Data Action Layer January 18, 2024

Speed up CIM review with LLMs

by Elizabeth Matson

Website CI Ms

Private equity firms deal with vast amounts of documents, many of which are unstructured and extremely long. For documents such as Confidential Information Memorandums (CIMs), the data may be presented differently in each iteration, and it can seem like the only way to find what you need is to search through it page by page.

That’s where large language models come in.

Increasing your team’s efficiency when reviewing CIMs is one of many ideal use cases for the power of LLMs and generative AI in financial services. But what does that more efficient workflow look like?

An average CIM can consist of 100+ pages or PowerPoint slides. Analysts tasked with extracting the data required to make an informed purchasing decision frequently rely on searching through documents for specific keywords. When they find the right data, they manually input it into their deal sourcing platform or CRM.

Here’s how we designed a CIM workflow at Alkymi:

CIM Workflow Diagram

Step 1: We used Alpha, our generative AI tool, to build a set of questions to be asked of each document sent through the workflow.

Step 2: The answer to each question needed to come from specific sections, so we added page filtering rules to ensure answers were pulled only from designated sections (for example, a question asking about the company’s EBITDA would pull an answer only from the “Financial Overview” section of the document). Using this method, the vast majority of data fields were populated automatically, with a 100% traceability rate, so analysts could easily verify the source of each answer within the document.

Step 3: If an answer wasn’t found in a particular section, our platform would flag it, so the analyst team knew that field needed further review.

Step 4: An analyst then searched for the value in the document, using the document viewer within our platform. If the answer was in the document, outside of its normally designated section (for example, if the EBITDA is instead listed in the “Company Snapshot” page), they were then able to quickly locate it.

Step 5: Next, they used our Smart Capture tool to populate the corresponding data field and link it back to where the value was found in the document, without the need to retype it manually.

Step 6: Once all the data was extracted and organized to the correct fields, our platform automatically validated and transformed it to the correct format. Each answer could be traced back to its source in the document.

Step 7: Lastly, the clean, validated data was sent directly to a deal sourcing platform via API, with no manual input required.

Using the power of LLMs to automate data extraction and relying on features like document search only for exceptions, we were able to build an automated workflow that reduced deal cycle analysis time from 90 days to 30 days.

Taking advantage of tools built with large language models can not only expedite the deal sourcing process, it can also give you the power to capture more data, without adding more team members to a project or taking more of their time. Reviewing a stack of CIMs can be a long and painful process, but it doesn’t have to be. 

Schedule a demo today to see how Alkymi can streamline your workflows.

More from the blog

February 22, 2024

Quarterly Statements Rule: 3 ways to ensure compliance

by Elizabeth Matson

The transition window for the SEC's new Private Fund Regulations is well underway. AI-powered workflows will help you stay compliant and improve efficiency.

December 20, 2023

2023: A year in review

by Harald Collet

2023 was a monumental year for Alkymi. Take a look at our top data points from the past year.

December 12, 2023

Using LLMs confidently in financial services workflows

by Elizabeth Matson

Tools powered by LLMs can help you dive deeper into your data. Learn how you can use them with more confidence that you're getting the right answer, every time.