For any CFO, controller, or financial analyst, the due diligence phase of a merger or acquisition is a pressure cooker. You are typically working with a data room containing thousands of unorganized files—scanned PDFs of historical tax returns, multi-tab Excel ledgers, board decks, customer concentration reports, and employment contracts. The expectation is that you will synthesize this chaos into a clear, risk-adjusted view of the target company within a matter of days, not weeks. The pain is real: manual data extraction consumes 60% of your time, leaving only a sliver for actual analysis. Worse, the pressure to identify material liabilities—off-balance-sheet items, revenue recognition irregularities, or customer churn risks—means that a single overlooked footnote can become a board-level catastrophe post-acquisition.
Claude Cowork, Anthropic’s collaborative AI workspace, changes this dynamic fundamentally. Instead of using a chatbot to ask generic questions, you build a persistent “coworker” that has read your entire data room, your internal accounting policies, and your past deal memos. It doesn’t just summarize; it cross-references, flags anomalies, and drafts sections of your Quality of Earnings (QoE) report in a format your investment committee expects. The key shift is from “prompt engineering” to “workflow design.” You are not asking for a one-off answer; you are delegating a structured analytical task with specific success criteria. This post provides two production-ready prompts you can adapt for your next deal, focusing on the anatomy of a prompt that yields CFO-grade output, not generic fluff.
Why Standard Chat Prompts Fail in M&A
The default approach—”Read this balance sheet and tell me if it looks good”—produces superficial output because it lacks context. The AI doesn’t know your firm’s risk appetite, the specific deal thesis (e.g., cost synergies vs. growth platform), or the accounting standards (GAAP vs. IFRS) applied. Furthermore, it doesn’t know what “good” looks like for your firm. A prompt that works for a blog post fails for a QoE memo. The solution is to provide a “success brief” and a “reference standard” inside the prompt itself. This forces Claude to operate like a first-year associate who has been fully briefed on your firm’s expectations, rather than a generic assistant.
Below is the first prompt template. This one is designed for the critical first pass: analyzing the target’s historical financial statements to isolate non-recurring items and normalize EBITDA. This is the foundation of any valuation discussion.
First, read these files completely before responding:
[target_financials.xlsx] — This contains the raw P&L, Balance Sheet, and Cash Flow statements for FY2023, FY2024, and FY2025.
[management_adjustments.xlsx] — This is my firm’s working file where I have manually flagged potential add-backs and deductions.
[prior_deal_memo.pdf] — This is a redacted example of a QoE section from a deal we closed last year in the same sector.
Here is a reference for what I want to achieve:
The goal is to replicate the structure and depth of the “Normalized EBITDA” table in the prior_deal_memo.pdf. Specifically, I need a table that lists each adjustment, the dollar amount, the fiscal year affected, and a one-sentence justification that would hold up during a vendor’s Q&A.
Here’s what makes this reference work:
The prior memo separates “Recurring vs. Non-Recurring” items clearly. It does not simply add back all one-time costs; it applies a “look-back” test to see if the expense occurred in 2 of the 3 years. If it did, it is considered recurring. It also explicitly excludes any add-backs related to owner’s discretionary expenses unless they exceed 3% of revenue.
The tone is neutral and factual, avoiding aggressive negotiation language. It uses specific account names (e.g., “Consulting fees – M&A advisory” not “miscellaneous”).
Here’s what I need for my version / SUCCESS BRIEF:
Type of output + length: A markdown table followed by a 3-paragraph narrative summary. The table must have columns: Line Item, FY2023, FY2024, FY2025, Adjustment Type, Rationale.
Recipient’s reaction: The IC should be able to trust the number without re-doing my work. They should see that I have already stress-tested the “recurring” classification.
Does NOT sound like: A generic list of “one-time expenses.” It must not sound like I am inflating EBITDA to make the deal look cheaper.
Success means: The final adjusted EBITDA figure is within 2% of the figure calculated by the seller’s independent advisor, or I have a documented reason for the variance.
My context file contains my standards, constraints, audience. Read it fully before starting.
DO NOT start executing yet. Ask clarifying questions first.
Give me your execution plan (5 steps max) before you begin.
This prompt works because it forces the AI to adopt a specific analytical framework (the “look-back” test) and a specific output format. Notice the explicit “Does NOT sound like” clause—this prevents the AI from defaulting to aggressive buyer-side tactics that would poison the negotiation. The “Success means” metric (within 2% of the seller’s number) gives Claude a target to optimize for, not just a task to complete. When you paste this into Claude Cowork, ensure you have uploaded the files and the context file (which should contain your firm’s specific materiality thresholds and accounting policy preferences).
Moving Beyond the Numbers: Qualitative Risk Flagging
Once the quantitative normalization is done, the next bottleneck is qualitative risk assessment. This involves reading the MD&A sections of the target’s financials, the customer contracts, and the employee agreements to spot hidden risks like contract renewal cliffs or key-person dependency. This is where most analysts spend sleepless nights, manually highlighting PDFs. The second prompt below addresses this by turning Claude into a risk extraction engine that maps qualitative findings directly to your deal risk register.
First, read these files completely before responding:
[top_10_customer_contracts.pdf] — These are the signed agreements with customers representing >5% of revenue each.
[management_interviews.md] — This is my transcribed summary of the CEO and CFO interviews regarding future growth and retention.
[risk_register_template.xlsx] — This is my firm’s standard risk register format with columns for Risk Category, Likelihood, Impact, Mitigation, and Owner.
Here is a reference for what I want to achieve:
I need a completed risk register that mirrors the granularity of the template but adds a “Contractual Clause Evidence” column. The goal is to link each risk to a specific clause or quote from the contract or interview, not just a general concern.
Here’s what makes this reference work:
The best risk registers we have produced use a “Probability x Impact” score (1-5 each) to prioritize. They also distinguish between “Mitigable” (we can fix with a new contract) and “Structural” (we cannot fix, must price into the deal). The language is precise, citing “Section 4.2 (Change of Control)” rather than “the contract says something about ownership.”
Here’s what I need for my version / SUCCESS BRIEF:
Type of output + length: A table with 15-20 rows, sorted by the Probability x Impact score (highest first). Include a separate column for “Recommended Next Step” (e.g., “Request estoppel letter” or “Negotiate cap on liability”).
Recipient’s reaction: The deal lead should feel they have a complete map of the landmines before they walk into the negotiation room. They should not be surprised by any risk.
Does NOT sound like: A generic list of risks like “customer concentration.” It must be specific to the contracts I uploaded, with direct quotes.
Success means: Every risk identified can be traced back to a specific line in a contract or a specific quote from the interviews. No “gut feel” risks are included.
My context file contains my standards, constraints, audience. Read it fully before starting.
DO NOT start executing yet. Ask clarifying questions first.
Give me your execution plan (5 steps max) before you begin.
The critical difference in this second prompt is the demand for “Contractual Clause Evidence.” This forces Claude to ground its output in the actual source material, reducing hallucination risk. By asking for a “Recommended Next Step,” you are converting the AI from an analyst into a junior deal strategist. This is where Claude Cowork excels—it can hold the entire context of a 200-page contract in memory and cross-reference it with interview notes, something a human would take a full day to do accurately.
My practical tip for your first deployment is to start with a “dry run” on a closed deal. Take a past acquisition where you already know the final risks and the normalized EBITDA. Run these prompts against that old data. Compare Claude’s output to the actual work product from that deal. You will likely find that Claude catches a few things your team missed, and you will also find areas where you need to tighten the “Success Criteria” in the prompt. This calibration step is essential before you use it on a live, time-sensitive deal. Do not skip it.
Next, try modifying the second prompt to focus on “Retention Risk” only. Upload the employment contracts of the top 10 earners and ask Claude to identify non-compete clauses and change-of-control triggers. This is a high-value, low-volume test that will quickly show you the ROI of this workflow.
Published on 17 August 2026 on growwithgpt.com