Management reports that explain their own variances
Finance teams spend nearly half their effort collecting and checking data, so the commentary a board reads is written last. An agent can trace each variance to its transactions and draft the explanation for a person to approve.
The short version
- The 2026 FP&A Trends Survey of 475 finance professionals found that 47% of the planning and analysis function's effort goes on collecting and validating data, and only 32% on insight and driving action 1.
- APQC's open benchmark puts the median at 10 calendar days from running the first trial balance to completing the period-end management report, across 3,287 organisations 2.
- Gartner's 2025 survey of 183 finance leaders found 59% using AI in their finance function; among those, the most common uses were knowledge management (49%), accounts payable automation (37%) and error and anomaly detection (34%) 3.
- An agent can reconcile the report to the trial balance, find every variance over your threshold, trace each one to the transactions behind it, draft the commentary with those transactions attached, and hold what the ledger cannot explain for the person who owns the cost.
- A person still approves every word before it reaches the board, and the measure is your own: hours spent on the report, the share of explanations accepted without rewriting, and the questions the board asks that the pack already answers.
The explanation is the part a board reads, and the part written last
A monthly management report has two halves. The first is the numbers: actuals against budget and last year, by business unit and cost line. The second is the explanation: why freight is over, why margin slipped in one branch, which of the movements will repeat. Directors and executives read the second half. It is also the half written last, by people who have spent most of the month on the first.
The 2026 FP&A Trends Survey, from FP&A Trends with 475 respondents, puts a number on that. Nearly half the planning and analysis function's effort, 47%, goes on collecting and validating data, and only 32% on generating insight and driving action 1. The same survey found that only 19% of respondents rate their data quality as advanced, and only 11% of organisations have real-time access to their data 1.
The time it takes shows in the calendar. APQC's open benchmark measures the calendar days from running the first trial balance to completing the period-end management report. Across the 3,287 organisations in its sample, the median is 10 days 2. That is a global, cross-industry figure, not an Australian one, and APQC does not publish the collection dates on the open page. But it fits the experience of most finance teams: the report reaches the executive team well into the next month, and the commentary is the last thing finished.
Why explaining a variance takes so long
The figure that is over budget sits in one place. The reason it is over sits in five: the ledger lines behind it, the supplier invoices, the purchase orders, the payroll export, and an email thread with the branch manager who knows what happened. Someone in finance has to open each of them, work out which items add up to the gap, and then write two sentences a director can read in ten seconds.
Multiply that by every variance over the threshold, in every business unit, and the commentary becomes a scramble. Explanations get shorter, vaguer and more alike from month to month: "timing", "higher activity", "one-off costs". Those words are not wrong. They are what gets written when there is no time left to find the transactions.
This is the part of the report an agent is suited to. Finding the transactions that make up a gap is a search through records the business already holds, with a sum at the end that either adds up or does not. Writing the first draft of the explanation from those transactions is a language task. Judging whether the explanation is the right one, and whether it matters, stays with a person.
Gartner's 2025 survey found that 59% of finance leaders report using AI in their finance function, up from 58% the year before, and named three uses taken up by more than a third of those using it: knowledge management (49%), accounts payable automation (37%) and error and anomaly detection (34%) 3. Explaining variances is close kin to the third. Finding an anomaly is step one; the board report needs step two, which is saying what caused it in plain words, with the evidence.
Smaller Australian businesses are turning to AI for analysis too. The National AI Centre's tracker found that 43% of Australian small and medium businesses reported some level of AI adoption over December 2025 to February 2026, and that data analytics and content generation were the most common uses, each named by 54% of adopters 4. The difference between that and a management report a board can rely on is the build around it.
Where the hours go in one month's report
Take a wholesale distributor with six branches. Its monthly report compares actuals with budget for each branch's cost lines, and finance explains every variance over $5,000 that is also more than 10% of budget. Most months that is about 40 variances. The business and its figures are invented; the steps will be familiar to any finance team.
2,880 minutes is 48 hours a month, most of it before anyone writes a word of commentary. The writing itself is the smallest line, and the first to be cut short when the month runs late.
What one explained variance looks like
The output that matters is not a dashboard. It is a single variance, explained to the dollar, with the transactions one click away. Here is the freight line for one branch in the same invented month, as the agent would set it out for a person to check.
The draft commentary reads: "Freight was $13,400 over budget. $6,200 is fuel surcharges from one carrier and is likely to continue; $5,400 is two extra runs for one customer order; $1,800 is last month's invoice posted late and will not repeat." A person checks it, edits it if needed and approves it.
Variance commentary by hand
- Variances are explained in the last days before the report goes out
- Each explanation means opening the ledger, invoices and email by hand
- Commentary drifts towards "timing" and "higher activity"
- A director who asks "what's in that number?" waits for an answer
Variance commentary with an agent
- Every variance over the threshold is traced to its transactions as soon as the ledger closes
- The draft explanation carries the transactions that add up to the gap
- What the ledger cannot explain goes to the person who owns the cost, with the evidence
- Finance edits and approves every explanation before it reaches the board
What we would build
The build is an agent that works inside the systems a finance team already uses: the ledger and its budgets, the purchasing and payroll records behind it, and the report the board receives. For a distributor that might be Xero, MYOB Acumatica, NetSuite or Microsoft Dynamics 365 Business Central, with the report in Power BI, Excel or a Word pack; the products are examples and the design does not depend on them. It does seven things.
- It reads the closed ledgerReads
The trial balance, budgets and last year's actuals for every business unit and cost line, and the transactions behind them: supplier invoices, purchase orders, journals and the payroll summary.
- It reconciles before it explainsChecks
Every table in the report is reconciled to the trial balance first. If a figure does not tie back, the agent stops and says where, rather than explaining a number that is wrong.
- It finds every variance that mattersFinds
Against budget and last year, using your thresholds, such as over $5,000 and over 10%. Nothing is left out because the month ran late.
- It traces each one to its transactionsTraces
Works out which transactions make up the gap and shows the sum. An explanation that does not add up to the variance is not offered as one.
- It drafts the commentaryDrafts
Two or three plain sentences for each variance, saying what caused it and whether it is likely to repeat, with the transactions attached so a reader can check them.
- It holds what the ledger cannot explainHolds
A write-off with no reason recorded, a cost with no order behind it or a gap that does not add up goes to the person who owns that cost, with a specific question and the transactions attached. It never posts a journal or changes a budget.
- It proves itselfMeasures
Hours spent on the report, the share of draft explanations approved without rewriting, the variances explained to the dollar and the board's follow-up questions, against three months' baseline taken before go-live.
In the worked example, the agent produces the tables and reconciles them, and a person checks the reconciliation in an hour. Of the 40 variances, it traces 32 to their transactions and drafts the commentary; a person reviews each in about five minutes, 160 minutes in all. The other 8 are held, such as a $9,600 stock write-off at Branch 5 with no reason recorded, and each takes about 25 minutes to resolve with the branch manager. That is 60 + 160 + 200 = 420 minutes: the month's report falls from 48 hours to 7, and the hours that remain are spent on judgement. Illustrative figures again; the shape is what matters.
Watch 40 variances explained two ways
A 30-second film, with no sound to turn on: a month's management report explained by hand, then by an agent that traces each variance and holds what it cannot explain for a person.
Read what the film shows
- By hand. Each month the report takes 18 hours to export, reconcile and lay out, and each of its 40 variances takes 45 minutes to trace, ask about and explain. That is 48 hours, with the commentary written last.
- With an agent. The agent reconciles the report to the trial balance, traces 32 variances to their transactions and holds 8. Branch 3's freight, $13,400 over, is explained to the dollar, and Kim approves the commentary.
- What changes. The finance team spends 7 hours a month on the report, not 48, and every explanation in it adds up to its variance and has a person's approval.
Keep a person's name on every explanation
The board pack goes out under the chief financial officer's name, and an agent's draft should never change that. The National AI Centre's Guidance for AI Adoption makes maintaining human control one of its six essential practices: a person should oversee an AI system in a way that matches how much autonomy it has and how high the stakes are, and there should be clear points where people can pause or override it 5. For a management report that means three rules. The agent drafts and a named person approves every explanation. The agent reads the ledger and never writes to it. And anything that does not add up is held, not smoothed over.
Start with your own report, not the software. Take last month's pack and count the variances it explained, the hours it took and how many of the explanations a director could check. Then take three months of those figures before anything changes. If the hours, the rewrites and the board's follow-up questions do not fall, the build has not worked, whatever a demonstration shows.
Questions for your board
The first three can be answered by the finance team before the next report.
- How many days after month-end does the management report reach us, and which step takes longest?
- How many variances does each report explain, and what threshold decides which ones?
- Could finance show us the transactions behind any explanation in the last pack, and how long would it take?
- Which of our explanations recur month after month, and are they still true?
- How much of the finance team's month goes on collecting and checking data rather than analysis?
- If an agent drafted the commentary, who would approve it, and what would it never be allowed to change?
See reporting that runs itself
Our page on automated reporting shows how management and board reports can be assembled and explained from the systems you already run, with a sample report and an agent you can try.
See AI reporting automationSources
- FP&A Trends, Closing the AI readiness gap: 2026 FP&A Trends Survey (9 July 2026)
- APQC, Cycle time in days to produce period-end management reports (Open Standards Benchmarking)
- Gartner, Gartner survey shows finance AI adoption remains steady in 2025 (18 November 2025)
- National Artificial Intelligence Centre, AI adoption insights: December 2025 to February 2026 (3 June 2026)
- National Artificial Intelligence Centre, Guidance for AI Adoption: Foundations (October 2025)