Charta is on Google Play: practise on the go. Get the app
Charta

Explain how investment managers put large language models and generative AI to use

CFA® Level I · Quantitative Methods · Introduction to Financial Data Science · QM.20.c

The idea

Generative AI and large language models are used across investment work mainly to handle unstructured text at scale. Analysts use them to summarise filings, research and earnings call transcripts, to extract figures and themes, and to score the sentiment of news and disclosures as inputs to models. They help draft reports and client communications and speed up coding for data work. Retrieval-augmented generation connects a model to a chosen document source, such as a firm's research library, by inserting relevant passages into the prompt, which keeps answers current without retraining and, where the system keeps track of its sources, lets answers cite them. In every use the output is a draft that a professional checks before it informs a decision or reaches a client.

A practice question

From the question bank, with the reason each wrong option is wrong.

Which is a typical use of a large language model by an equity analyst?

  1. Certifying reported financialsWhy not: Assurance on financial statements is an auditor's role and needs verification an LLM cannot give.
  2. Drafting summaries of earnings calls — correct
  3. Replacing compliance reviewWhy not: LLM output still needs human and compliance review; it does not remove that obligation.

Summarising long unstructured text such as call transcripts, filings and research is a common LLM use, with the analyst checking the output before relying on it.

A harder one

An analyst wants to score the tone of 5,000 company filings each quarter as positive, neutral or negative. A suitable use of an LLM is to:

  1. classify each filing's sentiment at scale — correct
  2. simulate future prices of each issuerWhy not: Price simulation is a Monte Carlo task, not the text classification the analyst needs.
  3. audit the figures reported in filingsWhy not: Verifying reported numbers is assurance work, not a question of tone.

LLMs can read large volumes of text and label sentiment consistently and quickly, producing a signal the analyst can validate against a hand-checked sample.

The hardest tier

A firm connects an LLM to its internal research library through retrieval-augmented generation. The main benefit is that answers:

  1. can no longer contain any hallucinationsWhy not: Retrieval reduces unsupported answers but cannot eliminate them; the model can still misstate sources.
  2. come from weights retrained on the libraryWhy not: RAG does not retrain the model; the documents are supplied as context at query time.
  3. draw on current firm sources that can be cited — correct

Grounding responses in retrieved documents makes them more current and verifiable and lets users check the cited sources, while the base model stays unchanged.

In the app: 6 practice questions on this outcome, a narrated walkthrough with diagrams, spaced review that brings it back before you forget it.

Take the free diagnostic →

More in Introduction to Financial Data Science