How to automate investment memo writing with AI
Paul Omenaca
@houmland
Investing in established companies can offer significant returns, whether by transforming a seasoned business for additional growth or capitalizing on existing market share. However, analyzing such opportunities typically requires synthesizing massive amounts of information—from shareholder letters and due diligence questionnaires to market outlooks and performance sheets. This process often runs 118 hours per deal, eating into resources and extending decision timelines.
That’s where AI can help. By automating first drafts and populating repetitive, structured sections of the memo, AI not only speeds up the creation of comprehensive reports but also provides an overarching view of the target company’s viability. This can reduce the memo-building workload by up to 40%, potentially saving over 42 hours per opportunity.
In this article, you’ll learn how to construct an AI investment memo generation tool that aggregates all the data about a potential private equity purchase. This will help you focus on the strategic insights and determine whether it’s worth pursuing.
How to build an AI investment generation tool
- Using the AI investment memo generation tool
- Add inputs
- Set up dynamic vector stores
- Plug in data sources
- Drop and link LLMs
- Publish the tool
- Set up the user interface
- Share it with your team
- Keep track of usage and analytics
- Improve your AI investment tool
Using the AI investment memo generation tool
Here’s how the AI investment tool will look like once you complete this tutorial.
Your team will be able to:
- Type the name of the company they’d like to research.
- Write additional instructions to steer the generation process.
- Add URLs that contain valuable information about the company. The tool will automatically visit those links and extract the data from them.
- Upload internal reports with company data that should be included in the memo.
Once the process is complete, you can download an AI-generated report based on all the information added and data available from your internal sources.
1. Add inputs
Reports depend on the amount and quality of the data you ingest into the AI models in the project. Since creating an investment memo requires breadth and depth, we should add multiple inputs to let your team control how much data they’d like to include into the report.
1.1 Adding a company name
The user will enter the target company name in the first input. We’ll use the input node that’s already present on the canvas.
Due to this project’s complexity, we’ll use Stack AI’s organization features. Click the edit node name icon at the top right of the node.
Change the node’s name to Company name and click Save.
1.2 Setting additional instructions
Your team may want to analyze a specific angle of the company data instead of creating a general report. Click the Inputs section to expand it, drag and drop an Input node onto the canvas.
Click the edit node name icon and change it to Additional instructions.
1.3 Uploading reports
Some company information may already be compiled in existing documents and files. Instead of copying and pasting the contents into an input field, we can add a tool to upload them. Click to expand the Input section, drag and drop a Files input node onto the canvas.
Make sure this file input is exposed in the user interface by toggling the Expose as input setting.
The name of this node can’t be edited, but we can add a note. Click the note icon on the top right of the node.
This opens a text area. It supports Markdown syntax: you can use it to format text. Write # Upload reports. The hashtag will set it to an H1 header, making it easier to see.
1.4 Including URLs
A part of the research might involve searching the web and finding pages that contain company data. Again, instead of visiting those pages, copying and pasting the content into the tool here, we can let your team paste the URLs—Stack AI will scrape the content and make it available in the report. Click to expand the Input section, drag and drop a URL node input field onto the canvas.
Untick the Enable URL as input checkbox. In this case, this will disable the node’s input handle within the project—we won’t need it.
As with files, the URL node’s name can’t be edited. Click the note icon on the top right of the node and write # Reference URLs.
1.5 More inputs
We’ll use 4 inputs for this tutorial, but you can add as many as you need, depending on the data types and instructions you want to pass into the AI models. Stack AI supports inputs for text, documents, URLs, audio files, images and YouTube videos.
_
2. Set up dynamic vector stores
In simple terms, a vector store is an AI-friendly database. It can store large amounts of data and retrieve the most relevant chunks based on a user prompt, forwarding them to an LLM to ground the response. Since the inputs we just added to the project may contain a lot of data, we’ll use a Dynamic Vector Store. This node was developed by Stack AI to quickly store and return the most relevant data based on the prompt.
On the left side menu, click to expand the Dynamic Vector Stores section. Then, drag and drop a Basic node onto the canvas.
The company name and additional instructions inputs shouldn’t contain a lot of data, so we don’t need to connect them to the vector store. Instead, connect the Files input field to the dynamic vector store’s data loader handle.
All uploaded files will be stored in this node. We’ll need a second one for the URLs, since the web page data can be extensive. Drag and drop another Dynamic Vector Store Basic node onto the canvas, close to the first one.
Connect the URLs input node to the second dynamic vector store’s data loader handle.
Both vector stores need a prompt to extract the most relevant data. To simplify the user experience, we can add a default prompt to retrieve the same kind of every time. Click to expand the Utils section, drag and drop a Default Message input onto the canvas.
Write the following in the Default Message node:
Extract company financial information, risk factors, product information and management remarks.
You can change this prompt depending on what kinds of information you want to extract from URLs and your documents. If they’re widely different, you can use two Default Message nodes with two different prompts instead.
Connect the Default Message node to the input query handles of both Dynamic Vector Store nodes.
3. Plug in data sources
Now that we’ve handled the user inputs, it’s time to connect other data sources to the project. For this tutorial, we’ll connect a web search and a Microsoft Sharepoint instance. You can connect as many data sources as needed to bring in all company information without exporting it from the platforms where it’s stored.
3.1 Starting a web search
On the left side menu, click to expand the Knowledge Bases section. Then, drag and drop a Web Search node onto the canvas, close to the Company name input.
This node needs a keyword to understand what to search for. Connect the company name input to the Web Search.
Connecting Microsoft Sharepoint
In the Knowledge Bases section, drag and drop a Microsoft Sharepoint node onto the canvas.
This node integrates with your Sharepoint instance, offering similar functionality to a dynamic vector store: Stack AI created a proprietary algorithm that efficiently searches for the most relevant information based on a prompt.
Read the documentation for instructions on how to connect Microsoft Sharepoint to Stack AI. When fully connected, you’ll see your Sharepoint resources inside the node.
To simplify user experience, we’ll add a default prompt to retrieve information from Sharepoint. On the left side menu, click to expand the Utils section. Then, drag and drop a Default Message node onto the canvas.
Connect the Default Message to the Sharepoint node.
Here’s the prompt that we’ll use for this tutorial:
Extract market growth, inflation, macroeconomic conditions.
This prompt is highly dependent on what your data source contains. If you’re storing other kinds of information here, consider adjusting this prompt to surface them instead.
Write the prompt in the Default Message input field.
Other data source connections
Stack AI offers a wide range of data connections with major platforms—you can see the full list in the Knowledge Bases and Databases sections on the left side menu.
Beyond Sharepoint, you can also bring in data from Google Drive, Dropbox, or AWS S3, for example. You can integrate all your data sources and reuse them across projects. For additional organization and security, you can create knowledge bases inside Stack AI and assign files and folders to each one, setting role-based access control for each.
All these nodes include the Stack AI proprietary search algorithm we mentioned earlier. As such, they can’t be connected to a dynamic vector store: they’re ready to be connected to LLMs directly.
4. Drop and link LLMs
In this tutorial, we’ll use a linear AI generation architecture, a chain of 5 LLMs, each tasked with writing a section of the investment memo.
4.1 LLM for executive summary section
On the left side menu, click to expand the LLMs section. Drag and drop an OpenAI LLM node onto the canvas.
The node drops on the canvas with GPT-4o mini selected as default. Click the dropdown and change it to GPT-4o for its advanced reasoning skills.
Let’s connect the inputs and data sources to this node. Start with the company name input and connect it to the OpenAI LLM node.
Connect the additional instructions input to the OpenAI LLM node.
Repeat for the Web Search node.
Continue by connecting the vector store containing the files to the OpenAI LLM.
Next, connect the vector store containing the scraped URL data to the Open AI LLM.
Finish by connecting the Microsoft Sharepoint node to the OpenAI LLM.
We can’t change the name of LLM nodes, but we can add a note. Click the note icon at the top right of the node and write Executive summary.
Copy and paste the following into the instructions input field:
_You are an AI assistant helping draft an investment memo for a PE fund._
_\- Write in bullet points. Only return the bullet points._
_\- You will focus on one section only._
_\- Be verbose._
_\- Your section is Executive Summary. Draft just that section._
_\- Don't mention the section you are writing as a title._
_Add citations at the end as: \[1\] www.google.com, \[2\] document.pdf_
Next, copy and paste the following user prompt into the corresponding input field:
_Company to analyze: {in-0}_
_Additional instructions: {in-1}_
_Market information to use: {knowledgebase-0}_
_Web scraped content: {vec-1}_
_Additional documents to be used: {vec-0}_
_Web search: {websearch-0}_
The values in curly brackets are variables. They’ll be replaced with the content generated by each corresponding node. If you add more data sources in the future, remember to connect them to the LLM and then add their variables to the prompt.
LLM for company overview section
Drag another OpenAI LLM node and drop it to the right of the first one.
Connect the first OpenAI LLM’s output to the second’s input handle.
You don’t need to connect all the previous data sources and inputs to the second OpenAI LLM. They’re all available through the connection to the first LLM.
This new node will work on the company overview section. Copy and paste the following system instructions onto the corresponding input field inside the node. Notice that it’s the same as the first one, except for the bolded part:
_You are an AI assistant helping draft an investment memo for a PE fund._
_\- Write in bullet points. Only return the bullet points._
_\- You will focus on one section only._
_\- Be verbose._
**_\- Your section is company overview. Draft just that section._**
_\- Don't mention the section you are writing as a title._
_Add citations at the end as: \[1\] www.google.com, \[2\] document.pdf_
The system instructions will already change the focus and behavior of this new OpenAI LLM. All that’s needed is to pass the variables with the data. Copy and paste the same user prompt into the corresponding input field:
_Company to analyze: {in-0}_
_Additional instructions: {in-1}_
_Market information to use: {knowledgebase-0}_
_Web scraped content: {vec-1}_
_Additional documents to be used: {vec-0}_
_Web search: {websearch-0}_
_
Edit the second OpenAI LLM’s note to display Company overview by clicking the corresponding icon at the top right of the node.
Finally, open the dropdown menu and select the GPT-4o model.
LLMs for the remaining sections
You can repeat these instructions as many times as needed for all the sections. As you do so, update the line in the system instructions that refers to the section that each LLM should focus on.
Here’s how this architecture would look like for an investment memo covering:
- Executive summary
- Company summary
- Market opportunity
- Investment recommendation
- Summary
Finish with an output node
Once you complete the LLM chain to create all the sections, connect the last LLM to the output node.
To include the results from all the LLMs, you can configure a template for the output. Click the settings icon at the top right of the output node.
Click the Add template button.
You can structure the output here and include any variables to populate each section. Here’s an example template you can start from:
_Investment Opportunity in_
_{in-0}_
_Executive summary_
_{llm-0}_
_Company Overview_
_{llm-1}_
_Market Opportunity_
_{llm-2}_
_Investment Strategy_
_{llm-3}_
_Summary_
{llm-4}
This node supports Markdown syntax, so you can turn each section name into a heading. You can do so by typing the hashtag before each header (#) or using the rich-text formatting tools at the top of the input field.
5. Publish the tool
On the top right of the screen, click the Publish button.
Every time you make changes to the workflow of the project, remember to click Publish to save them and make them available to your team.
6. Set up the user interface
Click the Export tab on the top left side of the screen.
Stack AI exports new projects as a Form interface by default—even though there are others, such as Chat Assistant or Batch. For this project, this is exactly what we need, so you can leave the default as it is.
Add a custom domain to make the tool easier to access. At the same time, you can add a name and a description of the tool to help your team understand what it can do and how to interact with it.
In the Fields section, you can show or hide input fields in the interface. In this case, we want all of them to be showing. Make sure all of them are activated by ticking the checkboxes. To improve user experience, change the Alias of each input to reflect what kind of information should be added. In this case:
- in-0 as Company name
- in-1 as Additional instructions
- url-0 as URLs for scraping
- doc-0 as Reports
- out-0 as Investment memo
Since this interface integrates with your internal data sources, you need to keep it secure from unauthorized access. Beyond high privacy and security standards already active at the platform level, Stack AI offers security controls to protect these tools. In the Security section, you can set password lock, SSO protection for organization or specific emails, and whitelist web pages where you’re planning to embed this interface.
When you’re done making changes to this interface, always click the Save Interface button at the top-right corner of the screen.
7. Share it with your team
Sharing this project with your team is as easy as copying the link at the top of the preview window and pasting it in an email or internal communication channel.
You can also share a copy of the project with other Stack AI builders on your team. Click the Share button and add their emails. They’ll be able to see and experiment with the project, but no changes they make will reflect on your project.
8. Keep track of usage and analytics
Stack AI keeps track of usage as your team interacts with the tool. Click the Analytics tab at the top-left side of the screen.
From top to bottom, you’ll be able to:
- Filter results in a date range
- Generate an AI report or download logs as a CSV
- A breakdown of runs, users, errors and tokens
- A list of recent tool runs. You can hide or show columns using the dropdown on the top right of this table
9. Improve your AI investment memo generation tool
After you share the AI investment memo tool with your team, keep an open line of communication to receive feedback and understand how to improve it. This can include improving the system and user prompts, adding more nodes or reconnecting them differently.
Beyond these changes, you can keep up with the latest AI models as they’re released. Stack AI makes it easy to upgrade. Simply click the settings icon in each of the LLM nodes…
… and change the provider and model dropdowns to do so.
Remember to run a test of the newer model to see how it changes the tool’s functionality, and click Publish to push the changes to the user interface.
Wrapping up
Creating a thorough investment memo requires a lot of research, plenty of analysis time and gathering all the details in a single place to help you make a decision. While AI can’t decide for you, it can help you gather, organize and explore all the data you need. This way, you don’t have to waste time in rote research tasks, instead freeing your mind to weigh whether the investment at hand is a good choice.
But this is just the beginning of all you can automate with Stack AI. Create a free account and explore our other tutorials: