Prompt Chain: From Raw Variance Data to Complete Management Report Narrative

What This Builds

A structured multi-step prompt chain that takes your raw variance data through four sequential AI steps (analysis, commentary, executive summary, Q&A prep) and produces a complete first-draft management report narrative in under 15 minutes. Instead of using AI for one task at a time, you connect the outputs of each step into the inputs of the next, creating a narrative pipeline that mirrors how a skilled analyst thinks through a management report.

Prerequisites

• Claude Pro account ($20/month, longer context needed for chaining)
• Your Monthly Close Commentary Assistant Claude Project set up (Level 3 guide)
• One full month of close data to use as the test case
• 1 hour to set up and test the chain

The Concept

A prompt chain is like an assembly line. Instead of asking one worker to build an entire car from scratch, you break it into stations: first the frame, then the engine, then the interior. Each station uses the output of the previous one.

For management reporting, the chain looks like this:

1. Station 1 (Analysis): AI reads raw variance data → identifies the top 5 most significant variances and their probable drivers
2. Station 2 (Commentary): AI uses the analysis output → writes line-by-line commentary for the full P&L
3. Station 3 (Summary): AI uses the commentary → writes the CFO executive summary
4. Station 4 (Q&A Prep): AI uses the summary → generates anticipated leadership questions with draft answers

Each step builds on the previous. The commentary is better because the analysis happened first. The executive summary is better because it's derived from coherent commentary rather than raw numbers.

Build It Step by Step

Part 1: Set Up the Four-Step Chain in Your Claude Project

Open your Monthly Close Commentary Assistant Claude Project (or create a new one called "Management Report Narrative Chain").

Add these instructions to the project if they aren't already there:

You are my financial reporting assistant. I will run a 4-step management report narrative chain with you each month. When I say "Step 1," "Step 2," etc., follow the specific instructions for that step.

STEP 1 — VARIANCE ANALYSIS:
When I provide a variance table and say "Step 1," your job is to:
- Identify the 5 largest variances by absolute dollar amount
- For each, suggest the most likely business driver (based on the numbers, not business context I haven't given you)
- Flag any variances that look like they could be data errors (extreme outliers, opposite sign from usual)
- Output: A numbered list of the top 5 variances with probable drivers. Do NOT write commentary yet.

STEP 2 — LINE-ITEM COMMENTARY:
When I say "Step 2" and provide or confirm business context, your job is to:
- Write a variance comment for every line item in the table using the standard format
- Incorporate any context I give you about one-time items or specific drivers
- Flag any lines where you're uncertain about the driver with [REVIEW]
- Output: A complete commentary file with one comment per line item

STEP 3 — EXECUTIVE SUMMARY:
When I say "Step 3," your job is to:
- Read all the commentary from Step 2
- Synthesize it into a 3-paragraph CFO executive summary (under 250 words)
- Lead with the headline result, follow with the 3 most significant drivers, close with any forward-looking context I've provided
- Output: A formatted executive summary

STEP 4 — Q&A PREP:
When I say "Step 4," your job is to:
- Based on the executive summary from Step 3 and any underperformance vs. budget
- Generate the 5 most likely challenging questions leadership will ask
- Draft a concise answer to each (under 50 words)
- Highlight which answers need my confirmation before the meeting
- Output: A numbered Q&A prep sheet

Save the project instructions.

Part 2: Prepare Your Chain Input Document

Before running the chain each month, prepare one document that will travel through all four steps. In a plain text file or Word document, organize:

CHAIN INPUT — [Month/Year]

VARIANCE TABLE:
[Paste your actuals vs. budget variance table — all rows]

ONE-TIME ITEMS / CONTEXT:
- [Line item]: [explanation of one-time driver]
- [Line item]: [explanation]
[Leave blank if none]

FORWARD-LOOKING CONTEXT:
- [Any context about Q2/remainder of year that should appear in the executive summary]
[e.g., "The 3 delayed West deals are expected to close in Q2 — $2.1M"]

Having this prepared before you start means the chain runs smoothly without interruption.

Part 3: Run Step 1 — Variance Analysis

1. Open a new chat in your Claude Project
2. Type: "Step 1. Here is the variance table for [Month]:"
3. Paste your variance table
4. Send

What you should see: A numbered list of the 5 largest variances with Claude's assessment of probable drivers. It may flag one or two lines as needing your attention.

1. Review Claude's analysis. Note any corrections:

   • "The #3 item (IT Infrastructure +$350K) is favorable because a planned server migration slipped to next quarter. It's a timing item, not a structural reduction."
   • "The #5 item appears to be a data error. I'll check with accounting."

This review takes 5–10 minutes and significantly improves Step 2.

Part 4: Run Step 2 — Line-Item Commentary

1. In the same Claude chat, type:

Step 2. Here is the context to incorporate:

One-time items:
- IT Infrastructure variance: Favorable timing — server migration shifted to next quarter. Comment should note this.
- [Any others]

[Paste your full variance table again if it's a long chain — Claude's context window can handle it]

1. Send

What you should see: A complete commentary file with one comment per line item. Items Claude is uncertain about are marked [REVIEW].

1. Review the [REVIEW] flags and correct them:

   • "The [Line Item] variance is unfavorable due to [specific context]. Update the comment."
   • Claude will revise the specific line instantly

2. When you're satisfied, copy the full commentary to your management report document.

Part 5: Run Step 3 — Executive Summary

1. In the same Claude chat, type:

Step 3. Forward-looking context: [paste any Q2/full-year outlook context].

1. Send

What you should see: A 3-paragraph executive summary that reads as a coherent narrative, not a list of disconnected facts. Because it was derived from the Step 2 commentary, the themes and language are consistent.

1. Review and edit for any organization-specific language or framing your CFO prefers.

Part 6: Run Step 4 — Q&A Prep

1. In the same chat, type: "Step 4."
2. Send

What you should see: A Q&A prep sheet with 5 anticipated questions and draft answers. Typically 2–3 will be ready to use; 1–2 will be marked as needing your confirmation.

1. Review each answer. Mark the ones you need to verify before the meeting.

Part 7: Assemble the Output

You now have four documents:

1. Variance analysis (your reference for the session)
2. Line-item commentary (paste into management report)
3. Executive summary (paste into slide 1 of the deck)
4. Q&A prep sheet (print and bring to the CFO meeting)

Total elapsed time from starting Step 1 to having all four: 12–20 minutes. Editing time: 20–30 minutes. Total close narrative work: under 1 hour.

Real Example: The February Close Chain

Input: Alex, a senior financial analyst at a $200M retailer, runs the chain at 9am on day 4 of the February close.

Step 1 output: Claude identifies that the top variance is a $1.2M favorable swing in Cost of Goods and flags it as unusual ("favorable COGS typically indicates lower volume or mix shift, confirm this is accurate"). Alex checks and confirms: a supply chain optimization drove a genuine cost reduction.

Step 2 output: 22 line-item comments, all under 50 words, in the correct format. 2 flagged as [REVIEW]. Alex corrects them in 5 minutes.

Step 3 output: A 210-word executive summary that opens with the headline (Revenue in line, EBITDA $0.8M favorable), covers the COGS story and a timing-related SG&A favorable, and closes with a full-year outlook note.

Step 4 output: 5 anticipated questions, including "Is the COGS improvement sustainable or one-time?" Alex confirms it's sustainable and adds supporting detail to the answer.

Total time: 18 minutes for the chain, 25 minutes editing. Alex presents to the CFO at 11am with everything ready.

What to Do When It Breaks

• "Step 2 commentary doesn't reflect the context I added in Step 1" → Repeat the context explicitly in Step 2: "Use this context in Step 2: [paste again]." Don't assume Claude remembers context from a previous turn if you skipped it.
• "The executive summary is too long" → Add to your instructions: "Step 3 output must be under 230 words. Count them." Claude will respect this constraint.
• "The Q&A prep questions are too generic" → Add to Step 4: "Focus your questions on the specific areas where we underperformed budget, not generic CFO questions." This makes the prep more targeted.
• "Claude loses context between steps in a long chain" → Paste the Step 2 commentary at the top of your Step 3 message. For very large reports (40+ line items), this is necessary.

Variations

• Simpler version: Run just Steps 2 and 3 (commentary and executive summary). Skip the analysis and Q&A steps if your close cycle is tight. You still save 90+ minutes.
• Extended version: Add a Step 5, "Board Narrative," that converts the executive summary into board-level language with less financial jargon and more strategic framing.

What to Do Next

• This week: Run the chain with last month's data as a test and compare the output quality to what you would have written
• This month: Use it for the first time on live close data; track how much editing you need
• Advanced: Combine this chain with the Power Automate automation (Level 4 guide) so Steps 1–4 run automatically when your variance file is saved

Advanced guide for Financial Analyst (Corporate) professionals. Claude Pro ($20/month) is required for the long-context conversations in this chain. Do not paste confidential unreleased financial data into AI tools without checking your company's data governance policy.