Technical Accounting Memo Workflow

When a technical-accounting team reaches a position on a question like this, the deliverable is an issues memo: a short document that spots the accounting issue, maps it to the guidance that governs it, works through the analysis, states a conclusion, and shows the alternative it ruled out. The format is deliberate. It forces the reasoning onto the page in an order a reviewer can follow, so a signer, an auditor, or the audit committee can trace how the position was reached rather than take it on faith. The first move is issue spotting: reading the facts and naming the question that actually has to be decided. Practitioners rarely read ASC 606 cold, either; they work from it alongside the Big 4 revenue handbooks (KPMG's Handbook: Revenue recognition and the PwC and Deloitte revenue guides), which collect the interpretive guidance and worked fact patterns the codification itself leaves terse.

Across those sources the standard for a memo is consistent, and it defines the finish line the rest of this module aims at. A supported conclusion has three properties. It is cited, so the position traces to a specific codification paragraph rather than to a confident assertion. It rests on the full set of criteria the guidance names, not on one convenient fact that happens to point the right way. And it addresses the alternative treatment, showing the position was reached by comparing options and ruling one out, rather than by settling on the first answer that appeared. Grounding each conclusion in the paragraph that governs it, and checking the conclusion against each criterion that paragraph lists, are the accountant's responsibility, and a fluent memo can read like authority while quietly skipping one.

You support technical accounting at Meridian Components, and sales has signed a new arrangement: Meridian will build a highly customized assembly line for Riverbend Fabrication for a fixed price of 2,400,000 dollars, constructed to Riverbend's specification over roughly seven months that span a period end, with milestone payments along the way. The controller needs a memo before the quarter closes answering one question: should Meridian recognize revenue over time as it builds, or at a single point in time when the line is delivered.

The issue is the timing of revenue, and under ASC 606-10-25-27 it turns on the custom-asset criterion: whether the assembly line has no alternative use to Meridian, and whether Meridian has an enforceable right to payment for work completed to date. Both conditions have to hold together for over-time recognition; one on its own does not settle it. That is the analysis the memo has to work, criterion by criterion and in order, rather than jumping to the answer that feels right.

This is a strong candidate for AI assistance because most of the work is reading and writing rather than arithmetic. Spotting the issue and mapping it to the governing paragraph is a checkable lookup; drafting the analysis and the position is exactly the kind of structured prose a model produces quickly. What stays with the accountant is the judgment: deciding whether the facts satisfy each criterion, and standing behind the conclusion and the citation. Naming that split, the checkable mapping on one side and the judgment plus prose on the other, is the design of this workflow, and Part Two builds it.

The most useful thing you can give a model for this task is an example of the finished shape. A prior supported memo on a similar issue shows the house structure, the depth of the analysis, and how a position is stated and defended. With that example in the folder, you are not asking the model to invent what a good memo looks like; you are asking it to produce more of a known form. You start with facts and guidance and are heading to a supported position, and the example anchors the far end.

An example carries what an instruction struggles to spell out: the tacit standard for how much analysis each criterion earns, where a citation belongs in a sentence, and how firmly a position is stated before it overreaches. The catch is that the model copies the example's habits as faithfully as its structure, so the memo you hand it should be one you would sign yourself. A weak exemplar teaches weak structure just as reliably as a strong one teaches a good one, which is why the anchor is a prior supported memo, not merely a prior memo.

A memo is mostly prose, not arithmetic, but it still splits into a checkable part and a judgment part, and separating them is the point. The checkable part is the mapping: for each spotted issue, which codification paragraph governs it. That is a lookup against the source a reviewer can confirm directly. There is also a small deterministic sliver: once over-time recognition holds, the amount recognized this period comes from a measure of progress, and a cost-to-cost measure is just costs incurred divided by total estimated costs, times the transaction price. That belongs in code so it recomputes and ties out, not in model prose that can quietly round or drift. The judgment part is weighing whether the facts satisfy each criterion in the governing paragraph, and reaching a position. That is where the accountant, and a deliberate critique step, add the value.

Because the judgment part is where confident errors hide, the workflow does not trust a single pass. It uses draft-critique-revise: the model drafts a position, then a deliberate critique treats that draft as a hypothesis and tests it, and only then is the memo revised into something to rely on. When a first draft is accepted as an answer, a skipped criterion or a mis-cited paragraph rides along unnoticed. When the draft is treated as a claim to be challenged, those errors have a place to surface.

The lab folder is deliberately small: the fact pattern, the simplified guidance excerpts, and, for the peer-review task, one AI-drafted memo to critique. Nothing else. A larger pile of standards and precedents would bury the paragraphs that actually govern and raise the chance the model cites something the memo does not support. Minimum context is the fuel: give the model the facts, the guidance it is allowed to cite, and the destination example, and keep outside material out unless you add it as a source.

Two pieces of scaffolding make this repeatable rather than a one-off. The first is a reusable memo prompt or skill: a saved instruction, in whatever assistant your firm has approved, that fixes the house structure (issue, guidance, analysis, position, alternative), states which guidance the model may cite and that it should cite nothing else, and points at the destination example. Saving it once means each memo starts from the same supported shape rather than from whatever the model improvises that day.

The second is a critique checklist the model runs against a draft, and that you run behind it: does each cited paragraph exist and say what the memo claims; is each criterion in the governing paragraph addressed rather than one asserted as sufficient; is any fact assumed that the folder does not establish; is the alternative treatment ruled out rather than ignored. Those four questions map to the four errors this module's peer-review task is built around, and Part Five turns them into the validation pass. A checklist is what keeps the critique from sliding back into a vague reread.

Download the folder and run the pattern in whatever AI you use. The folder holds the Meridian and Riverbend fact pattern, simplified guidance excerpts written for the lab, and an AI-drafted memo that contains planted errors. There are two tasks. First, draft a supported issues memo from the facts and the guidance: spot the issue, map it to the governing paragraph, reach a position, and name the alternative you considered. Second, peer-review the flawed memo and produce a redline that says, for each error, what is wrong, why, and what it should say. Cite only the guidance in the folder. The company, the counterparty, and each figure are fictional, and the paragraph numbers are simplified stand-ins for the lab.

The heart of validation is two passes. First the citation check: take each cited paragraph and confirm the number is right and the paragraph actually says what the memo claims. A single mis-cited criterion is enough to send the memo back. Second the logic check: confirm the conclusion rests on the full set of criteria the guidance names, not on one convenient fact, and that the alternative treatment is addressed and ruled out rather than waved away. Work the checklist below against your draft, and against the flawed memo, before you would ever call either one done.

Put numbers on the Riverbend facts to see why the two-criteria test is worth the care. The contract is fixed at 2,400,000 dollars and the build spans a period end, so the timing choice moves real revenue between quarters. Work the two conditions in order.

• Condition one, no alternative use. The line is engineered to Riverbend's plant and could not be readily redirected to another customer without significant rework, so Para 37 is satisfied on the facts as written. This is the easy half of the test.
• Condition two, enforceable right to payment. This is the half to evaluate rather than assume. The fact pattern states the contract entitles Meridian to payment for work completed to date plus a reasonable margin if Riverbend cancels, which is what Para 38 requires. A careful memo confirms that from the cancellation clause itself, not from the presence of milestone billings, because a billing schedule says little on its own about what is owed when a contract is torn up midway.

Both conditions hold, so Para 35 points to over-time recognition. Now the stakes. Suppose that by the period end Meridian has incurred 1,500,000 dollars of an estimated 2,000,000 dollars of total build cost. On a cost-to-cost measure the line is 75 percent complete, so over-time recognition records 0.75 times 2,400,000, or 1,800,000 dollars, of revenue this quarter. Point-in-time recognition would record zero until delivery and acceptance next quarter. The two-criteria test is not academic: it decides whether 1,800,000 dollars of revenue lands in this period or the next, which is exactly the kind of swing a reviewer and an audit committee will want to see grounded in Para 35 rather than assumed from a billing schedule.

The peer-review task hands you an AI-drafted memo on the same facts that reaches a fast over-time conclusion. A good redline catches four planted errors, each named with what is wrong, why, and what it should say.

• Error 1: wrong citation. The flawed memo cites Para 40 as the source of the over-time criteria. Para 40 is the point-in-time fallback; the over-time criteria are in Para 35. Grounding the over-time test in the point-in-time paragraph points the whole conclusion at the wrong rule, so the fix is to cite Para 35.
• Error 2: overstated single criterion. It claims that no alternative use, on its own, is sufficient for over-time recognition. Para 35 sets two conditions together: no alternative use AND an enforceable right to payment for work completed to date. One criterion alone does not support the conclusion, so the memo has to reach the right-to-payment condition as well.
• Error 3: assumed fact. It treats the milestone payments as equivalent to an enforceable right to payment that includes a reasonable margin on cancellation. Para 38 makes that a question to evaluate against the contract's cancellation terms, not a fact to assume from a billing schedule. The fix is to read the cancellation terms and support the conclusion, or flag it as unconfirmed.
• Error 4: omitted alternative. It states there is no other treatment to consider. The point-in-time alternative in Para 40 should be addressed and ruled out, not ignored, so a reader can see the position was reached by comparison rather than assertion.

The lab is clean by design: the guidance is short, the paragraph numbers are tidy stand-ins, and the errors are seeded. Real memos are messier, and three failure modes recur.

• The stale-version citation. Standards get amended, and a model trained before an amendment can cite a superseded paragraph that still reads plausibly, so a reviewer who does not trace the citation to the current codification risks filing an out-of-date rule.
• The legal-dependency stall. Whether an enforceable right to payment exists on cancellation usually turns on a lawyer's reading of the termination clause, so the accounting position stays provisional until counsel opines, and a memo that reads as fully settled can hide that the conclusion is waiting on an opinion outside the accountant's own desk.
• The confident fabrication. A model will sometimes invent a paragraph number or a criterion that is not in the standard but reads like boilerplate, and a reviewer skimming for fluency rather than tracing each citation back to the source tends to let it through.

That last failure is why the citation and logic checks stay in the workflow rather than being treated as optional polish; a memo can read as fluent and specific and still be wrong.

The place this gets tested hardest is the audit committee. When a memo like this reaches the committee, expect the questions to probe the soft joints: what happens to the conclusion if the cancellation terms do not give a right to payment with a margin, how sensitive is the position to the enforceability opinion, and what treatments were considered and rejected. A memo drafted for that room already answers them on the page: it leads with the conclusion, names the criterion that carries the most judgment, shows the point-in-time alternative ruled out, and lists the open points rather than hiding them. Preparing the Q&A is not a separate task; it is what a supported memo already contains.

Rate your own memo and redline against the rubric below. An honest rubric score shows which parts of the memo workflow hold up and which still need practice. Your scores roll up to the workflow maturity dashboard on the course hub.