AI playbook for Finance and Administration: what to delegate, what to keep

We wrote last time about Programmes and Monitoring, the first of the seven department playbooks we are publishing properly. Finance and Administration is next, and it is arguably the department where the case for delegation is easiest to make and easiest to get wrong at the same time. The volume of repetitive, rules-based work is enormous, and the cost of a single unnoticed error is measured in real money leaving the organisation.

This piece sorts the department's real work the same way as last time: what is genuinely worth delegating, and what never should be, whatever the efficiency case looks like on paper.

What the department actually spends its time on

Finance and Administration lives on invoice processing, transaction reconciliation, expense report drafting, budget-versus-actual tracking, payroll administration, procurement paperwork, and preparing financial reports for the board. Most of it is rules-based matching and drafting against numbers that already exist. A narrow but critical slice of it is the moment money actually moves, or a named person answers for whether it should have.

What's worth delegating

• First-pass reconciliation of transactions against the ledger. Flagging exceptions for a human to review is mechanical matching, not a judgement call, and it catches discrepancies faster than a tired analyst working through a spreadsheet at month end.
• Drafting routine expense reports from receipts already submitted and categorised. Turning already-categorised data into a formatted report is exactly the kind of task that should never eat an administrator's afternoon.
• First-draft budget-versus-actual summaries from clean accounting data. Producing the narrative first draft is aggregation, and a human still decides what the variance actually means before it reaches the board.
• Organising procurement paperwork into the file structure auditors expect. Filing and tagging documents correctly is clerical work with no financial judgement attached to it.

What never gets delegated

• Actually authorising a payment. Whoever approves money leaving the account has to be a specific, accountable human, every time, regardless of how routine the payment looks on the surface.
• Signing off the year-end financial statements. This is the one moment auditors and the board rely on a named person's judgement and signature, and that reliance cannot transfer to a tool.
• Investigating a suspected irregularity. The moment a number looks wrong in a way that suggests deliberate manipulation rather than an honest mistake, this becomes a human judgement call with legal and reputational consequences attached, not a data-cleaning task.

A tool can reconcile the numbers. It cannot answer for the money.

The trade-off finance carries more sharply than most departments

Finance carries a higher consequence for error than most departments, which makes the temptation to over-restrict delegation entirely a real one: refuse to delegate anything, and the department pays in staff hours spent on pure aggregation that never needed a human judgement in the first place. The honest answer is that the line in finance is unusually sharp precisely because money leaving the account is auditable and irreversible in a way most other departments' mistakes are not. That sharpness applies at the point money actually moves. It should not spill over into treating reconciliation and drafting, which never touch the account directly, as equally untouchable.

The practical takeaway

Delegate the mechanical matching and drafting that produces no money movement by itself; never delegate the moment a human authorises a payment, signs a financial statement, or judges a suspected irregularity.

No external statistic cited; this article presents an internal department playbook rather than third-party evidence.