specific handle spanker phone but if we create a skill on this knowledge then we will use it on other phones as well. Great. Now let's go ahead and create a second agent. Just a quick reminder if you're enjoying this video please subscribe and give me thumbs up so that I continue creating similar videos. For the second demo we will create a morning brief on autopilot. If we go back to cheat sheet. So here you can see that we will create a daily sustainable investment digest end to end and schedule. [6:48] So I had a client that told me that every morning at 9:00 she had daily brief with the team and she had to present all the sustainability news for the last 24 hours. However, for doing this she had to come to the office quite early in the morning and be in the office already 5 or 6. So she wanted to automate this process and I suggested to create this I agent that will find all the news will summarize the news and will send the news to her email so that she can read the news on her daily commute to work and do not have to be in the office so early. So here is a prompt and here we also divided the prompt in four different sections. [7:34] who you are, what I need, what to do, how to do it and restrictions. And also you can see that I am specifying an email name and in the client case it was client email because a draft will be sent to client email. So let's go ahead and copy this information. Okay. And once this is done, we will go back to cloud co-work and we will create a new project. So the reason we create a new project even for this is because we will need to schedule this process and we can't schedule a chat. So last time we use an existing folder because we had a folder with all the documents but in this case we don't have any document and we will ask agent to search the internet for the information. [8:21] So we will start from scratch and here we will just copy paste the prompt and we can give any name to this agent. Maybe we will call it sustainable agent. Next we will just click on create and you can see that it was created. So let's go ahead now and copy the questions. This is the first question generate today's brief and once we give this to the agent it will go ahead and will check all the yesi news based on our prompt. So again this prompt was created with a collaboration to the client because she wanted the news only for Nordic market. So we need to be very specific about this to cloud and once we create the first agent we will normally sit with the client and we will see what are the answers because in some cases she will say no I don't like this answer or the answer should be in this way and based on her feedback we will go back and update the instructions in order to get the reply that she wants to get and in the end if we do this several times we will end up with agent that is answering exactly the way that the client wants. [9:34] And you can see that in the right side cloud created a progress bar and there are four different steps. Search regulatory policy news, next search asset manager owner stewardship news, next search corporate sustainability disclosures and rank deduct and format the brief. So those are all the steps that agent will do. And under the context we have web search. So this can take some time and once this is done the next question that we will ask is draft an email with this brief to me and here is my address and save it in Gmail as a draft I can review before sending. So here because cloud needs to connect to my email and send me a draft email. [10:23] It needs to be already connected to my Gmail. Because if you do not connect your cloud co-work with your Gmail, then it can never send you an email. And in order to connect it, you need to go to customize on the left side. And then under customize, you can see that there are skills, connectors, and plugins. So here we will go to connectors. And under connectors, there is the Gmail connector. And you can see that it is done for me. If you haven't connected yet, then there will be the button that says connect and you will click on the connect. A new browser will pop up and you will authorize. So that is how you connect your cloud co work with your Gmail. Now let's close this and go back. We can see that we have an answer right now. [11:08] So the agent found all top five news that we asked for for the last 24 hours. This looks amazing. It is exactly the same format that we asked for. And again, if something doesn't look good, you will just go ahead and update the instructions until you have the answer you want to have. And here now we will ask draft an email with the brief to my email address. So let's try to send this. And once that is done, we will expect cloud to send the email. And here you could see that this is the connector and it will use the tool or the connector to send the draft email. This can take a couple of seconds. And once this is done in the next step, we will go and schedule this because we want to run this every weekday at 7:00 so that the client can read this on her commute to the work. [12:00] Okay, it says that this was sent. Let's go ahead and check if it was actually sent to my Gmail account. You can see that it worked and we even have all the sources as we asked. Great. [clears throat] So now let's go back to cloud and here we can ask to schedule this process. So you can see that it's still you can see that it is still using the connector because every morning it needs to create the draft and send to our email and it still needs to use the Gmail connector. And here we have schedule task daily sustainable investment group. You can see that it was prepared. So we'll click on schedule. Okay, we have an answer. It says the weekday 7:00 draft task already exists with the right schedule. [12:45] You can see that it says that this already exists. It will just refresh it. The brief will be generated every weekday at 7:00. Now let's start with demo tree. Here we will find an actual bug [clears throat] in Excel model. So if you ever worked in financial industry, you know that excels are still very common and in publicly traded funds and now or net asset value will be calculated every single day and this is basically the value for every share that investors will see before investing in the fund. So what is happening? Normally portfolio managers that are working in front office will calculate net asset value. At the same time back office will also calculate the net asset value. [13:30] And if there is some type of discrepancy between the value calculated by front office and back office then it needs to be resolved very quickly because these numbers is published for investors. So if you ever have this situation that we have in here that is discrepancy per share then it is very very important to solve it as quickly as possible and of course you can always go back and try to understand how this number is calculated and try to understand why there is a difference but we don't have the time so that is why it is very very efficient to use cloud in the cell and in order to activate cloud in the excel you can just click on start and then from there you will click on cloud and the chart chatbot will open in your Excel. [14:16] And this chatbot has access to all the formulas and all the information in every tab in your Excel sheet. So let's now try to solve this discrepancy. Here we will go back to our cheat sheet and we will try to get the prompt that I created. So it says the reconciliation cell on the nav tab shows a discrepancy between our model and the font administrators reported nav. trace through the model across all tabs and identify the source of the arrow. Now let's just copy this prompt and let's go back to our Excel sheet and pass it in here. And remember that you can use cloud on the Excel sheet that is running locally on your computer. Right now I opened Excel sheet with Microsoft 365 on my browser but it works perfectly even on your local computer. [15:05] So at the moment it is thinking and now it found what the issue is. It says that on cash liability which is this tab where we are summing the total liabilities in here it says that we are summarizing the numbers between B11 and B13. So basically we are summarizing those three and we didn't involve 14 and 15. And the reason that this can happen is because you normally will add more liabilities and then you will get to update the formula that is in row 16. So cloud found it and here we can just go ahead and fix it or you can ask cloud to correct it. It says want me to correct it and we can say yes please. So now cloud will go ahead and update the formula. [15:52] Okay, it says is now the sum between B11 and B15 and this is the sum. Let's go to nav and we can see now the discrepancy is just 0.00001 which is acceptable because it is very very little and here we have the note our model and the administrator should reconcile to within 0.01 01 which is then fine and this is how cloud can help to fix the issues very very quickly in Excel where you don't have enough time. Those were several user cases that I created for my clients couple of months ago. If you want me to create video about a certain topic, please comment down below and I [clears throat] will create a video. Hopefully you like this video and have a good day. [16:38] Bye.