Analyze a client's monthly financials and flag what deserves your attention

Traditional vs AI-native

The traditional way

1. Export the P&L and balance sheet by month from the accounting system.
2. Paste into an Excel template and calculate dollar and percent variances.
3. Scan for large or odd movements, then drill into the general ledger to see what's behind them.
4. Check relationships by eye: did cost of goods move with revenue, did payroll taxes move with wages?
5. Write notes and client talking points, often under deadline.

Where it hurts

• Most of the time goes to building the comparison, not thinking about it.
• Variance columns flag everything that moved, including the expected, so the real issue gets buried.
• Relationships between accounts are easy to miss when you look one line at a time.
• Every client gets the same template, even though each owner cares about different numbers.

The AI-native way

1. 1Automation

   Pull the data and the context

   Export the trial balance by month for the last 13 months and the GL detail for the current month, from the accounting system's reports or its API. Add a short client profile the accountant maintains: what the business does, seasonality, known one-time events this year, and what the owner watches.

   Tools: QuickBooks Online or Xero reports or API · A client profile document

2. 2Automation

   Compute variances in code

   A script or spreadsheet calculates change versus prior month, same month last year, and trailing average, plus a handful of ratios the firm cares about (gross margin, payroll as a percent of revenue, days sales outstanding). The model never does this arithmetic.

   Tools: Python, or an AI tool that runs code on your file · Excel template

3. 3AI

   Find what doesn't fit

   The model gets the computed variances, the ratios, the GL detail behind the largest movements, and the client profile. It separates expected movements (seasonality the profile describes) from unexpected ones, and looks across accounts: revenue up but merchant fees flat, wages up but payroll taxes flat, a new vendor with large round-dollar payments.

   Tools: Enterprise LLM

4. 4AI

   Rank items for the accountant

   Output is a short ranked list: likely misposting or cutoff issue, needs a client question, or real business change worth discussing. Each item cites the accounts, amounts, and GL lines behind it.

5. 5You

   Investigate and decide

   The accountant works the list: fixes misposts, confirms what is real, and decides which questions to send. Items that turn out to be expected get added to the client profile so they aren't flagged next month.

6. 6AI

   Draft the narrative and charts

   For the confirmed items, the model drafts a plain-English summary built around the numbers that client's owner cares about, plus two or three charts. The accountant edits and approves before anything goes to the client.

   Tools: Enterprise LLM · Charting in Excel or the AI tool

What you need

• Monthly trial balances for 13 months and current-month GL detail
• A one-page client profile: the business, seasonality, one-time events, what the owner watches
• Firm-defined review criteria and materiality thresholds
• An enterprise AI tool that can run code on a file, under a data agreement
• An accountant who reviews every flagged item

The stack

Named tools are examples, not endorsements. Check any tool against your own data rules first.

Copy the prompt

You are a senior CAS accountant reviewing a small business client's month before the client meeting.

You will receive:
1. PROFILE: what the business does, seasonality, known one-time events, and what the owner watches.
2. VARIANCES: a table already computed by code (account, current month, prior month, same month last year, trailing 12-month average, dollar and percent changes). These numbers are authoritative. Do not recalculate them.
3. RATIOS: gross margin, payroll percent of revenue, DSO, and any others provided, by month.
4. GL_DETAIL: current-month transactions for the ten largest movements.

Tasks:
A. Classify each significant movement as expected (explained by PROFILE), unexpected, or unclear.
B. Look across accounts for relationships that don't hold: revenue vs merchant fees or COGS, wages vs payroll taxes and benefits, new vendors, round-dollar or duplicate payments, items that look like they belong in another period.
C. Rank the unexpected and unclear items: "possible misposting or cutoff", "question for client", or "real change to discuss". Cite the accounts, amounts, and GL lines for each.
D. Draft up to five client questions in plain language.
E. Draft a 150-word summary for the owner built around what PROFILE says they watch. Mark it DRAFT.

Rules:
- Use only numbers in the input. If a number you need is missing, say so.
- Do not state a conclusion about why something happened unless the GL detail supports it. Otherwise phrase it as a question.

Example

Fictional data.

PROFILE: Three-location coffee roaster and cafe. Revenue dips in summer. Owner watches
cash, gross margin, and labor percent. New wholesale account signed in July.
VARIANCES (August):
  Revenue           $186,400   +9% vs July   +22% vs Aug last year
  COGS: Green coffee $51,200   +31% vs July
  Merchant fees       $3,020    -4% vs July
  Wages              $58,900    +6% vs July
  Payroll taxes       $2,110   -52% vs July
GL_DETAIL: Green coffee includes two invoices from Pacific Importers for $14,400 each, dated 8/12.

1. POSSIBLE MISPOSTING  Payroll taxes down 52% while wages rose 6%.
   Likely a missed or misclassified payroll tax liability entry for one pay period.
   Check the August payroll journal entries.
2. QUESTION FOR CLIENT  Two identical $14,400 invoices from Pacific Importers on 8/12.
   Duplicate, or two shipments? One invoice alone is more than the month's increase.
3. EXPECTED  Revenue +22% year over year, consistent with the new wholesale account.
   But merchant fees fell 4%: check whether wholesale is paid by ACH (would explain it).

DRAFT owner summary: Sales were up 22% over last August, mostly from the new wholesale
account. Before we talk margin, we're confirming one supplier invoice that may have been
entered twice. Labor was 31.6% of revenue, in line with July.

Guardrails

• Code computes every variance and ratio. The model classifies, explains, and drafts. Never take a number from model output that is not in the input.
• Nothing reaches the client until the accountant has investigated the flagged items and approved the summary.
• Client financial data goes only to an enterprise AI tool under a data agreement, or is anonymized first.

What can go wrong

How to prove it worked

Where this goes

Run the same review weekly, or every time the books sync, and monthly reporting becomes continuous. The client profile grows into a record of how this particular business works, and the summary becomes something the owner actually reads because it is written around what they care about.

Demo

Keep browsing

• Small impactCAS and bookkeeping

  Clean a messy bank or ledger export and get the formulas written for you

  Turn a CSV with merged headers, split descriptions, text-formatted numbers, and subtotal rows into a clean table you can pivot, with a check that nothing was lost.

  Saves 10 to 30 minBeginnerTask
• Large impactCAS and bookkeeping

  Run month-end close with an agent doing the first pass

  Close a client's books each month with an agent that reconciles, categorizes, proposes adjusting entries, and prepares the review package, while the accountant approves every change.

  Changes the workflowAdvancedAgent
• Medium impactTax

  Catch dropped carryforwards and missing schedules before prep starts

  Compare this year's documents and draft return to the last two or three years and surface what vanished, what was dropped, and what swung without an explanation.

  Saves 10 to 30 minIntermediateWorkflow