Mortgage completion is no place for autonomy

Completion is the moment a mortgage stops being paperwork and becomes money. The lender releases a large sum, often to a conveyancer's client account, on a fixed day, and the property changes hands. Once that happens, there is no practical way back. That makes completion the one step in a mortgage that should never be autonomous, however good the agent is at everything before it.

What makes completion different from the rest of the mortgage?

A mortgage involves weeks of work: application, document gathering, affordability checks, valuation, underwriting, offer, conveyancing. Almost all of it is preparation. Errors found during that time can be corrected, sometimes at the cost of delay.

Completion compresses everything into a single, irreversible event. Funds go out. The purchase completes. Other transactions in a chain, each depending on this one, complete too. If the money went to the wrong account, or the conditions were not met, the lender cannot pull it back without unwinding a property sale, which in practice rarely happens.

There is also time pressure. Completion dates are agreed in advance and missing one can collapse a chain of purchases. That pressure is exactly the condition under which shortcuts are taken and fraud succeeds.

Why is completion a fraud target?

Completion funds are large, they move on a predictable day, and they go to an account that is often specified in correspondence. That combination attracts fraud. A common pattern is intercepted email: a fraudster gains access to a conveyancer's or buyer's email and sends revised bank details shortly before completion. The new details look legitimate, arrive at the expected time, and redirect the funds.

An agent processing completion requests would read that email the same way a busy person might. It is plausible, timely and consistent with the transaction. The only defence is a rule that changes to payee details on a completion are never accepted from an external source, and that the destination account is confirmed against a record held independently.

Where can an agent help before completion?

The weeks before completion are full of work an agent can carry. Gathering and checking documents is the largest share; the playbook on how to gather documents for mortgage applications with an AI agent covers it.

Approval stays with a person, as explained in whether an AI agent can approve a mortgage. Release does too, for the reasons above. Everything else can be carried, in part or whole, by the agent.

Why set an autonomy ceiling, not just a depth?

Depth is normally earned: an agent proves a capability on the lender's own cases and gains depth as evidence supports it. That approach, often called graduated autonomy, is right for most work.

Completion is a case for a ceiling instead. An autonomy ceiling caps the depth a capability can reach, whatever the evidence. For mortgage completion release, the ceiling should be set at drafting. No amount of good performance on past completions should let an agent release funds on the next one, because the cost of a single failure is large, the event is irreversible, and the main threat, fraud through redirected payee details, is designed to look like a normal case.

That is not distrust of the agent. It is recognition that some actions should always have a person's name on them.

What controls belong around completion?

1. Destination verified independently. The account receiving funds is confirmed against a record the lender holds, not against correspondence received during the transaction.
2. No late changes from external sources. Any change to payee details close to completion is held and verified by a call to a number already on file.
3. Conditions confirmed. Every condition of the offer is shown as met, with the evidence, in the release pack.
4. Named release. A person releases the funds and their name, the time and the facts they saw are filed.
5. Separation. The person releasing is not the person who prepared the pack, and no agent can approve another agent's preparation.

What about smaller mortgage actions?

Not every mortgage action is completion. Many servicing tasks after completion are routine and bounded. Handling an overpayment, for example, involves receiving money rather than sending it, and the main risks are in applying it correctly and explaining the effect. The guide on whether an AI agent can handle a mortgage overpayment request shows how much more room there is for agent execution once the large, outbound, irreversible step is behind you.

Where Pryme fits

In Pryme, the agent's role sets the ceiling, Attest evidence sets the depth, and the lower wins. A lender can scope the role so completion release stops at drafting, and a named person always makes the release, while the agent's certified depth 3 still covers document gathering and condition tracking. See AI agents in banking: what they may and may not do.