Monthly Client Commentary With AI: A Draft-and-Review Workflow for Bookkeepers

Most small business clients get a profit and loss statement (P&L) and a balance sheet every month. Fewer get a plain-English note explaining what changed and why it matters. That note is often the first thing to go when month-end runs long, and it is the part of the package clients notice most.

It is also something the ledgers don't do well yet. A vendor review of AI for bookkeepers lists "explaining what a client's numbers mean" as something AI does not replace (SuperDupr, Aug 2026, vendor), and Intuit lists report generation in the Australian edition of its Accountant Suite as "coming soon" (Intuit Australia, May 2026, vendor). Meanwhile, practitioners are discussing clients who now expect a faster, fuller close (beancount.io forum).

Here is a draft-and-review workflow any small firm can run with a business AI account and the reports you already export.

The rule that makes this safe

AI drafts; your bookkeeper decides. The AI never guesses why a number moved. It describes what moved, and where the cause isn't clear from the data, it writes a question for the client. A person edits and approves every note before it is attached.

Step 1: Gather the inputs

For each client, after the books are closed:

• P&L for this month, last month and the same month last year.
• Balance sheet for this month and last month.
• Budget, if the client has one.
• A short client notes file: what the business does, known seasonal patterns, anything the client told you this month ("we hired two people," "we prepaid insurance").

Export from QuickBooks Online or Xero as a spreadsheet file (CSV or Excel). Keep the files in your firm's own storage.

Step 2: Set thresholds per client tier

Not every movement deserves a sentence. Flag a line only when it moves past both a dollar amount and a percentage. Example tiers (set your own):

Using both tests stops tiny accounts with big percentage swings, and big accounts with tiny percentage swings, from cluttering the note.

You can compute the flags in a spreadsheet before the AI sees anything. That keeps the math in a tool you trust and gives the AI a short, clean list to write about.

Step 3: Use a structured prompt

A prompt that works well has four parts: who the reader is, the facts, the rules, and the format. Example:

You are drafting a monthly note from a bookkeeping firm to a small business owner who is not an accountant.

Facts: [paste the flagged lines with this month, last month, same month last year] and [paste the client notes file].

Rules: Describe only what the numbers show. Do not guess causes. If a cause is in the client notes, you may state it. If not, write a short question for the client. Use plain words, no accounting jargon. Round to the nearest hundred dollars.

Format: Three to six short bullets on what changed, most important first. Then "Questions for you" with up to three questions. Under 200 words.

Run it in a business AI account your firm has approved for client data, not a personal chat account. Accounting technologist Jason Staats has said firms need paid ChatGPT Business for its data protections (CFOtech US, Jan 2026); the same logic applies to other providers' business plans.

Step 4: Review with a checklist

The bookkeeper reads the draft against the reports and checks:

1. Every number matches the report.
2. No cause is stated that isn't in the data or the client notes.
3. Nothing is missing that the client would expect (a big one-off, a loan payment, a payroll change).
4. Questions are ones you actually need answered.
5. Tone fits the client.

Then edit, approve and attach it to the monthly package.

What a good draft looks like

Here is a made-up example for a sample landscaping company (not a real client):

March at a glance

• Revenue was $48,200, up $9,100 on February and $4,300 on March last year.
• Fuel was $3,900, about $1,200 more than last March.
• Payroll rose $6,400 from February. Your notes mention two seasonal hires starting March 1.
• Cash ended at $21,500, down $7,800, mainly from the equipment loan payment.

Questions for you

• The $2,400 payment to Valley Equipment on March 14 is coded to repairs. Was this a repair, or new equipment we should record as an asset?

Short, factual, and the question at the end is often where the most useful client conversation starts.

Step 5: Pilot on five clients for two closes

Don't roll this out to everyone at once:

1. Pick five clients with different shapes: seasonal, steady, growing, messy, one with a budget.
2. Run the workflow for two month-ends.
3. Track how long review takes and what bookkeepers had to fix most often.
4. Tighten the prompt and thresholds based on those fixes.
5. Then decide which clients get commentary as standard and which get it as part of a higher tier.

Common mistakes

• Letting the AI explain causes. It will produce plausible reasons. Plausible is not the same as true.
• Too many bullets. If every line gets a sentence, the client reads none of them.
• Skipping the questions. Questions are where you catch miscoding and show you are paying attention.
• Pasting client files into a personal account. Your firm's AI rule sheet should list which accounts are allowed.

Why this matters for your fees

Commentary is the visible part of month-end. When the ledger agents make the keystrokes faster, the client can't see that; they can see a clear note. That makes it a natural part of a fixed monthly package rather than an hourly line. We go through the numbers in how to price bookkeeping when AI makes the work faster.

---

If you'd like help setting this up on your own client list, start a free conversation. We'll look at your month-end and tell you plainly whether a commentary workflow is worth building.