Most of it, yes. An AI agent can check a deposited cheque's image, confirm the payee, date, amount and signature fields, catch duplicates, and submit clean cheques for clearing at depth 3. Releasing funds before the cheque clears beyond the bank's policy, third-party cheques, and anything that looks altered or counterfeit stay with a named person.
What does processing a cheque actually involve?
A cheque deposit, whether by app image, machine or branch, goes through the same questions. Is it made out to the account holder? Is it dated within the period the bank accepts, and not post-dated? Do the amount in words and the amount in figures agree? Is it signed? Has this cheque been deposited before, here or elsewhere? And how much of the money should the customer be able to use before it clears?
Character recognition and rule-based tools already answer some of these; our article on robotic process automation covers where that kind of tool stops. An agent adds reading the cheque as a whole, comparing it with the customer's history, and explaining problems to the customer in plain language. That is useful, as long as the agent does not take on decisions that belong to fraud and credit staff.
Where does the agent have to stop?
Two lines matter most.
Fraud. A cheque with a payee name that looks overwritten, an amount that does not fit the customer's history, or a drawer account that has sent several cheques to new accounts this week is a fraud question. The agent flags it and holds the deposit. It does not decide whether the cheque is genuine.
Early funds. Most banks make part of a deposit available before the cheque clears. That is effectively short-term credit: if the cheque bounces, the bank is out of pocket. The agent applies the bank's availability rules exactly. It does not extend them because the customer asks or because the drawer is a well-known company.
What may the agent do, and what stays with a person?
| Step | The agent | Stays with a person |
|---|---|---|
| Reading the image | Extracts payee, date, amounts, signature, depth 1 | Images it cannot read |
| Field checks | Confirms payee matches, date in range, amounts agree | Mismatches the customer disputes |
| Duplicate detection | Checks for earlier presentments of the same cheque | Confirmed duplicates |
| Submitting a clean cheque for clearing | Submits at depth 3 | Nothing routine |
| Funds availability | Applies the policy's availability rules | Any exception |
| Third-party or endorsed cheques | Holds | Accepting under policy |
| Suspected alteration or counterfeit | Holds and flags | Fraud team |
| Returned (unpaid) cheque | Debits the credit under the clearing rules and tells the customer | Deciding whether fraud is involved |
An example: two cheques in one evening
Take a customer depositing two cheques through the app. The first is a 400 refund from an insurer, made out correctly, dated last week, with matching amounts. The agent checks it, finds no earlier presentment, submits it for clearing and makes the standard portion available under policy. The customer sees the expected date for the rest.
The second is for 3,500, from a private buyer of a car the customer sold online. The image shows the amount in figures has a slightly different ink density from the words. The agent does not say the cheque is forged. It tells the customer the cheque needs a closer check before the funds are available, holds the deposit, and sends the image and its notes to the fraud team. It also suggests the customer not hand over the car until the funds are confirmed, using approved wording.
What controls make it safe?
- Availability rules from policy. The agent reads how much to release and when from the policy library, version recorded.
- No override capability. The agent cannot release funds beyond the policy amount.
- Duplicate checks across channels. App, machine and branch deposits are checked together.
- Fraud flags held, not judged. Suspected alteration goes to a person with the image and the agent's reasons.
- A documented process. Each check, its order and its evidence are written down; see how to document an AI agent's process.
A cheque cannot be pulled back like an electronic payment; returns follow the clearing rules instead. For electronic payments, see processing payment recall requests with an AI agent. The general limits are in AI agents in banking: what they may and may not do.
Questions people ask
Can the agent tell the customer when the money will be fully available?
Yes, from the bank's clearing timetable and availability policy. It should say that funds can still be reclaimed if the cheque is later returned unpaid.
Can the agent accept a cheque made out to a slightly different name?
Small variations, such as a missing middle name, can be accepted under most banks' policies. Anything more than that, or a cheque payable to someone else, goes to a person.
What if the customer deposits a cheque that was already paid in at another bank?
The agent can catch it only if the bank can see the earlier presentment. Where it cannot, the clearing process will return one of them, and the agent tells the customer what happened.