Generative AI has tremendous potential to help save time and money.
One great example of a company using the technology to great advantage is Moody's, which recently expanded its capabilities with an agentic AI offering for financial information.
Moody's Corporation operates two main businesses: a credit rating agency and Moody's Analytics, which provides data analytics, research and workflow solutions to financial institutions. The company has spent decades building quantitative models for catastrophe risk and credit risk assessment. They had experimented with machine learning for years, but the technology struggled with unstructured data like PDF documents.
Everything changed in late 2022 when ChatGPT demonstrated breakthrough capabilities with unstructured data. Moody's quickly identified two opportunities: improving internal productivity and maintaining commercial relevance as customers adopted AI. The company first launched Research Assistant in December 2023, a conversational interface grounded entirely in Moody's proprietary content. It became their fastest-adopted product ever.
This week, Moody's officially launched Agentic Solutions, a suite of AI-powered tools designed to automate knowledge-intensive workflows across finance, risk and strategy. The platform leverages Moody's data covering more than 590 million global entities to coordinate specialized AI agents that can work together on complex tasks.
Among the complex tasks that Moody's customers perform is the preparation of credit memo that traditionally could take 40 hours or more. Agentic AI now dramatically reduces the time needed.
"We can produce a credit memo in a set of minutes, in two minutes you have a credit memo," Christina Pieretti, General Manager, Head of Moody's Digital Content and Innovation, told VentureBeat.
While the time reduction is noteworthy, the real story isn't just Moody's impressive efficiency gains. It's how the company's methodical approach to agentic AI offers a blueprint that other enterprises can follow. This proves especially valuable for organizations in regulated industries where accuracy and compliance are non-negotiable.
From conversational search to workflow automation
Moody's Agentic Solutions represent a significant evolution beyond their successful Research Assistant product. While Research Assistant functions as what Pieretti describes as conversational AI for Moody's content, the agentic approach tackles fundamentally different challenges.
The distinction comes down to scope and output. Research Assistant helps users find and consume information more efficiently through conversational queries. Agentic Solutions automate entire workflows and produce complete deliverables.
"There's a huge efficiency component, in terms of an agent that's looking for the different pieces of information, making sure that you have a comprehensive search of that information, and that you're putting those pieces together in the way that a human would normally look," Pieretti said.
This evolution from information retrieval to task automation represents a broader shift in enterprise AI applications. Rather than simply making existing processes faster, agentic AI fundamentally changes how work gets done by handling multi-step workflows that previously required significant human coordination and analysis.
The modular approach to agentic AI architecture
While many enterprises struggle with where to start with AI agents, Moody's has developed a modular approach. This method breaks complex workflows into manageable, specialized tasks.
"When we think about an agent, we are basically assembling a combination of... different kind of tasks, or what we call agents that are each doing their work, and then you're concatenating the results at the end to produce the output," Pieretti explained.
For credit memo generation, this means separate agents handling distinct functions. One identifies and verifies the correct company entity. Another extracts financial data from 10-K filings rather than third-party sources. Others handle peer analysis and risk assessment. This modular approach contrasts sharply with monolithic AI systems that attempt to handle entire workflows in one go.
The architecture mirrors successful patterns in modern software development, where microservices have replaced monolithic applications. But unlike traditional software, these AI agents can work in parallel. This dramatically reduces processing time while maintaining the specialized focus that ensures accuracy.
Model-agnostic strategy reduces vendor lock-in
Unlike some enterprise AI initiatives that bet heavily on a single large language model(LLM), Moody's has embraced a model-agnostic approach. The company evaluates different LLMs based on cost, context size, performance and instruction-following capabilities.
"We don't see our competitive advantage in developing large language models," Pieretti said. "We like to say we're model agnostic. And not only we're model agnostic, but we might end using different models for different things."
This strategy offers enterprises several advantages.
First, it reduces vendor lock-in and allows organizations to optimize costs by using the most efficient model for each specific task.
Second, it provides flexibility as new models emerge and capabilities improve.
Third, it forces organizations to focus on their actual competitive advantages rather than trying to build foundational AI capabilities from scratch.
For Moody's, that competitive advantage lies in their proprietary data covering more than 590 million global entities and their deep understanding of financial workflows. It does not lie in training language models.
Grounding prevents hallucinations in high-stakes environments
The new agentic AI solutions from Moody's benefit from the company's experiences building out its first gen AI product.
Moody's Research Assistant became their fastest-adopted product ever by focusing obsessively on accuracy over flashy capabilities. The approach offers lessons for other enterprises operating in high-stakes environments.
"We were very focused on not hallucinating," Pieretti said. "The way we approach that is everything was grounded on information from Moody's."
This grounding approach proves particularly crucial in regulated industries where inaccurate information can have compliance implications. Rather than allowing AI models to draw on their full training data, Moody's constrains responses to verified, proprietary information sources.The technique offers a template for other enterprises. Sacrifice some AI flexibility in exchange for reliability and auditability.
Another perhaps counter-intuitive finding was about research consumption. Users of Moody's Research Assistant consume 40% more research and data than they previously did, and they actually spend more time on the platform.
"You might think, well, that's counterintuitive, right? Because you're supposed to be saving time, but you're spending more time," Pieretti explained.
She noted that the reason why users spend more time is because the Research Assistant isn't just a simple search. It's a conversation about a topic and it will surface additional content that is relevant and as such the user ends up consuming more research.
This discovery reveals that conversational AI interfaces don't just make research faster - they make it more thorough by surfacing related information users might not have otherwise discovered.
Start small, focus on unique advantages
Perhaps the most important lesson from Moody's approach is their emphasis on starting small. They focus on playing to existing strengths rather than trying to revolutionize everything at once.
"If you are a vendor, you have to understand where you want to play in this agentic future. And where do you have a right to win?" Pieretti said. "We are spending a lot of time on where we can develop highly specialized agents…where we have unique features and they do not have to compete, for example, with something that Google or Amazon or Microsoft could develop with their eyes closed."
This philosophy extends to their customer recommendations as well. Pieretti noted that enterprises often struggle because they're trying to do everything at the same time and trying to build everything themselves."
The alternative approach involves identifying workflows where your organization has unique data, domain expertise or regulatory knowledge. These create natural advantages for AI implementation.
What this means for enterprise AI strategy
For enterprises looking to lead in AI adoption, Moody's success offers several actionable insights. Start with workflows where you have proprietary data advantages. Build modular agent architectures that can evolve with improving AI capabilities. Focus on accuracy and compliance over flashy features, especially in regulated industries. Resist the temptation to build foundational AI capabilities in-house unless that's truly your competitive advantage.
"This market is changing by the minute, right? So I think maybe the number one advice to everyone is, you know you could be the leader, one moment at the other moment, you're displaced if you don't have your eyes and your ears open," Pieretti said. "So I think you have to be very very observant of where the market is moving."