With Moody’s new generative AI tool, Moody's Research Assistant, you can save up to 27% of your time conducting research and analysis so you can focus on staying ahead of challenges and identify new opportunities faster than ever. 

Within seconds, you can synthesize vast amounts of Moody’s research, analyze data and generate custom reports. 

How does Research Assistant work?

In the most basic form, this powerful tool is an AI model that takes your prompt as a directive to go retrieve specific data and research from across Moodys.com to deliver a relevant output in seconds. 

As the saying goes, “garbage in, garbage out.” The quality of the input matters here.  

It all starts with a prompt. 

So, let’s start with the basics: What is a prompt?

According to Miguel Romao, Senior Director of Product Strategy for Research Assistant, “Prompting is being able to engineer a specific question to get a specific output from a Large Language Models (LLM).”  

He continues, “An LLM is a model that consists of a neural network that is trained to predict the likelihood of a word or sequence of words given the context of the preceding words.”  

How are LLMs trained to predict the next word?

LLM models are trained with a RAG, or an AI framework that combines the power of retrieval-based and generation-based models.   

Let’s take a closer look at the types of information you’ll gain access to in seconds with the power of Research Assistant.   

Moody’s Research Assistant leverages RAG using Moody’s proprietary data: 

• All reports across all segments published since 2020   
  
• 10 years of rating history and 3 years of key indicators for the Corporate, Banking, Insurance and Sovereign market segments   
• 3 years of adjusted and as-reported financials for all Moody’s rated corporate entities   
• 3 years of financials, peers and firmographic information on over 12,000 corporates that do not have an active public rating from Moody’s Ratings

Now that we know what a prompt is and the type of data that Moody’s Research Assistant has been trained on, let’s go back to prompting.  

The importance of prompt engineering to optimize the output that you desire

It’s important to understand that Research Assistant is not Google or a search engine that can return results off one word. Think of it more as a smart new hire with immediate access to research and data across Moodys.com and the ability to deliver an answer in seconds but needs specific direction on how you like things done.

To put this into a more tangible example, imagine you ask someone to make a peanut butter and jelly sandwich without any additional context.  They might bring it to you on pumpernickel bread when you were hoping for white, deliver it weeks out when you were looking forward to eating it on your lunch break today, or even choose to use strawberry jelly instead of grape.  The sandwich isn't necessarily incorrect, but the end result was in no way what you were looking for.

How does this apply to prompting in Research Assistant? 

As we mentioned before, prompting is about being specific in your request in order to generate an expected outcome. Just like in our example above, you need to bring in specifics around the data you are expected to receive, the timeframe and the format of delivery. 

So instead of just typing “Ford” like you would in a search bar, try adding some specifics like, “write a credit memo on Ford going back three years”.  Instead of “Ford’s ratings”, try “Make a table of Ford’s rating changes over the last 10 years”.

Try out your new prompting knowledge now!

Don’t have access to Research Assistant? Click on the link below.