Finance work suits AI for three reasons. The volume is high, the rules are written down, and every step is already audited. It is also where a wrong action costs real money, which is why the design matters more than the model.
The three workflows below share one rule: AI does the reading, matching, ranking and drafting. A named person makes every decision that moves money.
1. Supplier invoices: capture, match, approve
The problem. Accounts payable teams key invoice fields by hand and chase mismatches by email. The work is slow, and the audit trail lives in inboxes.
What changes. AI reads each invoice, matches it to the purchase order and goods receipt, and drafts a short note on anything that does not match. Clean matches within an agreed tolerance move to the approval queue. Exceptions go to the purchase order owner with the evidence attached.
Who signs. The AP lead approves every exception and every payment run.
What to measure. Cost and time per invoice, days from receipt to approval, the share of invoices with no manual touch, and late-payment fees.
2. Overdue invoices: who to call first
The problem. Collectors work the overdue list from top to bottom. High-value or high-risk accounts wait behind small ones.
What changes. Each morning, AI ranks overdue invoices by amount, age, terms, payment history and relationship, shows the reason for each rank, and drafts the first message.
Who signs. The collector sends, edits or skips every message. Payment plans and write-offs need the credit controller.
What to measure. Days sales outstanding, overdue value contacted in the first week, and promises to pay kept.
3. Blocked orders: explain and release
The problem. Orders sit blocked on credit, pricing or stock checks while someone works out why. Customers wait, and sales chase finance.
What changes. AI reads the order, the credit position, the pricing conditions and how similar blocks were resolved before, then explains the block in plain words and recommends the next step.
Who signs. The credit controller releases or holds every order. AI can only trigger follow-ups agreed in advance, such as asking sales for a missing purchase order.
What to measure. Time to release, orders blocked for more than 48 hours, and value on hold.
What AI should never do in finance
Write this list before you build, and agree it with the finance controller and compliance:
- Change supplier bank details or create a new supplier
- Release a payment or approve outside the agreed tolerance
- Agree payment terms, threaten legal action or write off debt
- Release a blocked order, change a credit limit or change a price
These limits are what make the rest safe to automate. They also make the audit conversation short.
Why the baseline comes first
In Camunda's September 2026 survey, 72% of process decision-makers said process-related challenges had caused AI initiatives to fail. A common thread is a pilot with no agreed starting point, so nobody can say whether it helped. In finance the baseline is usually already in the system: timestamps, edit history and the general ledger. Take it in week one, before anything changes.
Get the use cases
The short guide below covers these three workflows on four pages. Subscribers to our newsletter get the full playbook: the workflow before and after for each use case, the control points, a baseline sheet and the delivery path.