Many financial services companies are experimenting with AI through pilot programs, but several challenges remain for adoption. Key concerns include data security, the accuracy of large language models (LLMs) and the rigorous scrutiny from regulators regarding AI’s role in financial decision-making. Current use cases are largely internal, with some customer-facing chatbot solutions addressing noncritical service inquiries.
Recently, there has been growing interest across industries, including financial services, in generative-AI-powered agents, often referred to as “agentic AI.” Agentic AI relies on LLMs to perform tasks autonomously, using the models’ capabilities for natural language understanding, generation and interaction. Foundational LLMs such as Llama, Anthropic Claude, Mistral, GPT and Gemini are trained on vast amounts of text data (to the extent that they are running out of text data for further training) and can perform a variety of language-related tasks. These models, when integrated into agentic AI systems on a platform such as Snowflake, provide the foundation for understanding context, generating responses, driving for more automation and better efficiency and making more balanced real-time decisions based on the input they receive.
Agentic systems
In agentic systems, a coordinating agent is a central component that manages and directs the activities of other agents (or subsystems) to achieve overarching goals. This coordinating agent typically acts as a conductor, orchestrating actions, allocating resources and making high-level decisions. To carry out its tasks effectively, the coordinating agent can use tools, models or other agents that are either domain-oriented or task-oriented in complementary ways, ensuring that the system can both understand the context and execute specific actions to accomplish the desired outcomes.
Domain-oriented
Domain-oriented services specialize in understanding and processing language related to a specific industry, field or knowledge area. As such, they can generate responses or decisions that are more accurate, relevant and context aware within that field.
In agentic AI, domain-oriented models can help the system understand and act on specific tasks more effectively. In insurance, for example, such models could generate more accurate policy language, assess risk factors or interpret claims data, aligning with industry standards.
Task-oriented
Task-oriented services, on the other hand, are focused on achieving specific objectives. These models are trained to understand the sequence of steps needed to complete a specific task, and they integrate with broader systems to perform actions autonomously. Task-oriented models are designed not just to understand language but also to interact with tools, retrieve data and take actions in pursuit of a well-defined goal.
Cortex Analyst: High-accuracy SQL generation
Snowflake Cortex Analyst can be seen as a task-oriented agentic service because it is optimized for a specific goal, which is accurate SQL generation to retrieve data from a Snowflake table. To do that, Cortex Analyst performs these tasks:
-
Interpreting user intent: The model processes natural language input or contextual information (for example, what kind of data the user is looking for) and checks that the data exists to answer the question or go back to the user and ask for clarity.
-
Generating actionable SQL queries: Based on that interpretation, it creates SQL code that can be executed to retrieve or manipulate relevant data in a database. Its accuracy in the SQL generation comes from understanding the context of what is stored in each table.
-
Task completion: The end goal of this process is to generate a highly-accurate SQL that fulfills a specific query or data extraction need, which is a clear, defined task.
Agentic systems in insurance
For businesses, especially in complex industries such as insurance, to get actionable insights, they need to go beyond accessing databases to also tap into contracts, documents and other unstructured data that can be searched using another type of task-oriented service — a vector store.
By representing documents as numerical vectors, AI agents can quickly search and automatically read claims documents, extract relevant details (such as incident type, damage assessment, policy coverage, etc.) and generate SQL queries to retrieve additional data from a claims database or policy system.
An LLM could then be used to make recommendations, update the claim status or trigger further actions, all driven by the agent’s coordination between documents (unstructured data) understanding and database (structured) interaction.
Accenture and Snowflake’s role in facilitating agentic AI
Accenture and Snowflake are at the forefront of enabling agentic AI for enterprises, addressing both technological and strategic needs. Our approach focuses on:
-
Providing a unified data and AI platform with cutting-edge tools and capabilities that allows developers, data scientists and data teams to rapidly build and deploy AI models and applications with enterprise-grade security and governance, with the ease of Snowflake’s managed infrastructure
-
End-to-end unified governance, from ingestion to application, enables teams to deliver a new wave of data agents that use high accuracy out-of-the-box retrieval for structured and unstructured data
-
Guiding clients in selecting high-value use cases, leveraging Accenture’s expertise advising C-suite executives and driving value for clients through Snowflake’s unified AI data platform that simplifies the process
An all-in-one AI Data Cloud platform
Snowflake and Accenture offer a comprehensive solution for enterprise AI, designed to power agentic AI use cases effectively:
-
All your data, all in one place: Snowflake provides a scalable, unified platform to manage all your data needs — whether it be Snowflake tables or PDF files in object storage — in one place, reducing silos, enhancing data quality and promoting seamless integration.
-
Robust governance: Our governed platform enables data security and compliance from data ingestion to AI application, so that none of your data is used to train external models (except as you direct for fine-tuning). This makes Snowflake an exceptional choice for piloting and implementing new agentic AI use cases.
-
AI tools built for enterprise
-
Snowflake Cortex Search and Snowflake Cortex Analyst: Interact with enterprise data through conversational interfaces and provide secure access to foundational LLMs.
-
Document AI: Automates the processing of various document types, from standard forms to handwritten notes.
-
Snowflake ML: Allows data scientists to build, train and deploy machine learning models or customize embedding models directly within Snowflake using familiar programming languages such as Python, while leveraging Snowflake’s powerful data processing capabilities for scalable, efficient workflows.
-
-
Industry expertise: Accenture understands clients’ needs and how to solve problems using agentic AI.
By combining these capabilities, Snowflake and Accenture empower enterprises to customize their AI solutions while maintaining flexibility and control.
A strategic framework for AI adoption
Implementing agentic AI requires a structured approach. Here’s what we recommend for enterprises to navigate their adoption journey:
-
Start with use cases that unlock high business value: Focus on use cases where approximate accuracy is acceptable. Prioritize areas where faster decision-making or process acceleration provides more value than painstaking precision. Strike the balance between risk management and fast-to-market value creation.
-
Target low-regulation areas first: In highly regulated industries such as financial services, deploying AI in compliance-heavy areas can be slower due to strict standards of care and approval processes. Begin with less regulated domains to achieve faster results while gradually addressing regulated use cases as AI reliability improves.
-
Keep a human in the loop (for now): To mitigate risks related to explainability, customer service and accuracy, maintain human oversight in AI-driven processes. As the technology matures and reliability increases, you can gradually reduce human intervention. However, for regulated or high-stakes use cases, human review should remain a key component until the AI’s trustworthiness is fully established.
Use case: Gen AI Claims Agent by Accenture
Accenture has leveraged Snowflake’s platform to create a groundbreaking AI insurance claims agent, combining Snowflake’s tools to automate key parts of the claims process with Accenture’s expertise in agentic AI. This AI agent can review documents, summarize information, make claims decisions and generate personalized claims letters to clients, explaining the reasoning behind an approval or denial.
According to Mike Lao, a leader of Accenture’s data and AI team, “Underlying the Gen AI Claims Agent is Snowflake’s AI Data Cloud, which includes capabilities like Document AI that can process various documents, such as driver’s licenses, receipts and forms. We also use Snowflake’s Cortex Analyst, which helps the AI Claims Agent analyze data and policy documents to make informed decisions.”
Accenture’s focus on claims stems from insights into the financial services industry, where claims approval does not always require 100% accuracy. Interestingly, clients are often willing to accept partial payouts in exchange for faster processing. As Accenture found, there’s room for flexibility in claims payments — decisions can be appealed internally, avoiding regulatory intervention. This contrasts with the more stringent accuracy requirements of areas such as financial advice, which demands fiduciary responsibility. Claims, in comparison, are less regulated, offering insurance companies more leeway to adjust payouts within their margins. The ability to speed up payments can significantly enhance both the insurer’s bottom line and customer satisfaction.
To ensure reliability and client satisfaction, Accenture has included human ingenuity in the AI Claims Agent. Mike Lao emphasized, “Keeping a human in the loop is important. We are using AI to reinvent work processes to scale AI adoption in organizations and meet client expectations.”
The Accenture team in Manila, Philippines, brings deep expertise in claims payments supporting global insurance companies. This project marks the first step toward broader adoption of AI agents across various financial services sectors. As Kaushik GD, Snowflake’s Head of Financial Services in APJ, noted, “In the future, we expect to see AI agents assisting with financial planning and personal investment advice. However, this will take time, as both technical solutions and regulatory acceptance evolve.”
