AI agents in finance: what they can do today, and the controls they need
Agents go further than chatbots: they take steps on their own, such as chasing an invoice or preparing a reconciliation. Useful, as long as their limits are set as carefully as a new colleague’s.
A chatbot answers questions. An agent acts: it can read an inbox, match a remittance, raise a reminder, prepare a journal for approval or reconcile a supplier statement. Large finance teams are moving quickly. In Deloitte’s CFO Signals survey of North American chief financial officers this year, more than half named AI agents among their top transformation priorities.
Good first jobs for an agent
- CollectGather invoices and receipts from email and supplier portals.
- MatchPair remittances and bank lines with the invoices they settle.
- ChaseSend polite, escalating reminders for overdue invoices.
- PrepareDraft accruals, reconciliations and commentary for review.
The controls that matter
- An agent never pays money out, changes bank details or signs anything on its own.
- It works with its own login and only the access it needs.
- Each action is logged, with the reason, so a person can review it.
- Someone is named as responsible for each agent and checks its work weekly.
Where they struggle
Agents are least reliable where the right answer depends on context they cannot see: a customer relationship, a dispute, a one-off deal. Keep those with people, and use agents to clear the routine so people have time for them.
This insight is general information, not advice for your circumstances. Talk to us before acting on it.