Close and Reporting Acceleration

A final trial balance answers "what are the numbers." It does not answer "why did they move." That second question is the job of management's fluctuation analysis, usually shortened to flux: a written explanation of why each major line changed from the prior period, in language a CFO can take to the board. The Institute of Management Accountants frames this kind of management reporting as decision-useful analysis rather than a data dump, so the reader gets the drivers and their implications, told in a measured internal voice rather than a marketing one.

At Meridian the deadline is concrete. The October trial balance is final, and the CFO wants flux commentary by 2 pm, ready to drop into the board package. The analysis is not conceptually hard, but it is fiddly and slow. You have to find the movements that matter, work out what caused each one, write it up in a controller's voice, and make sure each figure ties. Little of that is arithmetic; most of it is judgment and language.

Materiality decides which movements are worth explaining, and it is more subtle than a size cutoff. SEC Staff Accounting Bulletin No. 99 cautions against treating a single percentage threshold, such as a five-percent rule of thumb, as the sole test of what is material; a quantitatively small item can still matter for qualitative reasons, for example when it changes an earnings trend, affects a loan covenant, or tips whether a target is met. FASB Statement of Financial Accounting Concepts No. 8, Chapter 3, frames materiality the same way, as an entity-specific aspect of relevance judged against what would influence a primary user of the statements. In practice a controller sets a working dollar threshold as a first screen and then overrides it upward or downward on qualitative grounds.

Tie-out is the discipline of tracing each figure in the narrative back to its source in the trial balance, so the words and the numbers do not drift apart. Both disciplines are settled before drafting because they are the analyst's judgment rather than something to delegate. A fluent draft that skips them can read beautifully and still be wrong.

The standard is set before any tool touches the data. Close-ready commentary has four properties. It is reconciled, so each figure ties back to the trial balance. It is materiality-screened, so it explains the movements that matter and stays quiet on noise. It is correctly toned, reading in a controller's measured voice rather than in marketing language. And it is free of invented drivers, so each cause named is one the data actually supports. In the lab the working threshold is 75,000 dollars, applied as a first screen and then read against the qualitative factors above.

The task suits AI assistance because, once the trial balance is final, most of the remaining work is language rather than arithmetic. The variances are a mechanical calculation with one correct answer each. The judgment, which movements deserve explanation, what defensible driver sits behind each, and how to phrase it in the house voice, is where the analyst adds value. That split, deterministic math on one side and judgment plus narrative on the other, is what the rest of this module is built around.

The most useful thing you can give a model for this task is an example of the destination. Last quarter's flux commentary is exactly that: it shows the house style, the level of detail, and the tone the CFO expects. With that example in the folder, you are not asking the model to guess what good looks like; you are asking it to produce more of a known shape. The journey is clear: you start with a trial balance, you are going to close-ready commentary, and the worked example anchors the far end.

Computing a variance is deterministic work: for a given prior and current balance, there is one correct dollar change and one correct percent. Asking a language model to produce those numbers in prose invites small, hard-to-catch errors. So the workflow puts the math where it belongs. A flux template computes every account's variance in dollars and percent, and the model narrates only the results. The model performs the non-deterministic work, the part with many acceptable phrasings: deciding how to describe each material driver and how to sequence the commentary.

No other design choice here matters as much: when the model recomputes figures in prose, two versions of the numbers can exist. When the template owns the math, there is one source of truth, and validation becomes a tie-out rather than a recalculation.

A single line makes the split concrete. Inbound freight in the Meridian folder reads $540,000 in September and $690,000 in October. The workflow carries that line from raw balances to one close-ready sentence, and each step has a clear owner.

• Template, deterministic. The dollar change is $690,000 minus $540,000, or $150,000. The percent is $150,000 divided by the prior-period base of $540,000, or 27.8%. There is one correct answer for each, so the template computes both and the model is told not to recompute them.
• Screen, mechanical. The $150,000 movement clears the $75,000 materiality threshold, so the line earns commentary. Had it come in below $75,000, the workflow would drop it at this step and it would not reach the model at all.
• Driver, judgment. Reading the cause is the non-deterministic part. The month's revenue gain was price with flat volume, so units did not rise, and the folder's rate detail shows carrier rates climbed. That makes the freight step a rate-driven margin headwind, not a cost that scaled with sales.
• Narrate, non-deterministic. The model turns the computed figures and the confirmed driver into one measured line: "Inbound freight rose $150,000 (27.8%) on higher carrier rates; with revenue up on price and volume flat, this is a margin headwind rather than a cost that moved with sales."

The lab folder is deliberately small: the current trial balance, the prior trial balance, the flux template, and the one worked example. Nothing else. A larger pile of schedules would bury the accounts that actually moved and raise the chance the model latches onto something irrelevant. Minimum context is the fuel: give the model exactly what the task needs, and the destination example so it knows the shape of the answer.

The worked example does work here that written instructions tend to miss. It encodes the tacit house style, the level of detail, and the sequencing a controller would struggle to spell out as a rule, and the model tends to match a shown example more reliably than it follows an abstract description of one. That is why the folder carries a finished commentary rather than a longer style guide.

The five moves are not a one-off. Because the close runs the same way each period, the prompt that carries these instructions is worth saving as a reusable skill or prompt template: the destination example, the deterministic-math rule, the minimal-folder list, and the house-voice guidance, packaged so next month starts from the same known shape rather than a blank page. Standardization is exactly what close benchmarking rewards, and a saved skill is how you standardize the AI step.

Pair it with a short close-ready checklist the analyst runs before sign-off: tie each figure in the draft back to the trial balance, confirm each named driver against the folder or flag it as unconfirmed, screen out any line below the materiality threshold, and check tone and rounding. The skill makes the draft repeatable; the checklist makes the review repeatable. Together they keep the human checkpoint from becoming a rubber stamp.

Download the folder and run the flux pattern in whatever AI you use. The folder holds Meridian's October and September trial balances, a flux template with the variances already computed, and last quarter's commentary as your style example. Let the template hold the math, have the model draft commentary for the material movements only, and then come back for validation. The company and every number are fictional.

The heart of validation is the tie-out: take each figure in the commentary and follow it back to the trial balance or the flux template. A single untied number is enough to send the whole package back. Confirm, too, that variance percents are computed on the prior period, that every account above the 75,000 dollar threshold is addressed, and that nothing below it got its own paragraph. Work the checklist below against your draft before you would ever call it done.

The exemplar below is what you would drop into the board package. It leads with the material drivers, ties each figure to the trial balance, distinguishes one-time from run-rate, and stays silent on the noise.

Revenue. Total revenue rose modestly, driven by Industrial Fittings, which increased $410,000 (5.0%) on the list price increase that took effect in October. Volume across all three lines was essentially flat, so the month's growth is price rather than demand. The other two revenue lines moved within the immaterial range.

Cost of goods sold. Inbound freight rose $150,000 (27.8%) on higher carrier rates. Because the revenue increase was price-driven with flat volume, the freight step is a margin headwind this month rather than a cost that moved with sales. Materials and direct labor were within the immaterial range.

Operating expenses. Engineering payroll rose $87,000 (10.6%) on two hires who started in October; this is a run-rate increase and should be expected to persist. Legal and professional expense rose $150,000 (157.9%) on a one-time settlement; excluding that item the line was roughly flat, and the increase should not be extrapolated. All other operating lines moved within the immaterial range and are not broken out.

Bottom line. The month's story is a price-led revenue gain partly offset by a freight-rate headwind and a step up in engineering run-rate, with a single one-time legal item inflating operating expenses. Underlying operating performance, excluding the settlement, was steady.

The lab is clean by design: the anomalies are seeded, the folder is minimal, and materiality is stated. Real closes are messier, and the workflow tends to strain in three specific places.

• Materiality that shifts late. A line you judged immaterial in an early draft can cross the threshold after a late adjusting entry, so commentary built before the trial balance truly settles can miss a driver that the reviewer then catches.
• Drivers that live outside the ledger. The trial balance shows freight rose but not why; the real cause, a renegotiated carrier contract or a one-off expedite, often sits in procurement's inbox, and the model will fill that gap with a confident guess if you do not confirm it first.
• Reclass movements that are not economic. A reclassification between two accounts can make one line look like it spiked while the offsetting line quietly fell, and a model reading only the balances can narrate a "spike" that reflects no real change in the business.

On top of these, the model will sometimes produce commentary that is fluent, specific, and wrong, which is exactly why the tie-out and driver checks are not optional. The workflow does not remove the controller's judgment; it removes the mechanical drudgery so the judgment has more room.

Rate your own commentary against the rubric below. Scored honestly against the rubric, the result shows where the flux workflow is solid and where it still needs work. Your scores roll up to the workflow maturity dashboard on the course hub.