Sage Finance Intelligence Agent: Advanced AI for Financial Decision-Financial Decision-Making and Automation

Introduction to the Sage Finance Intelligence Agent Panel

During this session, titled "Inside the Finance Intelligence Agent: Turning AI Capabilities into Confident Decisions," Kevin Wallace, a Senior Product Marketing Manager at Sage with $5$ years of experience, introduced a panel focused on the emergence of AI in the financial ecosystem. He was joined by Shneuryshakar (referred to as Srihari), the Product Manager for AI, and Rohit Kumar, also a Product Manager for AI. The session aimed to elevate the "AI IQ" of participants by discussing industry shifts and the specific product roadmap for Sage Intacct. Kevin Wallace shared his dual perspective as both a financial professional with experience in FP and A and as a small business owner navigating three boutique fitness studios. He emphasized that the introduction of AI must provide confidence through explainability, verification, and human control. The goal is to move beyond simple automation to a system where humans set guardrails and approve AI actions, ensuring that all financial data remains transparent and auditable for CFOs and controllers who are ultimately responsible for the bottom line.

Strategic Impact of AI and Automation on Finance Workflows

Kevin Wallace outlined the primary goals of AI at Sage, shifting the focus from specific features to the overarching impact on daily workflows. The central transformation is the shift from "doing" to "reviewing," allowing professionals to spend less time on mundane, repetitive tasks and more time ensuring accuracy. Key objectives include increasing accuracy, reducing risk, and accelerating visibility. In the context of decision-making, such as hiring, firing, or ordering supplies, managers require the most up-to-date information accessible in real-time. The panel discussed how AI helps increase understanding of financial data through a more unified experience. Wallace differentiates the ecosystem into work done inside Sage Intacct—utilizing the Finance Intelligence Agent (FIA), close automation, AP automation, Sage Intelligent Time, and outlier detection—and work done outside the platform. For external work, Sage provides REST APIs, the Sage Data Cloud for zero-copy access in Snowflake, and the recently announced Sage Contact IP, which allows secure connections to AI-based applications with existing Sage permissions.

Panel Discussion: Reimagining Finance in an AI-First World

During the panel discussion, Rohit Kumar addressed what it means to reimagine finance for an AI-centric world. He drew a parallel to the shift from Google’s "$10$ blue links," which left the burden of discovery on the user, to generative AI like ChatGPT, which provides direct answers. In finance, this means moving away from static dropdown menus and buttons toward natural, proactive language interfaces. Srihari forecasted that over the next $2\,\text{to}\,3$ years, finance leaders should expect a significant reduction in the "time to insight" and "time to action." He categorized these shifts into three areas: less hunting for data across distinct modules, faster first-pass analysis (such as scenario modeling or classifications), and a unified workflow where users can execute actions directly from the point of insight. Kevin Wallace noted that while professionals often have $10$, $20$, or $30$ items on their to-do lists, AI allows them to reach the items historically left on the backburner with greater efficiency.

Finance-Specific AI Requirements and the Philosophy of Trust

Rohit Kumar highlighted the critical differences between general-purpose AI and AI for finance. While platforms like ChatGPT may hallucinate or provide approximate answers, finance requires absolute accuracy. As Aaron Harris mentioned in the keynote, being off by even $1p$ or $1\,\text{cent}$ prevents books from closing. Consequently, Sage AI focuses on transparency, explainability, and controllability. Kumar described himself and fellow product managers as "gatekeepers" or "villains" who must reject flashy, hype-driven features if they do not meet rigorous standards for scale, accuracy, and trust. Srihari added that in finance, "almost right is wrong." While spelling errors in an AI-generated email are trivial, errors in financial numbers carry immense risk. The panelists emphasized that AI should level the playing field, but the human element remains the differentiator. Kevin Wallace referenced Kelly Wright, a professor and former EVP of Sales at Tableau, who noted that as data and research become commoditized through AI, it is the people who design workflows and build solutions who create value.

Functional Capabilities of the Finance Intelligence Agent (FIA)

Srihari detailed the foundations of the Finance Intelligence Agent (FIA), which acts as a conversational partner built across six core modules: AP, AR, GL, OE, EO, and Cash Management. The agent focuses on four primary capabilities: retrieval, analysis, proactive insights, and action. Retrieval involves pulling records and navigating transactions without manual searching. Analysis includes identifying trends and anomalies. Proactive insights involve FIA flagging risks—such as expiring discounts or overdue invoices—without being prompted. Finally, the agent can execute actions like paying bills, approving transactions, or sending reminders. The FIA intends to compress the traditional loop of logging in, exporting data to spreadsheets, creating pivot tables, and then returning to the system to act. Instead, it offers a single, unified experience with deep links back to specific records and a centralized view of all pending approvals across the system.

Technical Architecture and the Agentic Workflow

Rohit Kumar provided a "$20,000\,\text{to}\,10,000\,\text{foot}" view of the FIA’s backend architecture. When a user asks a question in Copilot, the query first passes through a Sage Arbiter or content moderation system. This layer screens for harmful content, jailbreaking attempts, and ensures the query is accounting-related. Once cleared, an orchestration layer extracts the user's intent and creates a plan, selecting the appropriate agents or "skills" (e.g., data query skill or analysis tool). The Sage Agent then executes this plan, utilizing the Model Context Protocol (MCP) or APIs to pull data. The agent can self-correct in an agentic loop if initial results are incomplete. Before returning the final answer, the Arbiter checks the response against Sage trust and communication standards. The system is designed as a "glass box," providing users with a "How Copilot worked this out" tab that reveals the specific objects, parameters, and logic used to generate the answer. Every action taken is auditable, recording exactly who approved a specific payment and when ($\text{minute/hour}$).</p><h3>Case Studies and Future Roadmap</h3><p>Srihari demonstrated a future-state scenario where a user logs in at$8:02\,\text{AM}$to find a proactive insight that two discounts are expiring tomorrow. By$8:03\,\text{AM}$, the user follows the alert to a detailed panel. At$8:04\,\text{AM}$, the user asks for a recommendation on which bills to pay given a budget of$10,000\,\text{USD}$. By$8:05\,\text{AM}$$, the action is executed and shared with the team. Future developments coming in the next few months include broader finance coverage, stronger reasoning capabilities, and the introduction of a Copilot inbox and workspace. This inbox will proactively notify users of tasks like overdue invoices or pending approvals. Further developments include trend analysis, personalized contextual conversations, and the ability to draft succinct email templates for payment reminders. Rohit Kumar concluded by explaining that the content moderation service is being fine-tuned specifically for the finance domain; for example, while an off-the-shelf LLM might block a request to see a balance sheet as sensitive, the Sage Arbiter understands that a authorized finance professional should have access to that record.