How to prepare fixed asset additions and depreciation with an AI agent

The agent finds candidate additions in capital and work-in-progress codes, proposes each with its invoice, order and project, suggests an asset class and useful life from your policy, and simulates depreciation. The accountant approves every capitalisation, every useful life and every in-service date. The agent posts the monthly depreciation run only once certified, from the approved register.

What does fixed asset accounting involve, and who owns it?

Spending on assets that will be used for more than a year, such as equipment, vehicles, fit-out and some software, is capitalised rather than expensed, then depreciated over its useful life. Each month the team must decide what new spending is capital, add it to the fixed asset register with a class, cost, life and in-service date, move completed projects out of work in progress, record disposals and run depreciation.

The fixed asset accountant, often part of the general ledger team, owns the register. The financial controller owns the capitalisation policy: the threshold below which spend is expensed, the asset classes and their default lives. Project managers know when assets go into use, which is the date that starts depreciation.

Whether an agent can run depreciation on fixed assets is the narrow question. This playbook covers the wider monthly cycle, inside the bounds set out in AI agents for finance.

What does the agent prepare?

1. Agent, depth 1: collects candidates. Invoices coded to capital or work-in-progress accounts, and expense invoices above the capitalisation threshold with descriptions that suggest an asset.
2. Agent, depth 2: proposes each addition. Invoice, purchase order, project code, description, cost, proposed class and useful life under the policy, and the evidence for each.
3. Agent, depth 1: checks work in progress. Projects with no new cost for several months, or marked complete in the project system, are listed for transfer to assets in use.
4. Agent, depth 2: proposes disposals. From sale invoices, write-off requests and asset verification results, each with the net book value and gain or loss.
5. Agent, depth 2: runs depreciation in simulation. The month's charge by class and cost centre, compared with last month, with every change explained.
6. Person, fixed asset accountant: approves additions, transfers and disposals. Including class, life and in-service date.
7. Agent, depth 3 once certified: posts the depreciation run. From the approved register only, within a variance limit against the simulation.
8. Person, controller: reviews the roll-forward. Opening cost and depreciation, additions, disposals, charge, closing, tied to the ledger.

What does a proposed addition look like?

Take an illustrative logistics company that buys three forklifts for 90,000 in total on one purchase order, plus a 4,000 invoice for delivery and installation. The agent proposes one addition per forklift at 31,333 each, because delivery and installation are part of the cost of getting the assets ready for use. It suggests the plant and machinery class with the policy's default life, and asks the depot manager for the date the forklifts went into service rather than using the invoice date. A separate 600 invoice for a first service is left as expense, with a note saying why.

What does the named person approve, and what evidence remains?

The fixed asset accountant approves three judgements: is this capital spend, what kind of asset is it and how long will it last, and when did it go into use. Each approval is recorded against the proposal with any change made. For a sample of additions, they also confirm the asset exists.

Policy reference. Each proposal cites the policy clause for threshold, class and life. Evidence: policy version on the proposal.

Source documents attached. Invoice and order on every addition. Evidence: links on the register entry.

Simulation before posting. Evidence: the simulation report against the posted run.

Roll-forward tie-out. Evidence: a signed roll-forward reconciled to the ledger.

Where does it usually go wrong?

Repairs capitalised. An invoice for maintenance on an existing machine is coded to capital. Repairs restore an asset; they do not improve it. The agent should flag words like "repair", "service" and "replacement part" for a person to judge.

Depreciation before the asset exists. Additions are given the invoice date as the in-service date, so depreciation starts before the equipment is installed. Ask the project owner for the date.

Work in progress that never moves. A finished fit-out sits in work in progress for a year, undepreciated. The agent's monthly list of stale projects stops this.

Double counting. An asset accrued at month end and then invoiced is added twice. The agent should match additions to accruals and to earlier additions from the same order.

Forecast disconnect. Planned capital spend is in the forecast, but its depreciation is not. Feed the agent's simulation forward into rolling forecasts and into the budget so the charge does not surprise anyone.

What should you measure?

• Proposals changed by the accountant, by type of change.
• Assets in work in progress beyond a set age.
• Depreciation run variance against the simulation.
• Audit adjustments on fixed assets.
• Assets not found during physical verification.

Where Pryme fits

In Pryme, proposing additions, running the simulation and posting depreciation are separate capabilities, each certified on your own past months. Invoices arrive as external documents and the approved register as sealed, so the run posts only from approved data. Proposals are held for the accountant and every decision is filed.