Close week has a specific texture for every CPA and finance professional who lives through it: the numbers close, the variances stack up, and then comes the part that takes almost as long as the reconciliations โ converting all of that into prose that a managing partner, a CFO, or a board member can actually read and use.
The short version: Using Claude, CPAs and senior finance professionals can move from raw variance notes and line-item data to clean, well-structured, partner-ready financial commentary in a fraction of the drafting time โ while keeping their analytical judgment firmly in the driver's seat. The Leverage Years teaches this system specifically for finance professionals who want to spend close week on analysis, not on staring at a blank comment box. The workflow does not replace your understanding of why the numbers moved; it converts that understanding into clear, professional communication faster.
The constraint worth stating upfront: Claude produces prose, not analysis. The variance narrative that emerges is only as good as the analytical notes you feed it. The system works because you do the hard thinking first โ Claude handles the conversion from thinking to writing. That division of labor is not a limitation; it is the correct model for professional financial communication.
Who this is for
- Senior accounting and finance professionals (CPAs, controllers, finance managers, VP Finance) who personally write or heavily edit financial commentary during the close cycle.
- Public accounting professionals drafting management discussion, period-end memos, or client deliverables under time pressure.
- Finance directors and CFOs who review close-week packages and want a consistent communication standard from their teams.
- Professionals in mid-size and enterprise environments who produce monthly, quarterly, or period-end reporting for internal or external audiences.
This is not for you if: you are looking for AI to perform variance analysis, read your GL, or explain what caused a movement without your input. Claude does not have access to your systems and cannot independently analyze financial data. It converts your analysis into polished communication โ nothing more.
The System: Six Steps From Raw Notes to Finished Commentary
The close cycle has a natural rhythm. This system fits inside that rhythm without adding a new phase โ it replaces the drafting phase with something faster and more consistent.
Step 1: Complete your variance analysis first
The system starts after you have done the work. You have your trial balance, your prior period comparatives, your budget-to-actual reports, and your knowledge of the business โ you know why gross margin contracted, why SG&A ran above plan, why the EBITDA bridge tells a different story this quarter than last. You have notes. They may be in a spreadsheet comment, a working-paper annotation, a bullet list in a notepad, or your own shorthand.
The notes do not need to be polished. They need to be accurate and complete for the variance narrative you are going to produce. If you don't understand a variance yet, do not proceed to the Claude step โ the system does not substitute for understanding, it converts understanding into communication.
Step 2: Structure your notes before prompting
Raw notes benefit from a light structure pass before going to Claude. This is not a full write-up โ it takes two to three minutes. The structure you want:
- What the line item or category was (Revenue, COGS, specific SG&A line, EBITDA, working capital, etc.)
- The movement (favorable/unfavorable, dollar amount, percentage โ at the level of precision your audience needs)
- The primary driver (the business reason you have identified: volume, price, timing, one-time item, category shift, etc.)
- Any qualifications (unusual item, reclass, offset in another line, expected to normalize in next period, etc.)
Even in rough bullet form, this structure gives Claude what it needs to produce accurate commentary. Absent this structure, Claude will produce generic or vague prose โ readable but not specific to your numbers or your period.
Step 3: Choose the audience and format for the output
Different audiences require different commentary styles. Before you prompt, decide:
- Who will read this? A managing partner who wants the business story in plain English? A board audit committee that wants technical precision and full disclosure? An operations team that needs cause-and-effect without accounting jargon?
- What format? A bulleted variance table with brief commentary? A narrative MD&A section? A standing section of a board deck? A memo for a client review meeting?
- What length? One sentence per line item? One paragraph per section? A full narrative?
These decisions go into the prompt, not into a post-hoc editing session.
Step 4: Build the role + task prompt
The same principle that applies to legal drafting applies to financial commentary: the more specifically you describe the role and the output, the closer the first draft lands.
A useful role prompt for financial commentary:
You are a senior finance professional with deep experience writing management commentary for period-end reporting. You write in a clear, direct, professional register โ no padding, no hedging without cause, no vague language like "challenging environment" or "headwinds." You explain financial movements in plain business language, state the driver first, and qualify where relevant. You do not invent numbers or create explanations for variances I have not described. If my notes are ambiguous, you note the ambiguity rather than guessing.
Then the task prompt:
Using the variance notes below, draft the [monthly/quarterly] management commentary for the [period] close. The audience is [managing partners / CFO and board / client management]. Format: [bullets with brief commentary per major line / narrative paragraphs by category / table with analysis column]. Tone: [professional and direct / accessible for non-finance readers / technically precise]. Keep the total length [under X words / under X lines per category]. Here are my notes: [paste structured notes]
Specify the audience. Specify the format. Specify the tone. Specify the length. Every dimension you leave open is a dimension where Claude will make a choice you may need to edit back.
Step 5: Review the first output for accuracy, not just style
When Claude returns the draft commentary, your review has a specific priority order. Address these in sequence:
Accuracy first. Does every figure mentioned match your source? Does every stated driver accurately reflect your analysis? Does any sentence imply a cause-and-effect relationship that isn't supported by your notes? Claude will not fabricate numbers if your prompt supplies them โ but it will occasionally rephrase a driver in a way that subtly changes the meaning. Catch this before anything leaves your desk.
Completeness second. Are all material variances addressed? Is anything in your notes that should appear in the commentary absent from the draft?
Tone and audience fit third. Does this read at the right level for the audience you specified? Partner-ready commentary for a law firm management meeting reads differently than a board-level MD&A, which reads differently than an operations review brief.
Precision of language fourth. Finance has a vocabulary that matters: "favorable," "unfavorable," "variance," "deviation from plan," "prior period," "year-over-year," "non-recurring." Confirm the draft uses these terms correctly and consistently.
Step 6: Finalize, verify, and publish
Run your standard numerical verification pass: any figure that appears in the commentary should tie to a source document. This is not new โ this is what you always do. The draft coming from Claude does not change the tie-out requirement; it just means the prose is ready faster.
Once verified, the commentary is yours. It reflects your analysis, your period, your numbers. Claude produced the sentence structure; you supplied the substance.
Example: A Simplified Variance Commentary Pass
Suppose a controller has the following notes after a monthly close:
Revenue โ favorable $280K vs. plan. Volume-driven, primarily from two large orders that pulled forward from next month. Pricing in line with plan. Not expected to recur at this level next period.
COGS โ unfavorable $195K vs. plan. Mix shift toward lower-margin product line. Labor stable. Material costs slightly elevated on one SKU due to supplier pricing โ expect to normalize next quarter.
SG&A โ unfavorable $110K vs. plan. Two items: (1) annual software renewals accelerated into this period from next โ $75K, will not recur. (2) Travel above plan by $35K โ Q3 client activity, expected.
EBITDA โ favorable $85K net of the above.
Using the structured notes above and the prompt approach in Step 4, Claude would return commentary along these lines:
Revenue closed favorable to plan by $280K, driven by higher volume from two orders that accelerated from the following period. Pricing was in line with plan. The pull-forward is not expected to repeat at this level; next-period revenue should return to the planned run rate.
Cost of Goods Sold ran $195K unfavorable to plan, primarily reflecting a mix shift toward the lower-margin product line. Labor costs were on plan; material costs on one SKU were modestly elevated due to supplier pricing adjustments and are expected to normalize in Q4.
SG&A was unfavorable by $110K to plan, attributable to two items: annual software contract renewals accelerated from the subsequent period ($75K, non-recurring), and travel expense above plan by $35K due to client-facing activity in the quarter, consistent with Q3 seasonality.
EBITDA was favorable to plan by $85K on a net basis, with revenue upside partially offset by the COGS mix and SG&A timing items described above.
The reviewer's job: confirm every figure, verify each causal statement is accurate, adjust the tone for the specific audience, add any context the notes didn't include. That review takes four to eight minutes for a section like this. Starting from a blank page, this same section typically takes twenty to thirty minutes.
Checklist: Pre-Prompt Quality Pass
Before sending variance notes to Claude, confirm each:
- All material variances are included in the notes (nothing major is missing)
- Each variance has an identified driver โ not just the direction and magnitude
- Any non-recurring, one-time, or reclassified items are clearly flagged
- Dollar amounts and percentage movements are present at the level of precision the audience expects
- Prior-period or budget reference is clear (vs. plan, vs. prior year, vs. prior month)
- Notes do not include client names, employee names, or proprietary identifiers that should not appear in the commentary
- Format, audience, and length are decided before building the prompt
What Claude handles well in close-week work โ and what it doesn't
Use Claude for:
- Converting accurate, structured variance notes into polished narrative or MD&A prose
- Producing multiple versions of the same commentary at different technical levels (board-level vs. operations review vs. full audit committee)
- Drafting the framing sections of a management report (executive summary, period overview, key themes)
- Producing consistent formatting across a multi-section report when your own notes are inconsistent in style
- Writing the transition sentences between sections that make a report read as a coherent document rather than a stack of separate analyses
- Drafting cover memos for client financial deliverables, where the tone needs to be professional and relationship-appropriate
Never use Claude for:
- Performing the variance analysis itself โ Claude cannot read your GL or your reports
- Creating explanations for variances you have not analyzed and understood
- Any period where numerical accuracy has not been independently verified
- Generating forward-looking statements or projections without your explicit instruction and oversight
- Producing commentary for external regulatory filings without full professional review and signoff
The second list is not a failure of the tool. It is the correct division of labor: your analysis goes in, professional communication comes out.
Pitfalls to avoid on your first close cycle
Pitfall one: Sending undifferentiated data instead of analytical notes. If you paste a raw trial balance or a variance table with no explanatory notes, Claude will produce generic, data-narrative prose ("Revenue increased by $X compared to plan") without the business drivers. That is a starting point for your analysis, not a replacement for it. The system requires your analysis first.
Pitfall two: Skipping the audience and format specification. Generic output follows generic instructions. If you don't tell Claude the audience is a private equity board requiring precise technical language and full disclosure, it will default to a readable but generic middle register. A two-sentence audience description in the prompt costs thirty seconds and makes a material difference to the output.
Pitfall three: Treating the first draft as final. The draft is a very good starting point. It is not reviewed work product. Every figure needs to tie; every causal statement needs to be verified against your notes. The review is the professional work โ it is where your judgment and your signoff live.
Pitfall four: Over-prompting for perfection on the first pass. If the first output needs structural correction, send a follow-up prompt correcting the specific structure before doing substantive edits. If it needs a tone adjustment, ask for that specifically. Building iteratively is faster than trying to write a perfect one-shot prompt.
From one-time use to a repeatable operating system
The single-close use of this system is useful. The compounding value comes from building a library of close-cycle prompt templates.
After your first successful use, save the role prompt and the task structure in a working document. Label it specifically: "Monthly management commentary โ non-technical audience," "Quarterly board MD&A โ audit committee format," "Client period-end memo โ solo practitioner." Strip any period-specific content; save only the structural elements.
By your third or fourth close cycle using documented templates, the prompt setup time drops to under sixty seconds per section. The consistency across periods improves โ which matters for audiences who are comparing this quarter's report to last quarter's.
You also start to notice patterns in what your notes need to include to produce useful commentary. The discipline of preparing for the Claude step makes your own analytical notes more structured, which has benefits independent of the drafting workflow: better working papers, cleaner review notes, faster internal review cycles.
This is what a close-cycle operating system looks like when it is actually installed rather than improvised each month.
If you want the complete version of this system โ the full prompt vault organized by section type, the multi-audience adaptation framework, the review protocol, and how to extend it to client deliverables and board reporting โ The Leveraged CPA and Finance Professional covers it entirely, built for senior finance professionals who are doing real close-cycle work.
Not sure which course fits your work? Take the 6-question course selector.
You can also browse the full range of courses at The Leverage Years courses page.
Frequently asked questions
Will Claude fabricate numbers in the variance commentary?
Not if your prompt supplies the numbers explicitly. Claude does not have access to your financial systems; it only works with what you include in the prompt. If your structured notes contain the figures, Claude uses those figures. Where your notes are ambiguous, a well-constructed role prompt instructs Claude to flag the ambiguity rather than fill it in. The risk of numerical fabrication is near zero in a well-structured workflow โ but your tie-out review remains mandatory regardless, just as it would be for any draft prepared by a staff member.
How do you handle periods with complex or unusual items?
More explanation in your notes, not less. For a one-time item, a reclass, a significant estimate revision, or an acquisition effect, write out the explanation in your notes the way you would explain it to a colleague โ what it is, why it happened, what it does and does not signal about the underlying business. Claude will incorporate that explanation accurately. Unusual items require more note-level detail, which results in more accurate commentary output.
Can this workflow support external reporting, not just internal management commentary?
With a higher review standard, yes. The prompt and note quality that works for an internal management report works as a starting point for MD&A sections in external reporting. The difference is the review intensity: anything that goes into an external filing or audited document requires full professional signoff, numerical tie-out to source, and compliance review as appropriate. Claude's draft compresses the first-draft phase; it does not substitute for the external reporting review process.
Does this workflow work for tax commentary and client memos, not just management reporting?
Yes, with appropriate adaptations to the role prompt and task structure. The system is designed for period-end variance commentary, but the underlying approach โ role prompt, structured notes, audience/format specification, accuracy review โ applies to any professional financial communication. Tax client memos, planning summaries, and post-filing communications all follow the same pattern. The prompt vault approach described in the article applies: build a template for each document type you produce repeatedly.
How does the review step work on a compressed close timeline?
The review step for this workflow takes four to ten minutes for a typical management commentary section โ significantly less than the blank-page drafting time it replaces. On a compressed timeline, the net effect is more time available, not less. If the close timeline is so compressed that a ten-minute review is not feasible, the problem is the close timeline, not the workflow. The review step is the professional standard that makes the output reliable โ it is not optional regardless of time pressure.
What happens when the numbers are still uncertain at the time commentary is needed?
Draft with what is confirmed; flag explicitly what is preliminary. Claude can produce commentary that states "subject to final close" or "pending resolution of [specific item]" if you instruct it to do so in the prompt. Provisional commentary with appropriate qualifiers is better than delayed commentary. When final numbers are available, a targeted prompt revision produces a final version quickly from the working draft.