Demand for AI agents in finance functions is concentrated in three areas:

• workflow orchestration for month-end close and reconciliations;
• exception handling, for finding anomalies and identifying risks; and
• pulling out information such as policy, audit trails and prior decisions.

That’s according to Ciaran Cosgrave, CEO of Nearform, an Ireland-based agentic AI consultancy targeting the financial services, wealth and insurance sectors.

AI agents work best when focusing on depth, as opposed to breadth, he says. “The most successful firms go deep on one high-value process and prove impact end-to-end, rather than running dozens of shallow pilots.”

Know the risks of AI agents

As we’ve previously covered, AI agents can ‘go rogue’ and cause serious damage to systems and databases if they are not set up or managed properly. Setting up appropriate safeguards will help with this.

Using narrow-focus, ‘vertical-specific’ tools can help to reduce the risk of hallucinations, with specific regulatory and compliance requirements built directly into their workflows, according to Joris Van Der Gucht, founder and CEO of Ravical. “If you invest in vertical-specific solutions, hallucination becomes more under control.”

Example use cases

AI agents can be used to manage (with assigned humans-in-the-loop) month-end reporting, working perpetually behind the scenes to reduce the pressure on finance teams in the latter parts of the month. Here are some simple ways in which agents can be deployed.

Supplier invoice verification

According to Cosgrave at Nearform, invoice reconciliation for high-volume clients is one area where AI agents particularly drive innovation. Using contextual reasoning, the agent can flag discrepancies between supplier invoices, purchase orders and receipts early in the cycle — such as non-matching figures or name variations — that require human attention.

“An agent can be set up to match thousands of invoices to POs and receipts. It isolates only the breaks and drafts the supplier query. The accountant clears exceptions instead of trawling line by line, turning days into minutes,” he says.

Categorising transactions

AI agents can continuously monitor cross-border accounts and transactions in real time, highlighting balance mismatches and putting forward potential remedies ahead of month-end.

Ensuring the month-end close is ‘closed’

AI agents running constant anomaly-detection algorithms throughout the month could pick up errors much earlier than during the final review. For example, if a regular monthly subscription invoice due by the 28th of the month is yet to arrive, the agent then alerts the team.

Detecting long-term issues

“Businesses lose up to 30% of revenue annually to inefficiencies, which is exactly the type of structural waste these systems are designed to address,” notes Ciaran Cosgrave at Nearform.

“AI agents can continuously scan for cost leakage, automate routine decision loops and simulate financial scenarios using live data.”

For example, an AI agent could model the margin impact of a pricing change across every product line before the CFO commits to it. Agents can also potentially connect operational and financial data and present it in ways that could surface otherwise unnoticed opportunities, says Cosgrave. “AI applied in the right way, can help the CFO who explains the last quarter become the one shaping the next one.”