Can an AI agent run depreciation on fixed assets?

Yes, the monthly run is well suited to an agent. Depreciation applies fixed methods and lives to an asset register, so once the register agrees to the ledger, an AI agent can post the charge at depth 3. Decisions about the inputs, such as useful lives, impairments, disposals and method changes, are drafted for the financial controller to approve.

What makes the monthly run safe to hand over?

The calculation itself is set when each asset is added. Cost, date brought into use, useful life, residual value and method are recorded once and then applied every month. A straight-line asset costing 60,000 over five years depreciates by 1,000 a month until it is fully written down. Nothing in that needs judgement after the first day.

Errors come from the register, not the arithmetic. An asset added with the wrong life. A disposal never recorded, so a sold vehicle keeps depreciating. An asset fully depreciated but still charged. These are checks the agent can run every month before it posts.

What conditions must hold before the agent posts?

• The register agrees to the ledger. Cost and accumulated depreciation by category reconcile to the fixed asset accounts.
• Additions for the period are approved. New assets have been capitalised and set up with an approved life and method. The set-up process is covered in how to prepare fixed asset additions and depreciation.
• The period is open.
• No asset is past its life or below its residual value. These are flagged, not charged.
• The total charge is within the expected range. Based on the prior month plus additions less disposals.

If all hold, the agent posts. If any fails, it drafts the run and holds it.

What does the agent decide, and what does the controller decide?

The separation of duties is between whoever sets the inputs and whoever runs the charge. Changing an asset's life changes profit for every remaining year. The person who approves that change should not be the budget holder who benefits from a lower charge, and the agent cannot make the change itself.

An illustrative example

Take a fitness chain with about 4,000 assets across 25 gyms. At month end the agent reconciles the register to the ledger and finds they agree.

It calculates the monthly charge of 212,000 and checks it against the expected range: the prior month's 205,000 plus new equipment at two refurbished gyms. Within range. Before posting, it finds 18 treadmills at one site that were reported as scrapped in a maintenance log three months ago but are still in the register. It removes them from the run, posts the slightly lower charge and drafts disposal entries for the 18, with the maintenance log attached.

It also notices that a gym closed six weeks ago, but its fit-out assets are still depreciating over their original lives. That is an impairment indicator. It flags it to the controller, who decides on an impairment review.

What controls make it safe?

• Inputs approved by people. Lives, methods and residual values are set by the controller under the asset policy.
• Register reconciled first. No posting from an unreconciled register.
• Range check on the charge.
• Disposal and impairment flags. From maintenance logs, site closures and sale records.
• Every run filed. Register snapshot, calculation and posting.

Where does depreciation sit in the close?

It is one of the routine tasks in the month-end close, typically run after additions are approved and before the management accounts are drawn up. Like running payroll, it repeats every period on fixed rules, which is what makes it suitable for an agent. Unlike payroll, it moves no money, so it can be posted directly once the inputs are right. The wider controls are in AI agents for finance: books you can audit.