Variance commentary needs a person's judgement

An agent can calculate every variance in the management accounts and draft a first explanation for each. The reasons put in front of management still need a person's view. Commentary is not arithmetic; it is a claim about why the business moved, and only someone who knows the business can say whether it is true.

What can the agent do well?

The mechanical half of variance analysis is substantial, and agents handle it well. They can compare actuals with budget, forecast and prior periods at any level of detail, rank variances by size, and trace each one to the ledger entries behind it. They can spot that a cost line moved because of one large invoice, a timing difference in an accrual, a reclassification or a run of smaller transactions.

They can also draft explanations: "Marketing spend was 40,000 over budget, mainly due to an invoice of 35,000 from a new agency posted in this period." That is a useful first draft. It is factual, traceable and quick to check. Whether an AI agent can do variance analysis is answered yes for this part of the work.

The full process is set out in how to prepare variance commentary for management accounts with an AI agent.

Why do the reasons need a person?

Because the ledger tells you what moved, not why. The agent's draft above is true, but it is not the explanation management needs. Management needs to know: was the new agency planned and the budget wrong, or was it unplanned spend? Is it a one-off or the start of a run rate? Did it produce results? Should the forecast change?

Those answers live outside the ledger: in conversations with budget holders, in decisions taken at meetings, in knowledge of what the business is trying to do. A management accountant gathers that context as part of the job. An agent reading the books does not have it, and if it guesses, the guess can be plausible and wrong.

There are specific traps:

• Timing versus substance. A variance caused by an invoice arriving early looks the same in the ledger as one caused by overspending. The commentary should say which.
• Several causes. A revenue variance may combine price, volume and mix effects. An agent may name the largest and miss that the others are moving in the opposite direction.
• Management action. Commentary often needs to say what is being done about a variance. Only management knows.
• Sensitivity. Some reasons, such as a dispute with a customer or a restructuring, need careful wording or should not be in a wide-circulation pack at all.

What does a good draft look like?

One that separates what the agent knows from what it is inferring, so the person can see where to add judgement.

Leaving the last two blank, or clearly marked as hypotheses, matters more than it might seem. A draft that fills every field with confident prose invites the reviewer to accept it. A draft that shows its gaps invites the reviewer to fill them.

How do you stop the draft becoming the commentary?

This is the real risk. Under month-end pressure, a management accountant receives forty well-written explanations and approves them with light edits. The agent's hypotheses become the official reasons, without anyone having checked them with the budget holders.

Several practices help. Require the person to confirm or replace every "why" field, so approval cannot happen without engaging. Route material variances to the relevant budget holder for a line of input before the commentary is finalised. Track how often the agent's hypotheses are accepted unchanged; if almost all are, check a sample against what budget holders actually say. And keep the agent's draft and the final commentary as separate records, so it is always clear which words were the agent's and which were a person's.

Who signs the commentary?

The person who will answer for it at the management meeting. Usually that is the management accountant who prepared the pack, reviewed by the financial controller or finance director. If a board member asks about a variance, the answer comes from a person who understands it, not from an agent's draft.

This mirrors a pattern found across regulated work. Agents gather and prepare; people judge and answer. The same logic appears in compliance, where enhanced due diligence still needs a person, and in payments, where decisions such as scam reimbursement need a person. In each case, the agent's draft is valuable because it lets the person spend their time on judgement rather than on assembly.

What changes for the finance team?

The work shifts from producing commentary to testing it. Instead of building variance schedules and writing explanations from scratch, the management accountant receives a traced, structured draft and spends their time talking to the business about the reasons. That is the part of the job that adds most value, and the part most often squeezed when time is short.

It also improves consistency. Every variance above the threshold is analysed the same way, with the same trace, every month. Gaps in commentary, such as a material variance with no explanation, become visible instead of quietly missing.

What should you do differently?

Let the agent produce the numbers, the trace and a clearly labelled hypothesis. Require a person to write or confirm every business reason, ideally with input from the budget holder. Keep the agent's draft and the final words distinct in the record. Commentary that management can rely on is part of AI agents for finance: books you can audit.