How to Apply AI in Finance and Asset Management

TL;DR

AI can support finance teams by reviewing due diligence documents, curating sustainability news, and detecting spreadsheet anomalies in NAV calculations. The demonstrated approach starts with explicit, domain-expert instructions, limits analysis to approved documents, and improves outputs through repeated analyst feedback before turning the refined method into a reusable skill.

Transcript

Super excited today to be speaking with Annie. She's an AI consultant based in Europe that works with banks and private equity. And today she's going to walk through insane use cases for finance and asset managers. She's going to show you live builds, guys. This is very practical and very hands-on if you are in the finance space. So super excited f... Read More

Key Insights

• Due diligence is the process investment companies use to research a fund or company before investing and determine whether it is what it claims to be. Analysts compare provided information, including key performance indicators, with annual reports describing activity during previous years.
• Traditional due diligence is document intensive because analysts may collect several hundred pages in a data room. Reading those materials and identifying risks or red flags can take months, which creates a practical opportunity for AI-assisted research using an approved document collection.
• The quality of AI analysis depends heavily on explicit instructions. Annie structures instructions around who Claude is, what it needs to do, how it should complete the task, and which rules or constraints it must follow, including using only supplied project documents.
• Domain experts are essential to configuring a finance AI workflow because due diligence analysts understand the questions, evidence, and clarity required. Annie does not write instructions entirely alone, but works directly with analysts to translate their working methods into project guidance.
• A controlled document folder provides the research context for the due diligence assistant. The demonstrated project references locally downloaded public materials from a fund website, allowing the analyst to ask focused questions against a defined collection rather than an unspecified information source.
• Iterative evaluation is the core implementation process. Annie works with analysts for one or two weeks, collects their reactions to each answer, revises unclear or incomplete instructions, and repeats the cycle until the analysts are satisfied with the solution.
• Reusable skills preserve a refined research method beyond one investment target. After the questions, instructions, and feedback process produce satisfactory results, Claude can create a skill based on the conversation for application to another fund or company.
• Interactive due diligence research is not presented as an AI agent because the user continues to ask questions directly. Annie defines an agent as a workflow containing several tasks that acts on its own, a structure reserved for the later sustainability news example.

Questions & Answers

Q: How can AI support due diligence in asset management?

AI can act as a research partner by reviewing a defined folder of fund or company documents and answering the questions that due diligence analysts normally investigate. The analyst can use it to compare provided claims, key performance indicators, and general information with annual reports, while looking for risks, red flags, unclear statements, or discrepancies across the materials.

Q: What documents should an AI due diligence project use?

The demonstrated project uses documents downloaded from the investment target's website and saved in a local folder. These can include public information about the fund, key performance indicators, general fund information, and annual reports. The project instructions explicitly require Claude to use only the documents supplied to the project, keeping its analysis tied to the selected collection.

Q: How should instructions for a finance AI assistant be structured?

Instructions should identify who the AI is, what work it needs to perform, how it should perform that work, and which rules and constraints apply. The example adds a rule requiring the assistant to rely only on the documents provided in the project. Explicit instructions matter because they are supplied as context during each conversation.

Q: Why should domain experts help design an AI finance workflow?

Domain experts understand what analysts actually need from the system, including the main questions they ask, the evidence they consider important, and the level of clarity expected in an answer. Annie therefore works with analysts instead of drafting every instruction independently. Their feedback guides revisions until the responses match the practical requirements of due diligence work.

Q: How do analysts improve AI answers during implementation?

Analysts test the project with their normal questions and explain which parts of each answer are useful, unclear, or incomplete. The project instructions are then updated and the same process is repeated. Annie describes working with analysts for one or two weeks and continuing this feedback loop until they are satisfied with the quality of the answers.

Q: Why create a reusable Claude skill after testing?

A reusable skill preserves the method developed through repeated questions, feedback, and instruction changes. The original project may focus on one particular fund, but the resulting process can then be applied to another fund or a small SaaS company. This avoids rebuilding the complete research approach whenever analysts begin evaluating a different investment target.

Q: Is the demonstrated due diligence assistant an AI agent?

The due diligence assistant is not classified as an AI agent in the discussion because the analyst interacts with it by asking questions. Annie says an AI agent requires a workflow with several tasks and needs to act on its own. The later sustainability news summarizer is presented as the example that comes closer to that agent-based structure.

Q: What security precautions are recommended for financial AI work?

Annie recommends that enterprise clients in regulated financial environments avoid personal Claude accounts for sensitive work and consult their IT departments before sharing information. She recommends Pro or Enterprise accounts because a data processing agreement is included, providing more control over data, including obligations concerning breach reporting and requests to delete information.

Summary & Key Takeaways

• The first use case turns Claude into a due diligence research partner. Analysts connect a folder containing public fund documents, define the desired review process, and ask targeted questions about risks or inconsistencies. This approach reduces the burden of manually reading hundreds of pages, a process that can otherwise take months.

• Effective project instructions answer four questions: who Claude should act as, what it must accomplish, how it should perform the work, and which rules constrain it. Annie develops these instructions with analysts because domain experts know the relevant questions, required evidence, and standards by which the resulting answers should be evaluated.

• The solution improves through repeated collaboration with analysts over one or two weeks. Their feedback identifies unclear or incomplete responses, which leads to revised instructions and further testing. Once the team is satisfied, the established method can become a reusable skill for researching another fund, a small SaaS company, or another investment target.