Finance teams are typically plagued with a great deal of “busy work,” according to the leaders of Ramp, who are betting that new forms of artificial intelligence will ease the burden that is placed on humans.
“Do you like approving expenses? Are you actually reading policies over and over again?” Subham Agarwal, Ramp’s product manager and marketing leader, told American Banker.
Ramp, a
The market for
What Ramp and Intuit are up to
Ramp is using technology from OpenAI, the firm that powers ChatGPT, among other AI-related software, to build Ramp Agents. These agents will use context-aware reasoning to manage workflows, working independently to enforce spending policies, prevent violations and improve the underlying technology over time.
Based on user settings, the AI agents will be designed to approve low-risk expenses or provide a recommendation with rationale to the approver, issue alerts for suspicious receipts and invoices, answer employee questions about spend policy, suggest edits to policies based on usage and feedback; and spot trends that signal fraud or careless spending, according to the company.
In testing, the AI agents caught 15 times more policy violations than humans, Agarwal said, arguing that finance departments at businesses are getting more work as corporate payments get more automated and complex.
“The work isn’t in their job description. You have managers and controllers that are looking at transactions and are making judgements on individual expense reports,” she said.
Ramp’s initial agent AI is designed to assist with expense management. Future AI agents will be programmed to help with reconciliation, accounting and other functions.
Another business-focused technology firm, Intuit, has also developed AI agents to address different jobs inside businesses.
One is a “payments agent” that “automatically” performs invoice tracking and transaction categorization and sends invoices and alerts. Other agents manage accounting tasks such as reconciliation and bookkeeping; while finance agents cover reporting, KPI analysis and scenario planning. These agents reside on a dashboard that includes access to human agents and other business payment and accounting functions.
Intuit is positioning its AI agents as a technology layer that simplifies working with all the existing software programs small businesses use. “We’ve reached a point of overdigitization. It feels like there’s software for everything,” Dave Talach, a senior vice president at Quickbooks, told American Banker. “Business owners are swimming in a plethora of software tools but they are all bespoke and don’t work together.”
Agentic AI for small business
Software that supports businesses is a particular growth area for agentic AI. Mordor reports that in 2024, fewer than 1% of organizational software applications harnessed the potential of agentic AI. Most AI agents still lean heavily on user prompts. However, by 2028, Mordor expects this landscape will shift dramatically, with projections suggesting that nearly one-third of all enterprise software will embrace agentic AI.
“While agentic AI is in the early stages of development, banks and payment fintechs do face pressure to have a strategy for how their products will fit into an emerging world of agentic commerce,” Christopher Miller, lead analyst for emerging payment technologies for Javelin Strategy and Research, told American Banker.
Read more about artificial intelligence. (https://www.americanbanker.com/artificial-intelligence).
Miller tells clients (of whatever size) that it is critical for companies in all parts of the payments industry to understand what agents are, what agents can actually do today, what agents might be able to do tomorrow and how those changes will impact consumer behavior. The development and deployment of AI in financial services is uneven and inconsistent, according to Miller, who says there are at least least three different types of exploration going on in financial services.
One is “agentifying” back-office functions, such as business process automations, internal facing agents, functions such as fraud detection or data analysis, and supporting customer service reps. A second area of exploration is using agentic AI to upgrade consumer-facing tools that complete or enable commerce.
“For example, a general-purpose agent can manage finances, make payment selections and perform shopping tasks,” Miller said. A third use is the addition of AI capabilities to existing products, generally under the guise of an “agent” or “assistant” tool that makes using an existing product easier.
Intuit’s announcement fits that category, Miller said. “Software companies that sell financial products to consumers and businesses will certainly face marketing pressure to include agentic features, while customers, particularly business customers, will continue to ask how automation technologies in all of their forms are saving time and improving results, and will require evidence to pay premiums for agent capabilities,” Miller said.
Intuit’s Talach said there are different levels of comfort with AI, so agentic AI will not be for everyone. “People will move at their own pace,” he said.
