# The Advisory Letter Generator

A four-step prompt chain that turns a client's financials into a client-ready advisory letter — with verification built in.

This is the working core of advisory work: read the numbers, find what matters, explain it in plain English. AI drafts the narrative. You own every conclusion. Plan 45–60 minutes the first time, 20 minutes once you have the rhythm.

**Confidentiality first.** Client financial data is confidential under AICPA Code Rule 1.700.001. Before you paste anything into an AI tool, do one of two things: anonymize the data (Step 1 shows how), or confirm the tool is approved under your firm's confidentiality policy. If any input comes from a tax return, stop — tax return information carries use and disclosure restrictions under IRC Sec. 7216. Get written consent or keep that data out of the prompt.

---

## Step 1 — Prepare the inputs

Pull these from your accounting system after the close. Do not analyze an unreconciled GL — the letter will inherit every error in it.

**From the P&L:** current period and the same period prior year. Include revenue by line if the chart of accounts supports it, COGS, gross profit, payroll, rent, and total operating expenses.

**From the balance sheet:** current period end and prior period end. Cash, AR, inventory, AP, credit cards, loans, and equity.

**Your notes:** anything you already know that the numbers won't say. A one-time equipment sale. A new loan. An owner draw spike. Two or three lines is enough.

**Anonymize before pasting:**

- Replace the client name with "the Company." Strip the EIN, addresses, and all customer, vendor, and employee names.
- If you want another layer, multiply every dollar figure by a constant (say 1.37). Ratios and trends survive scaling. Note the factor in your workpapers so you can convert back.
- Confirm the basis: cash or accrual. State it in the prompt. It changes how AR and AP findings read.

## Step 2 — The analysis prompt

Paste your prepared data where marked.

```
You are helping an accountant analyze a small business client's
financials. The statements are [cash/accrual] basis, prepared
internally, post-close.

[PASTE P&L, BALANCE SHEET, AND PRIOR-PERIOD COMPARATIVES HERE]
[PASTE YOUR CONTEXT NOTES HERE]

Do three things:

1. Compute these ratios for current and prior period: gross
   margin, net margin, current ratio, AR days, AP days, and
   debt-to-equity. Show the formula and the exact input figures
   for each one.
2. Describe the trends: what moved, by how much, in which
   direction.
3. Give me the three findings that matter most to this business
   owner, ranked by importance. For each: what the numbers show,
   why it matters, and one question I should ask the owner.

Rules: Use only the data I provided. Do not cite industry
benchmarks unless I supply them. If a calculation requires data
I did not give you, say so instead of estimating.
```

Read the output before moving on. If a ratio looks wrong, it probably is — AI does arithmetic unreliably. You will recompute everything in Step 4, but catch obvious misses now.

## Step 3 — The narrative prompt

In the same conversation:

```
Draft a client advisory letter from the three findings above.

Audience: the business owner. No accounting background. Define
any term they would not know.

Structure:
- Opening: the period covered and the basis of preparation.
- The three findings, in plain English, one short section each.
- Recommended next steps: specific, dated where possible.
- A caveats block that states: these figures come from
  internally prepared statements and are not audited or
  reviewed; this letter is for discussion, not assurance; and
  the analysis reflects the periods shown, not a forecast.

Tone: plain and direct. No filler. Under 600 words.
```

The caveats block is not boilerplate to soften the letter. It is the honest boundary of what internal statements can support. Keep it.

## Step 4 — The verification pass

The draft is raw material. This step makes it yours. Work the checklist in order:

- [ ] **Recompute every figure** in your workpapers or a spreadsheet. Every ratio, every percentage, every dollar amount. No AI-computed number reaches a client unrecomputed.
- [ ] **Trace each claim to the actuals.** "Margin fell because payroll grew faster than revenue" — open the GL and confirm it.
- [ ] **Test the findings against your judgment.** You know this client. If the AI ranked the wrong thing first, reorder or replace it. The letter carries your name, not the model's.
- [ ] **Hunt for invented benchmarks.** Any line like "healthy businesses keep AR under 30 days" that you did not supply gets cut or sourced.
- [ ] **Verify any compliance statement** against the authoritative source — IRC, IRS publications, FASB ASC, or your state board rules — before it ships.
- [ ] **Restore the real numbers and the client name.** If you scaled in Step 1, convert back and recheck.
- [ ] **Read it aloud once.** If a sentence sounds like something you would never say to this client, rewrite it.

## Troubleshooting

**The AI sounds certain when it shouldn't be.** Reply: "Restate the findings with your confidence in each, and list every assumption you made." Cut anything it cannot tie to the data you provided.

**The AI cites a benchmark you never gave it.** That breaks the Step 2 rules. Reply: "Remove every benchmark not present in my data." If you want benchmarks in the letter, pull them yourself from a source you trust and supply them in the prompt.

**The findings feel generic.** The model is missing context only you have. Add two or three lines about the client's situation to your Step 1 notes and rerun Step 2.

**The numbers don't tie to your statements.** The problem is upstream. Recheck your close and reconciliations, fix the source, and paste again. The chain cannot rescue bad inputs.

---

Part of The Accountant & Bookkeeper Advisory Library — vetslearn.com/libraries/accountant