Skip to main content

Practitioner-Grade Intelligence

AI in the Books: Workflows for Firms

Every use case here is built to actually work — not just sound good. Security-tiered and practitioner-reviewed so you know what’s safe before you start.

Built ByJim Smart, BI Developer

Security Guardrails

Client information is confidential. Before using any workflow, check the security tier to ensure you aren’t violating professional ethics or exposing privileged data.

Read Governance Policy
Tier 1: Unrestricted

No client data involved. Safe for any public or free AI platform.

Tier 2: Anonymize

Requires stripping client identifiers. Safe on paid/pro platforms after scrubbing.

Tier 3: Local / Enterprise

Raw confidential data. Requires local LLMs (Ollama) or enterprise-grade platforms with a signed DPA.

AICPA Rule 301 applies. Feeding identifiable client data into free consumer AI may violate your professional ethics obligations. A 2026 federal court held that exchanges with public AI platforms are not privileged — a direct warning for any professional using free AI with client data. When in doubt: anonymize first or use Ollama locally. It’s free and data never leaves your machine.

Governance Policy

AI Data Governance for Accounting Firms

The security tier system on this page follows a principled framework. Here is how it works, what it means in practice, and how to implement it across your firm.

Why Tiers Exist

The central question in any AI workflow is whether the input contains information that could identify a specific person, business, or tax situation. That answer determines where data can safely go — not the sensitivity of the analysis itself.

Tier 1: Unrestricted

Internal firm processes, templates, and general professional knowledge. No identifiers, no financial data. Free AI platforms are appropriate — there is nothing confidential in the input.

Tier 2: Anonymize

Dollar amounts, account categories, and financial patterns with client identifiers removed. Meaningfully different from confidential data, but free consumer AI tools may use inputs for model training. Paid business tiers contractually exclude your data from training.

Tier 3: Local / Enterprise

Names, SSNs, EINs, tax return detail, IRS correspondence, payroll with employee names. Requires local AI (Ollama, free, no internet) or enterprise platforms with a signed Data Processing Agreement.

What “Anonymizing” Actually Means

Removing the client’s legal name from a report header is the minimum. It is not always sufficient. For Tier 2 compliance, work through this checklist before pasting anything:

  • Remove legal name, trade name, and DBA from all headers and footers
  • Search for the owner's name — it often appears in company name fields
  • Check memo fields for vendor names that identify the client
  • Confirm no Social Security numbers, EINs, or bank account numbers appear
  • Review dollar amounts — a single unusual transaction may identify the business
  • If uncertain whether any of the above applies, escalate to Tier 3 protocols

Platform Selection Quick Reference

Data TypeFree Claude / ChatGPTClaude for WorkChatGPT BusinessM365 CopilotOllama (local)

Internal firm process

Tier 1 — Green

Anonymized financial data

Tier 2 — Yellow

Raw client data (SSN, EIN, returns)

Tier 3 — Red

✗*✗*✗*

* Red-tier workflows require a signed Data Processing Agreement (DPA) with any cloud provider before use. Confirm in writing.

AICPA Rule 301 — What It Requires

AICPA Code of Professional Conduct Rule 301 requires members to protect confidential client information obtained in the course of professional services. Using a free consumer AI tool that retains inputs — even temporarily — is a disclosure risk. Several state boards have issued informal guidance treating AI platforms as third parties under confidentiality obligations.

A 2026 federal court held that exchanges with public AI platforms are not protected by privilege — an important signal that courts are treating AI-shared data as non-confidential.

When You Are Uncertain

Default to a higher tier. If you are not sure whether something is yellow or red, treat it as red and use Ollama locally.

The cost of running Ollama on a case that could have gone to ChatGPT Business is zero. The cost of putting raw client data on the wrong platform is a Rule 301 violation at minimum, and a professional liability exposure at worst.

Ollama is free, runs on most laptops, and takes under 10 minutes to set up at ollama.com.

Firm Implementation Checklist

Before rolling any AI workflow out to staff, work through these six items.

1

Establish a written firm AI tool policy documenting which tools are approved for each data tier

2

Subscribe to at least one paid business-tier platform (Claude for Work or ChatGPT Business) for Tier 2 work

3

Install Ollama locally for Tier 3 work — it is free and data never leaves the machine

4

Brief all staff on the three-tier framework before they begin using AI on client work

5

Review engagement letters — consider whether client consent language is needed for AI-assisted services

6

Consult your professional liability carrier — many are now publishing guidance specific to AI use

01
Tier 1: Unrestricted

Month-End Close Checklist Generator

Build a firm-specific checklist in 10 minutes instead of 2 hours.

month-endclosechecklistworkflowstaff training

The Problem

Most close checklists are generic templates pulled from the internet or stored in one person's head. When that person is out, steps get missed. When you onboard new staff, training takes weeks instead of days. AI generates a firm-specific, phased checklist from a description of your process in under 10 minutes — one you can actually hand to someone.

What You Need

  • Any AI platform — no paid tier required, no client data is involved
  • 10–15 minutes to describe your close process
  • List of entity types your firm closes (LLC, S-corp, nonprofit, etc.)
  • Optional: your current checklist if you want to improve an existing one

Execution Roadmap

1

Write down your entity types and current close steps

Before opening AI, spend 5 minutes listing the entity types your firm closes (S-corps, LLCs, nonprofits, HOAs) and the recurring tasks your team does every month — bank recs, AP aging review, accrual entries, payroll reconciliation, intercompany eliminations. You don't need to be exhaustive. AI will fill in standard steps and flag gaps.

2

Paste the prompt and your process description

Tell AI what your firm closes, your current steps, and what role the checklist is for — staff accountant, senior, or manager. The role matters. A staff checklist is task-level. A reviewer checklist is judgment-level. Mixing them produces something neither person will use.

3

Review every line before you finalize

Read each item. AI includes standard steps you may have missed, but also steps that don't apply to your client mix. Delete what doesn't fit. Add the firm-specific items AI cannot know: your naming conventions, file locations, report formats, and who signs off on what.

The Engineered Prompt

Paste directly into any AI chat
You are helping a CPA firm build a month-end close checklist for internal use.

Firm profile:
- Entity types we close: [e.g., S-corps, LLCs, nonprofits, HOAs]
- Accounting software: [QuickBooks Online / Xero / Sage / other]
- Close deadline: [e.g., books due by the 15th of the following month]
- This checklist is for: [Staff accountant / Senior accountant / Manager review]

Our current close steps (describe what you know — even if incomplete):
[DESCRIBE YOUR PROCESS HERE]

Generate a complete month-end close checklist that:
1. Covers all standard close tasks for the entity types listed
2. Groups tasks into phases: Pre-Close, Core Close, Review, Client Deliverables
3. Includes a clear owner for each task (Staff / Senior / Manager)
4. Flags high-risk steps — timing cutoffs, audit trail items, common errors
5. Is formatted as a table: Task | Owner | Due | Risk Flag | Done

After the checklist, list any questions you would ask before finalizing it for this firm.

Expected Outcome

AI returns a phased close checklist in table format with owner assignments, timing guidance, risk flags, and a short list of clarifying questions. A 2-hour documentation task becomes a 15-minute review.

S

Practitioner Perspective

Sydney Smart, CPA, MST

The value here is not the checklist itself — it is the conversation the prompt forces you to have. When I walked through this with our own close process, I caught three steps we had been handling inconsistently for two years. The output AI produced was not perfect, but describing our process out loud exposed the gaps. That is where the real return is.

Known Constraints

  • AI does not know your firm's naming conventions, file locations, or approval workflows. Every internal reference must be added by you after the fact.
  • The checklist will include standard steps that may not apply to your client mix — inventory reconciliation steps if you mentioned product businesses, for example. Delete what does not fit.
  • Do not treat the output as final without a senior review. AI produces a strong starting point, not a peer-reviewed procedure manual.
  • Update the checklist when your process changes. A checklist that reflects how you closed 18 months ago is worse than no checklist because it creates false confidence.

Security Deployment

No client data is involved in this workflow. You are describing your internal firm process — entity types, software, role structure, timing. Free AI platforms are appropriate here. If your description of the close process happens to include specific client names, remove them before pasting — but that level of detail is not necessary for this prompt to work.

Claude.ai (free)Free
ChatGPT (free)Free
Microsoft CopilotFree / included with M365
02
Tier 2: Anonymize

Cash Flow Narrative for Ownership

Turn a statement of cash flows into a paragraph an owner can actually read.

cash flowownership reportingmonth-endnarrativeQuickBooks

The Problem

Owners see the cash flow statement and ask the same two questions every month: 'Why is cash down if we're profitable?' and 'Where did the money go?' Writing a clear answer takes 20 minutes and often gets skipped because it feels low-priority at month end. AI drafts the plain-English narrative in under 90 seconds. Your job is confirming it reflects what you know about the business.

What You Need

  • Statement of Cash Flows from QuickBooks Online, Xero, or your GL system
  • Remove the client's legal name from the report header before exporting or copying
  • Two periods if available — current month and prior month, or YTD vs. prior YTD
  • Claude for Work, ChatGPT Team, or Ollama — see security note before you start

Execution Roadmap

1

Export the Statement of Cash Flows

In QuickBooks Online: Reports → Statement of Cash Flows. Set the date range to the current month or YTD period. Before copying the report, delete or replace the client's company name in the header with a generic label like 'Client A.' The dollar totals and category labels contain no personally identifiable information.

2

Add the context AI cannot see in the numbers

The context field is what separates a useful narrative from a generic one. Before pasting, write one or two sentences about what actually happened: 'We made a $40,000 equipment purchase in March.' / 'AR collections were slow — outstanding balance grew by $22,000.' / 'Owner took a distribution of $35,000.' Without this, AI will describe the numbers but not explain them.

3

Review the narrative before sending

Read the output paragraph aloud. If it sounds like a financial statement footnote, it is too technical for an ownership summary — ask AI to simplify. If a number is wrong, check whether you pasted the right period. The goal is a paragraph the owner reads instead of skims.

The Engineered Prompt

Paste directly into any AI chat
You are helping a CPA prepare a plain-English cash flow summary for a business owner who is not an accountant.

Below is the Statement of Cash Flows for [CURRENT PERIOD] and [PRIOR PERIOD] (client name removed).

Context about this period: [ADD WHAT YOU KNOW — examples: "We purchased a vehicle in January for $42,000." / "Owner took a $30,000 distribution." / "Collections were behind — AR grew by $18,000." / "No major one-time items — the variance is operational."]

Your tasks:
1. Identify the top 2–3 drivers of the cash change this period
2. Write a 3–5 sentence plain-English summary for a business owner — explain operating vs. investing vs. financing in plain language
3. Answer the question 'Why is cash down even though we're profitable?' if operating income is positive but cash decreased
4. Flag any line items I should verify or explain further to the owner

Format:
— Top cash drivers (bulleted, largest first, dollar amounts included)
— Owner summary (plain-English paragraph, no accounting jargon)
— Items to verify or clarify

[PASTE YOUR STATEMENT OF CASH FLOWS HERE]

Expected Outcome

AI returns the top cash drivers ranked by dollar impact, a 3–5 sentence owner-ready narrative, and a short list of items to verify. The explanation that used to get skipped now gets written every month.

S

Practitioner Perspective

Sydney Smart, CPA, MST

The question I hear most often after distributing financials is not about the income statement — it is about cash. 'We made money. Why is cash down?' This workflow does not replace that conversation, but it makes the written explanation consistent enough that the owner actually reads it before they call. That alone saves 20 minutes most months.

Known Constraints

  • AI cannot distinguish between a timing difference and a real cash problem. A large AR increase looks the same as a collection failure in the numbers. You have to tell it which one it is in the context field.
  • If the client name appears anywhere in the export — header, footer, company field — remove it before pasting. Dollar totals and category labels are fine. Named company data is not.
  • Free consumer ChatGPT and free Claude.ai are not appropriate for financial statement data. Use a paid business tier or run Ollama locally.
  • AI will produce a plausible-sounding narrative even if the numbers are wrong. Verify the period totals match your GL before sending the explanation to ownership.

Security Deployment

Remove the client's company name from the report header before pasting. What you are sharing is cash category totals and dollar amounts — no tax ID, no individual names, no account numbers. That puts this in the yellow tier. Do not use free consumer AI tools (free ChatGPT or free Claude.ai) for financial statement data — both may use inputs for model training by default. Safe options: Claude for Work ($25/user/mo), ChatGPT Team ($30/user/mo), Microsoft Copilot M365 (included with most Business plans), or Ollama (free, local).

Claude for Work$25/user/mo
ChatGPT Business$25/user/mo
Microsoft Copilot M365Included with most M365 Business plans
Ollama (local)Free
03
Tier 2: Anonymize

Budget vs. Actual Variance Narrative

Write the management commentary in 90 seconds instead of 30 minutes.

budgetvariancemanagement reportingmonth-endcommentary

The Problem

Budget-to-actual reports tell management what happened. The variance narrative tells them what it means. Writing that commentary from scratch — picking the right variances to highlight, drafting clear explanations, flagging what needs action — takes 25–35 minutes per client. AI drafts a complete management commentary in under 2 minutes when you give it the numbers and the context.

What You Need

  • Budget vs. Actual P&L report from QuickBooks Online, Xero, Sage, or your GL system
  • Remove the client's company name from the report header before copying
  • Current month and YTD columns if available — both periods improve the narrative
  • Claude for Work, ChatGPT Team, or Ollama — see security note before you start

Execution Roadmap

1

Export the Budget vs. Actual report

In QuickBooks Online: Reports → Budget vs. Actuals. Set the date range to the current month and include YTD columns if available. Before copying, replace the client name in the header with 'Client A' or a generic label. The revenue and expense categories, budget amounts, and dollar variances contain no personally identifiable information.

2

Note the context before you paste

Identify 2–3 variances that have a known explanation — a delayed hire, a one-time equipment purchase, a revenue shortfall on a specific contract. Write them down before you open AI. The prompt has a context field; what you put there is the difference between a narrative that sounds like it knows the business and one that reads like a template.

3

Review the commentary for accuracy and tone

Read the output commentary against what you know about the business. AI will accurately describe the numbers but may misread significance — it cannot tell you whether a 12% revenue shortfall is alarming or expected. Your review adds the judgment layer before this goes to management.

The Engineered Prompt

Paste directly into any AI chat
You are helping a CPA prepare a budget-to-actual management commentary for the period ending [MONTH/YEAR].

Below is a Budget vs. Actual P&L report (client name removed). Columns: Account | Budget | Actual | Variance $ | Variance %

Context about this period: [ADD WHAT YOU KNOW — examples: "Hiring for the operations role is delayed by 6 weeks — salary variance is planned." / "Revenue shortfall on Contract XYZ — they pushed the project to Q3." / "Equipment purchase in June was unbudgeted but approved by ownership." / "No unusual items — all variances are operational."]

Your tasks:
1. Identify the top 5 favorable and unfavorable variances by dollar amount
2. Write a 4–6 sentence management commentary suitable for a CFO or ownership group
3. Separate planned variances from unexpected ones based on the context I provided
4. Flag any variances I should investigate further before finalizing the commentary

Format:
— Top variances table (Account | Budget | Actual | Variance $ | Favorable/Unfavorable)
— Management commentary (paragraph format, plain English, no jargon)
— Variances to investigate

[PASTE YOUR BUDGET VS. ACTUAL REPORT HERE]

Expected Outcome

AI returns a ranked variance table, a 4–6 sentence management commentary, and a list of items to investigate. A 30-minute writing task becomes a 5-minute review.

S

Practitioner Perspective

Sydney Smart, CPA, MST

The clients who use budget-to-actual reports well are the ones who get a written explanation with the numbers, not just the numbers. This workflow makes that possible every month without it taking 30 minutes. The one thing I always do before sending: I read the commentary out loud and ask whether it sounds like something I would say about this specific business. If it sounds generic, I add one specific sentence that could only be true for this client.

Known Constraints

  • AI reads the numbers you give it. If a budget line is wrong because of a data entry error or an unapproved budget revision, the commentary will repeat the error.
  • The context field is what separates a planned variance from an unexplained one in the output. Without it, AI flags everything as potentially problematic. With it, the commentary distinguishes between 'as expected' and 'needs attention.'
  • Management commentary is a communication document, not just a math exercise. Read the output for tone — if it sounds defensive about an unfavorable variance, soften it. If it is too soft on a real problem, sharpen it.
  • AI cannot apply materiality judgment. A $500 variance at 40% might get the same attention as a $50,000 variance at 3%. Apply your own materiality threshold before deciding what to include.

Security Deployment

Remove the client's company name from the report header before pasting. What you are sharing is account category labels and dollar amounts — no tax ID, no individual names, no bank account data. Use a paid business AI tier or Ollama locally. Free consumer tools are not appropriate for financial statement data.

Claude for Work$25/user/mo
ChatGPT Business$25/user/mo
Microsoft Copilot M365Included with most M365 Business plans
Ollama (local)Free
04
Tier 2: Anonymize

AP Duplicate Payment Detection

Scan for duplicate payments before close without a dedicated tool.

accounts payableduplicate paymentsinternal controlscloseQuickBooks

The Problem

Duplicate payments are easy to miss in high-volume AP — same vendor, slightly different invoice numbers, a few days apart. Most small firms do not have software that flags them automatically. A manual review takes 45 minutes and rarely happens consistently. AI can scan an AP transaction export for duplicate patterns in under 2 minutes and return a prioritized list of items worth checking.

What You Need

  • AP transaction detail report from QuickBooks Online, Xero, or your GL system
  • Date range: at minimum the last 90 days — broader range catches more patterns
  • Remove the client's company name from the report header
  • Claude for Work, ChatGPT Team, or Ollama — see security note before you start

Execution Roadmap

1

Export the AP transaction detail report

In QuickBooks Online: go to Reports and search for 'Transaction List by Vendor.' Set the date range to the last 90 days. Include columns: Date, Vendor, Invoice Number, Amount, Payment Date. Before copying, replace the client's company name in the report header with 'Client A.' Vendor names and invoice numbers are business transaction data — not personal client information.

2

Paste and specify what to look for

Tell AI the time period and ask it to look for the three most common duplicate patterns: same vendor and same amount within 30 days, same vendor and same invoice number, and same vendor and similar invoice number with the same amount. A fourth check — same amount across different vendors on the same day — catches rare but high-risk journal entry errors.

3

Review the flagged items in QuickBooks

Pull up each flagged transaction in QuickBooks. Most will be legitimate — installment payments, recurring subscriptions, separate deliveries billed at the same amount. What you are looking for is the same invoice paid twice, which is rarely intentional. Mark each flag as resolved or escalate before close.

The Engineered Prompt

Paste directly into any AI chat
You are helping a CPA review accounts payable transactions for potential duplicate payments.

Below is an AP transaction list covering [DATE RANGE] (client name removed). Columns: Date | Vendor | Invoice Number | Amount | Payment Date

Please scan for the following duplicate patterns and report each match:
1. Same vendor + same amount, payments within 30 days of each other
2. Same vendor + same invoice number, regardless of date
3. Same vendor + same amount + invoice numbers that differ by only 1–2 characters (e.g., INV-1042 and INV-1024)
4. Same dollar amount from different vendors on the same calendar day (flag only if amount exceeds $500)

For each match found:
— List both transactions side by side: Date | Vendor | Invoice # | Amount
— Assign a risk level: High (exact invoice number match), Medium (same amount within 30 days), Low (similar invoice number)
— Add a one-line note on what to check in the source system

Sort results by risk level, highest first. If no duplicates are found, confirm the scan was completed and state which patterns were checked.

[PASTE YOUR AP TRANSACTION LIST HERE]

Expected Outcome

AI returns a ranked list of potential duplicate payment matches sorted by risk level, with side-by-side transaction comparisons and a note on what to verify in QuickBooks. A 45-minute manual review becomes a 10-minute verification task.

S

Practitioner Perspective

Sydney Smart, CPA, MST

I ran this on a client's AP register after month-end close and found two duplicate payments that had been sitting there for 10 weeks — same vendor, different invoice numbers that turned out to be the same invoice submitted twice by the vendor's AP department. The client got a credit on the next statement. That is not a common outcome, but the scan costs 10 minutes and the alternative is not finding it until the vendor calls or the audit.

Known Constraints

  • AI can only detect patterns in the data you give it. If the AP export is incomplete — missing a date range, excluding certain payment types, or capped at a row limit — duplicates in the missing data will not be found.
  • Most flagged items will be legitimate: recurring subscriptions billed monthly at the same amount, installment payments, split invoices for the same vendor. Do not escalate without verifying in the source system first.
  • Invoice number formats vary widely by vendor. AI may miss duplicates where one vendor uses numeric-only invoice numbers and another uses alphanumeric. A manual spot-check on your largest vendors by dollar volume is still good practice.
  • This workflow detects payment-level duplicates. It does not detect vendor master duplicates (same vendor entered twice under slightly different names) or fraudulent vendor creation. Those require separate controls.

Security Deployment

The AP transaction export contains vendor names, invoice numbers, and dollar amounts. It does not typically contain client employee names, Social Security numbers, or bank account details — which keeps this in the yellow tier. Remove the client company name from the report header. Do not paste data that includes bank account numbers, routing numbers, or wire transfer details — that escalates this to a red-tier workflow. Use a paid business AI tier or Ollama locally.

Claude for Work$25/user/mo
ChatGPT Business$25/user/mo
Microsoft Copilot M365Included with most M365 Business plans
Ollama (local)Free
05
Tier 1: Unrestricted

Chart of Accounts Cleanup Review

Audit a bloated COA and get a consolidation plan in 15 minutes.

chart of accountsQuickBookscleanupreportingaccounting setup

The Problem

Chart of accounts bloat is one of the most common problems in small business accounting — and one of the least often fixed. Duplicate accounts, over-segmented expense categories, inactive accounts cluttering reports, and misclassified account types all make financial reporting harder to read and easier to misinterpret. Cleaning it manually takes hours. AI reviews a full COA and returns a prioritized consolidation plan in minutes.

What You Need

  • Any AI platform — no paid tier required, no client financial data is involved
  • Chart of Accounts export from QuickBooks Online, Xero, or your GL system (account names and types only — no balances required)
  • 30 minutes to review and implement changes after AI returns the plan

Execution Roadmap

1

Export the Chart of Accounts

In QuickBooks Online: Accounting → Chart of Accounts → Export to Excel (or copy the list). You only need two columns: Account Name and Account Type (Asset, Liability, Equity, Income, Expense). You do not need balances, account numbers, or transaction history for this review. The COA structure itself contains no confidential client information.

2

Add context about the business

Before pasting, tell AI the entity type (LLC, S-corp, nonprofit), the industry, and the size of the business (solo operator, 5 employees, 50 employees). A COA appropriate for a single-member LLC looks different from one built for a 15-person professional services firm. Context prevents AI from recommending consolidations that would eliminate distinctions that actually matter for this client.

3

Review the consolidation plan before making changes

Do not merge or delete accounts in QuickBooks before reviewing AI's suggestions carefully. Some accounts that look redundant are separated intentionally — for tax purposes, for job costing, or for loan covenant reporting. Any merge in QuickBooks that affects historical transactions cannot be fully undone without a data restore.

The Engineered Prompt

Paste directly into any AI chat
You are helping a CPA review a client's Chart of Accounts for cleanup and consolidation opportunities.

Client context:
- Entity type: [LLC / S-corp / C-corp / Nonprofit / Sole proprietor]
- Industry: [e.g., HVAC contractor, dental practice, retail, software company]
- Size: [e.g., solo operator / 3 employees / 20 employees]
- Accounting software: [QuickBooks Online / Xero / Sage / other]
- Problem we are trying to solve: [e.g., reports are hard to read / too many accounts / duplicates / unclear account names]

Below is the Chart of Accounts (Account Name and Type only):
[PASTE COA HERE]

Please review and provide:
1. Duplicate or near-duplicate accounts that should be merged (list both account names)
2. Accounts that appear miscategorized by type (e.g., an expense coded as an asset)
3. Over-segmented areas where 4–6 accounts could become 1–2 without losing reporting value
4. Account names that are unclear or inconsistent with standard chart naming conventions
5. Accounts that are likely inactive and candidates for archiving

Format your output as a prioritized action list:
— High priority (errors that affect financial statement accuracy)
— Medium priority (consolidations that improve readability)
— Low priority (cleanup and naming conventions)

For each item, explain the rationale in one sentence.

Expected Outcome

AI returns a prioritized cleanup plan organized by severity — errors first, readability improvements second, cosmetic fixes last. Each recommendation includes a one-line rationale you can share with the client or use to decide whether to act.

S

Practitioner Perspective

Sydney Smart, CPA, MST

I do this review every time I take on a new bookkeeping client and usually find the same three problems: 10–15 expense accounts that overlap enough to confuse anyone reading the P&L, at least one account coded to the wrong type, and accounts from a prior owner or prior software migration that have never been used. This prompt does the initial scan in minutes. The judgment call — which ones are safe to merge — still requires someone who knows the client.

Known Constraints

  • Some accounts that look like duplicates are intentionally separate — for job costing, 1099 reporting, or loan covenant tracking. Do not merge based on AI's recommendation alone. Verify the intent before making changes.
  • Merging accounts in QuickBooks affects historical transaction classification. Once merged, the history cannot be unsplit without a data restore. Always back up before making chart changes.
  • AI reviews the structure — it cannot see the transaction history behind each account. An account with only 2 transactions in 3 years might be worth archiving, or it might be used for a quarterly tax accrual. You need that context.
  • Account type errors (expense coded as asset, for example) may have been there for years and affect prior period financial statements. Fix them carefully, and consider whether an amended report is needed.

Security Deployment

The Chart of Accounts contains account names and account types — no dollar balances, no transaction history, no client-identifying information beyond the account structure itself. Free AI platforms are appropriate for this workflow. If the client name appears in the account names (which occasionally happens), remove or replace it before pasting.

Claude.ai (free)Free
ChatGPT (free)Free
Microsoft CopilotFree / included with M365
06
Tier 2: Anonymize

Client Meeting Prep from Prior Notes

Turn last quarter's notes into a ready agenda in 5 minutes.

client meetingsagendaadvisorymonth-endclient communication

The Problem

Preparing for a client financial review meeting means re-reading prior meeting notes, pulling the key issues that were open, checking whether they were resolved, and drafting an agenda that covers both the financials and the outstanding items. When you have 6 clients with reviews in the same week, that prep time compounds. AI can turn a set of prior notes and a current P&L summary into a ready agenda in under 5 minutes.

What You Need

  • Your notes from the last 1–2 client meetings (exported from your notes app, CRM, or typed summary)
  • A summary of the current period financials — key line items only, not the full P&L
  • Remove the client's legal name and replace with 'Client A' before pasting
  • Claude for Work, ChatGPT Team, or Ollama — see security note before you start

Execution Roadmap

1

Gather your prior meeting notes and current period highlights

Pull your notes from the last meeting — action items, open questions, decisions made. Also pull 3–5 key numbers from the current period: revenue vs. prior month, a notable expense change, cash position if relevant. You do not need to paste the full financial statements. A bulleted list of highlights is enough.

2

Replace the client name before pasting

Search your notes for the client's legal name, trade name, and any employee names mentioned. Replace them with 'Client A' or a neutral label. The financial context and discussion items are what AI needs — not the name.

3

Review the agenda before the meeting

Read the output agenda once. Add the items AI could not know — a question the owner texted you last week, a referral conversation you want to have, a pricing discussion you have been putting off. Remove items that are resolved or no longer relevant. The agenda should be 4–6 items, not a transcript of your notes.

The Engineered Prompt

Paste directly into any AI chat
You are helping a CPA prepare for a client financial review meeting.

Client context (name removed):
- Entity type: [LLC / S-corp / other]
- Industry: [e.g., HVAC contractor / dental practice / retail]
- Meeting frequency: [monthly / quarterly]
- Meeting length: [30 minutes / 60 minutes]

Prior meeting notes (client name removed):
[PASTE YOUR NOTES FROM THE LAST 1–2 MEETINGS]

Current period highlights:
[PASTE 3–5 KEY FINANCIAL ITEMS — e.g., "Revenue up 12% vs. last month. Payroll variance of $6,200 due to new hire. Cash decreased $18,000 — equipment purchase. AR balance higher than usual."]

Please prepare:
1. A 4–6 item meeting agenda with estimated time for each item
2. A status summary for any open action items from prior meeting notes (resolved / still open / unknown)
3. 2–3 questions I should ask the owner during the financial review
4. One sentence I can open the meeting with that frames the current period in plain English

Format:
— Opening line
— Agenda (item | time | owner: CPA or Client)
— Open items status
— Questions to ask

Expected Outcome

AI returns an opening line, a timed agenda with owners assigned, a status update on prior action items, and 2–3 questions to ask the owner. Meeting prep drops from 20 minutes to a 5-minute review.

S

Practitioner Perspective

Sydney Smart, CPA, MST

The part of this that saves the most time is not the agenda — it is the open items status. When you have notes from 3 or 4 prior meetings, finding the action items that were never closed takes longer than writing the agenda. Asking AI to pull them out and flag their status catches the things that fell through the cracks between meetings. That is where client relationships get damaged quietly.

Known Constraints

  • AI can only assess open items as 'resolved' or 'still open' based on what appears in your notes. If an item was resolved verbally but not documented, AI will mark it as open. Your review catches this.
  • The 3–5 financial highlights you paste determine the quality of the agenda. Vague inputs ('revenue was fine') produce vague agendas. Specific inputs ('revenue up 12% but gross margin declined 3 points') produce the questions worth asking.
  • Meeting agendas have a political layer AI cannot see — topics the owner is sensitive about, relationships with co-owners, pending decisions that are not ready to discuss. Your review inserts that layer.
  • Do not skip removing the client name. Meeting notes often contain employee names, vendor names, and personal financial details that belong in the yellow tier — not the red tier, but not shareable on free consumer tools.

Security Deployment

Meeting notes often contain client financial details, owner names, and business decisions. Replace the client's legal name and any employee names with generic labels before pasting. What you are sharing is context and discussion items — not tax returns, SSNs, or bank account data. Use a paid business AI tier or Ollama locally. Free consumer tools are not appropriate for client-specific financial context.

Claude for Work$25/user/mo
ChatGPT Business$25/user/mo
Microsoft Copilot M365Included with most M365 Business plans
Ollama (local)Free
07
Tier 3: Local / Enterprise

IRS Notice Response Drafting

Draft a professional CP2000 or balance-due response in 10 minutes.

IRStax noticesCP2000correspondencetax practice

The Problem

IRS notices require a timely, accurate written response — but drafting one from scratch means re-reading the notice, researching the applicable code section, structuring the argument clearly, and formatting it for IRS correspondence standards. For high-volume practices, this task alone can consume 30–60 minutes per notice. AI drafts the response structure and language in minutes. Your job is adding the facts, verifying the accuracy, and signing it.

What You Need

  • The IRS notice — CP2000, CP501, CP503, CP504, or letter type — with notice date and response deadline
  • The relevant tax return information for the period in question
  • Ollama running locally OR a paid enterprise platform with a signed Data Processing Agreement
  • 30 minutes total: 10 for setup and AI drafting, 20 for your review, verification, and signature

Execution Roadmap

1

Identify the notice type and the IRS's specific claim

Before drafting a response, read the notice carefully and write down: the notice type (CP2000, CP501, etc.), the tax year in question, the IRS's specific assertion (unreported income, math error, balance due), and the response deadline. These four items go into the prompt. A vague description of the notice produces a vague draft response.

2

Decide which platform to use — this is a red-tier workflow

IRS notices contain the taxpayer's name, SSN or EIN, tax year, and in many cases dollar amounts tied to specific income items. This is raw confidential client data. Use Ollama locally (free, data never leaves your machine) or a paid enterprise platform with a signed Data Processing Agreement. Do not paste IRS notice data into free consumer ChatGPT or free Claude.ai under any circumstances.

3

Review every line of the draft before sending anything

AI drafts the structure and the standard IRS correspondence language — opening reference line, factual summary, legal basis, requested resolution, enclosure list. You add the specific facts, verify that cited code sections are accurate for the period in question, and confirm the dollar amounts match your records. Sign and send only after full review.

The Engineered Prompt

Paste directly into any AI chat
You are helping a CPA draft a professional response to an IRS notice on behalf of a client (taxpayer information will be added separately before mailing).

Notice details:
- Notice type: [e.g., CP2000 / CP501 / CP503 / CP504 / CP3219A / Letter 12C]
- Tax year: [e.g., 2023]
- IRS assertion: [Describe what the IRS is claiming — e.g., "Unreported 1099-NEC income of $12,400 from a single payer" / "Balance due of $3,200 after return processing" / "Mathematical error on Schedule C"]
- Our position: [Describe the taxpayer's position — e.g., "The income was reported correctly on Schedule C" / "The 1099 was from a related-party transaction that was already included" / "Agree with the balance due — requesting a payment plan"]
- Supporting documentation available: [e.g., copy of filed return, 1099, bank statements, prior correspondence]

Please draft a professional IRS response letter that:
1. Opens with the standard IRS correspondence reference line format
2. States our position clearly in the first paragraph
3. Provides a factual summary of the issue and the taxpayer's position
4. Cites the applicable IRC section or IRS publication if relevant
5. Requests the specific resolution: adjustment, abatement, payment plan, or confirmation of no change
6. Closes with an enclosure list placeholder for the supporting documents

Leave [TAXPAYER NAME], [SSN/EIN], and [DATE] as placeholders — I will complete those before printing.

Expected Outcome

AI returns a complete IRS correspondence draft in standard format with placeholders for taxpayer-specific information. A 30–60 minute drafting task becomes a 15-minute review and completion task.

S

Practitioner Perspective

Sydney Smart, CPA, MST

The IRS response letter is one of the most time-consuming tasks in a tax practice because practitioners write from scratch every time. The structure is almost always the same: here is the notice, here is our position, here is the support, here is what we want you to do. AI produces that structure in 2 minutes. What takes the remaining 30 minutes is the same regardless of AI: verifying the facts, confirming the legal basis, and attaching the right documents. This workflow cuts the drafting time, not the thinking time.

Known Constraints

  • AI is not a tax attorney. The draft provides structure and standard IRS correspondence language. The legal accuracy of the position — whether the taxpayer's argument is correct and well-supported — is your professional judgment, not AI's.
  • Tax law changes. Code section citations AI produces should be verified for the specific tax year in question. A citation that was accurate for 2021 may not apply to 2023.
  • IRS deadlines are real. If you are close to the response deadline, do not let the drafting process delay a timely response. A partial response filed on time is better than a complete response filed late.
  • Never paste the actual IRS notice into a free consumer AI tool. The notice contains the taxpayer's SSN or EIN, which triggers federal privacy obligations. This is a red-tier workflow — use Ollama locally or a paid enterprise platform with a signed DPA.

Security Deployment

IRS notices contain the taxpayer's name, SSN or EIN, tax year, and income details. This is the definition of raw confidential client data — it is a red-tier workflow. Do not use free consumer ChatGPT or free Claude.ai. Do not use standard Claude for Work or ChatGPT Team without first confirming a signed Data Processing Agreement is in place with your firm. The safest option is Ollama running locally — it is free, requires no subscription, and data never leaves your machine. Instructions for installing Ollama are at ollama.com.

Ollama (local)Free
Claude for Work + DPA$25/user/mo
ChatGPT Enterprise (with DPA)Enterprise pricing
08
Tier 1: Unrestricted

Engagement Letter First Draft

Draft a new engagement letter for any service scope in under 5 minutes.

engagement letterclient onboardingpractice managementrisk management

The Problem

Writing a new engagement letter means either modifying an existing one (and hoping nothing important gets left in or taken out) or starting from scratch. Either path takes 20–30 minutes for a standard engagement. When you are adding a service, taking on a new client type, or updating outdated language, the risk of errors compounds. AI drafts a complete engagement letter from a service description in minutes — one you edit rather than write.

What You Need

  • Any AI platform — no paid tier required, no client-specific data is needed
  • Description of the service scope, entity type, and fee structure
  • Your firm name and state of licensure (for jurisdiction-specific language)
  • 15–20 minutes to review and customize the draft before sending

Execution Roadmap

1

Define the service scope before opening AI

Write down: the service type (monthly bookkeeping, tax preparation, CFO advisory, audit), the entity type (LLC, S-corp, nonprofit), the deliverables, the fee and billing cycle, and any specific exclusions — what you are explicitly not doing under this engagement. The exclusions section is where most engagement letter disputes originate. Being specific here prevents ambiguity later.

2

Paste the prompt with your firm details

Include your firm name, state, service scope, deliverables, fee structure, and exclusions. Add any client-specific terms you want included — a specific reporting deadline, a data access requirement, a special termination clause. You are not pasting client data. You are describing a service.

3

Review against your professional liability standard

Read the output carefully before sending to any client. Check: is the scope of services clearly defined? Are the exclusions explicit? Is the fee and billing cycle unambiguous? Is the termination clause reasonable for both parties? If your firm uses standard engagement letter templates reviewed by counsel, compare the AI draft against yours — do not replace a reviewed template without a legal review.

The Engineered Prompt

Paste directly into any AI chat
You are helping a CPA firm draft a professional engagement letter for a new client service.

Firm details:
- Firm name: [YOUR FIRM NAME]
- State of licensure: [STATE]
- Engagement start date: [DATE or 'Upon execution']

Service scope:
- Service type: [e.g., Monthly bookkeeping / Annual tax preparation / Fractional CFO / Compilation / Payroll services]
- Entity type: [LLC / S-corp / C-corp / Nonprofit / Sole proprietor]
- Deliverables: [e.g., Monthly P&L, Balance Sheet, and bank reconciliation by the 15th of the following month]
- Reporting period: [Monthly / Quarterly / Annual]
- Fee: [e.g., $600/month flat fee, billed on the 1st / $2,500 fixed fee for 2024 tax return]
- Payment terms: [e.g., Net 15 / Auto-charge on file / Due upon delivery]

Exclusions (what is not included):
[LIST WHAT YOU ARE NOT DOING — e.g., tax planning, audit, payroll, CFO advisory, prior period corrections]

Additional terms:
[ANY SPECIFIC TERMS — e.g., client must provide bank statements by the 5th / 30-day termination notice required / late fees of 1.5% per month]

Please draft a complete engagement letter that includes:
1. Scope of services (with clear deliverables and timeline)
2. Exclusions section
3. Client responsibilities (what the client must provide and when)
4. Fee and payment terms
5. Term and termination clause (including how either party can end the engagement)
6. Confidentiality clause
7. Limitation of liability clause
8. Signature block for both parties

Format as a formal business letter. Leave [CLIENT NAME] and [CLIENT ENTITY] as placeholders.

Expected Outcome

AI returns a complete engagement letter draft with all standard sections, appropriate professional language, and placeholders for client-specific information. A 25-minute writing task becomes a 10-minute review.

S

Practitioner Perspective

Sydney Smart, CPA, MST

I use this primarily for service add-ons — when a bookkeeping client wants to add payroll or when an existing tax client is moving to advisory services. Writing a new engagement letter for a scope change used to mean pulling an old one, trying to remember what I changed last time, and hoping I caught everything. This produces a clean draft I can edit in 10 minutes. I always compare it against my standard reviewed template before sending, but the starting point is cleaner than what I would draft under time pressure.

Known Constraints

  • An AI-drafted engagement letter is a starting point, not a final document. Have your professional liability insurance carrier or legal counsel review any new engagement letter template before using it in practice.
  • Engagement letter standards vary by service type. A compilation engagement letter has different required language than a tax preparation engagement letter under AICPA standards. Make sure the draft reflects the applicable professional standards for the service.
  • The exclusions section is where most engagement disputes start. Read it carefully and make sure every service you are not providing is explicitly listed. 'Bookkeeping does not include tax preparation' sounds obvious — until a client assumes it is included.
  • AI does not know your state's specific CPA licensing requirements, mandatory disclosure language, or fee disclosure rules. Review the draft against your state board's requirements before finalizing.

Security Deployment

No client data is involved in this workflow. You are describing your firm, your service scope, and your fee structure — not the client's financial information. Free AI platforms are appropriate here. Leave [CLIENT NAME] and [CLIENT ENTITY] as placeholders in the prompt output and fill them in after the letter is finalized.

Claude.ai (free)Free
ChatGPT (free)Free
Microsoft CopilotFree / included with M365
09
Tier 2: Anonymize

GL Entry Anomaly Review

Flag unusual journal entries in a specific account before the audit or close.

general ledgeraudit prepjournal entriesinternal controlsanomaly detection

The Problem

Reviewing a general ledger account for unusual entries — odd amounts, off-cycle timing, entries with no memo, large round-number postings, or credits in an account that normally only has debits — is one of the most tedious tasks in accounting. Done manually, a thorough review of a single account takes 20–40 minutes. AI scans a GL detail report for anomaly patterns in under 2 minutes and returns a prioritized list of entries worth investigating.

What You Need

  • GL detail report for the account or accounts you want to review (QuickBooks, Xero, Sage, or your GL system)
  • Date range matching the period under review — at minimum one full fiscal year
  • Remove the client's company name from the report header before exporting
  • Claude for Work, ChatGPT Team, or Ollama — see security note before you start

Execution Roadmap

1

Export the GL detail for the target account

In QuickBooks Online: Reports → Transaction Detail by Account. Select the account and set the date range. Include columns: Date, Transaction Type, Reference Number, Memo, Amount, Balance. Before copying, replace the client company name in the header with 'Client A.' The transaction amounts and dates contain no personally identifiable information.

2

Specify what kind of anomalies matter for this account

Not all anomalies are equally relevant for every account type. A large round-number debit in a legal expense account means something different than the same entry in a payroll account. Tell AI the account type and what normal activity looks like for this account — this improves the relevance of the flags.

3

Investigate flagged entries in the source system

For each flagged entry, open the transaction in QuickBooks and review: who entered it, what it was coded to, whether it has supporting documentation attached, and whether the memo explains the posting. Entries with no memo and no documentation are the highest priority for follow-up, especially in asset accounts or near period-end.

The Engineered Prompt

Paste directly into any AI chat
You are helping a CPA review a general ledger account for unusual or anomalous journal entries.

Account details:
- Account name: [e.g., Legal & Professional Fees / Owner's Draw / Other Expenses / Prepaid Assets]
- Account type: [Expense / Asset / Liability / Equity]
- Period: [e.g., January 2025 – December 2025]
- What normal activity looks like: [e.g., "Monthly payments to 2 law firms, typically $1,500–$4,000 each" / "Weekly owner draws of $3,000" / "Occasional vendor invoices, usually under $2,500"]
- Review purpose: [e.g., Audit prep / Year-end close / Internal controls review / Suspected misclassification]

Below is the GL detail for this account (client name removed). Columns: Date | Type | Ref # | Memo | Amount | Running Balance

Please flag entries that match any of these anomaly patterns:
1. Round-number amounts over $1,000 (e.g., $5,000, $10,000, $25,000)
2. Entries with a blank or generic memo (e.g., 'Adjustment,' 'Correction,' 'See attached')
3. Entries posted on the last 1–3 days of a month or quarter
4. Credits in an account that normally has only debits (or vice versa)
5. Amounts significantly larger than the typical transaction range I described
6. Multiple entries on the same date to the same account from different sources

For each flagged entry:
— List: Date | Ref # | Memo | Amount | Anomaly Pattern
— Assign risk level: High (multiple flags or near period-end), Medium (single flag), Low (minor deviation)
— Add a one-line question to investigate in the source system

Sort by risk level, highest first.

[PASTE YOUR GL DETAIL REPORT HERE]

Expected Outcome

AI returns a prioritized list of flagged GL entries with anomaly pattern labels, risk levels, and investigation questions for each. A 30-minute manual scan becomes a 10-minute review of flagged items.

S

Practitioner Perspective

Sydney Smart, CPA, MST

I use this most for accounts where I know something looks off but cannot identify the entry quickly — legal expenses that are higher than usual, owner draws that don't match the regular schedule, or a prepaid account with a balance that doesn't clear the way it should. AI does not replace the judgment call on what is actually wrong, but it gets me to the right 6 entries out of 200 instead of reading all 200 looking for the 6.

Known Constraints

  • AI flags patterns — it cannot tell you whether a flagged entry is an error, fraud, or a legitimate transaction that happens to look unusual. Your investigation in the source system is the next step for every flag.
  • Period-end entries are almost always flagged. Accruals, adjusting entries, and closing entries are by nature posted on the last days of the period. Review flagged period-end entries separately from the rest — most will be expected.
  • Blank memo fields are the highest-value flags because they represent entries that cannot be self-documented. If a flagged entry has no memo and no attached document, that is a real gap regardless of whether the dollar amount is unusual.
  • AI reviews the transactions you give it. If the GL export is missing a transaction type (manual journal entries excluded, for example), those entries will not be flagged. Confirm your export includes all transaction types for the account.

Security Deployment

GL detail reports contain transaction amounts, reference numbers, and memo text — no employee SSNs or client personally identifiable information in most cases. Replace the client company name in the report header before pasting. If any memos contain individual names, Social Security numbers, or account numbers, remove those before pasting — that would escalate this to a red-tier workflow. Use a paid business AI tier or Ollama locally.

Claude for Work$25/user/mo
ChatGPT Business$25/user/mo
Microsoft Copilot M365Included with most M365 Business plans
Ollama (local)Free
10
Tier 2: Anonymize

Payroll Variance Explanation

Turn a 40-minute write-up into a 90-second review.

payrollQuickBooksmonth-endvarianceownership reporting

The Problem

Every month-end close, someone asks why payroll changed. Writing the explanation from scratch — staring at two months of numbers, calculating percentages, drafting something a business owner can actually read — takes 30–45 minutes. AI produces the first draft in under 90 seconds. Your job becomes reviewing it, not writing it.

What You Need

  • QuickBooks Online with Payroll (or any payroll platform with a summary export)
  • Two months of Payroll Summary reports — use 'Group By: Totals Only' inside the report to remove employee names
  • Claude for Work, ChatGPT Team, or Ollama — see security note before you start
  • 10 minutes the first time. 3 minutes every month after.

Execution Roadmap

1

Export the right report

In QuickBooks Online: go to Reports and search for 'Payroll Summary.' Set the date range to the current month. Under the 'Group By' dropdown inside the report, select 'Totals Only' — this removes individual employee rows from the export. What remains is category-level data: wages, taxes, benefits, deductions, with no personally identifiable information attached. Export to Excel or copy the table. Repeat for the prior month.

2

Paste this prompt

Fill in the context field before you hit send. Without it, AI writes a technically accurate but generic explanation. With it, AI writes the right one.

3

Review, adjust, send

Read the owner summary aloud. If it sounds like a press release, edit one or two sentences. The math is done. The structure is done. Your job is making sure the explanation matches what you know about the business.

The Engineered Prompt

Paste directly into any AI chat
You are helping a CPA prepare a payroll variance explanation for a client's ownership group.

Below are two months of payroll summary data in this format:
Payroll Category | [Prior Month] Total | [Current Month] Total

Context: [ADD WHAT YOU KNOW — examples: "We hired 2 employees mid-month." / "One employee took unpaid leave." / "Q2 bonuses were paid this month." / "No planned changes — this variance is unexpected."]

Your tasks:
1. Calculate the dollar and percentage change for each category
2. Identify the top 3 variance drivers by dollar amount
3. Write a 3–5 sentence plain-English explanation for a business owner who is not an accountant
4. Flag any line items I should verify before sending

Format your response:
— Top drivers (bulleted, largest first, dollar + %)
— Owner summary (plain English paragraph, no jargon)
— Verify before sending (bulleted)

[PASTE YOUR TWO MONTHS OF DATA HERE]

Expected Outcome

Claude returns variance drivers ranked by dollar amount, a clean owner-ready paragraph, and a short list of items to double-check. A 40-minute task becomes a review task.

S

Practitioner Perspective

Sydney Smart, CPA, MST

This workflow earns its time savings when the variance has a clear driver — a new hire, a bonus cycle, a benefits change. Where it falls short is when the numbers don't tell the whole story on their own. Use this to write the explanation faster. Don't use it as a substitute for understanding what actually happened. Those are two different jobs.

Known Constraints

  • AI works from what you give it. If a line item is miscategorized in QuickBooks, the explanation will repeat that error. Review the category totals before pasting.
  • The context field is not optional. Without it you get a description of numbers. With it you get an explanation.
  • AI cannot tell you whether a variance is a problem or expected. It explains what the numbers say. The judgment is yours.
  • If the variance has no obvious cause in the summary report, dig into the detail reports first. AI cannot find what you did not give it.

Security Deployment

The Total Only column setting removes employee names. What you are pasting is wage categories and dollar totals — not a personnel record. Do not use free ChatGPT or free Claude.ai — both may use your inputs for model training by default. Safe options: Claude for Work ($25/user/mo), ChatGPT Team ($30/user/mo), Microsoft Copilot M365 (included with most Business plans), or Ollama (free, local, data never leaves your machine). If you need employee-level detail with names or individual salaries, that is a red-tier workflow — use Ollama locally or obtain explicit client consent first.

Claude for Work$25/user/mo
ChatGPT Business$25/user/mo
Microsoft Copilot M365Included with most M365 Business plans
Ollama (local)Free
11
Tier 2: Anonymize

1099 Vendor Eligibility Review

Flag which vendors need a 1099 before the January deadline, without reviewing every line manually.

1099year-endvendoraccounts payableQuickBooks

The Problem

At year-end, most small firms have hundreds of vendor payments to review for 1099 eligibility. QuickBooks flags some automatically, but it misses payments categorized to the wrong accounts, vendors set up as corporations when they should not be and cash payments that did not run through the normal AP workflow. A full manual pass takes 90 minutes or more and still leaves gaps.

What You Need

  • QuickBooks Online: Reports → Transaction List by Vendor, date range: full calendar year. Include: Vendor, Transaction Type, Account, Amount.
  • Filter to exclude credit card payments — QBO excludes these from 1099s automatically. Include checks, ACH and bill payments.
  • Remove the client company name from the report header before copying.
  • Have the client vendor W-9 file or notes on entity type (individual, LLC, corporation) nearby for cross-reference.

Execution Roadmap

1

Export the vendor transaction summary

In QuickBooks Online, go to Reports and search Transaction List by Vendor. Set the date range to January 1 through December 31. Include columns: Date, Transaction Type, Vendor, Account, Amount. Filter out credit card payment types — QBO excludes those from 1099 reporting automatically. Before copying, replace the client company name in the report header with Client A.

2

Paste and ask AI to flag eligibility

Tell AI the payment threshold is $600 and ask it to flag vendors who: received $600 or more in non-employee compensation, rents or attorney fees during the year; appear to be individuals or single-member LLCs based on name format; and were paid via check or ACH, not credit card. Ask it to separate clearly ineligible vendors (incorporated names ending in Inc., Corp. or Ltd.) from uncertain ones that need W-9 verification.

3

Cross-reference flags against W-9s and QBO settings

Pull up the flagged vendors in QBO under Expenses → Vendors. Check the Track payments for 1099 checkbox and confirm the entity type matches what you have on file. For uncertain vendors, compare against W-9s in your files. Any vendor flagged by AI but missing a W-9 should receive a W-9 request before January 15.

The Engineered Prompt

Paste directly into any AI chat
You are helping a CPA review vendor payments for 1099-NEC and 1099-MISC eligibility.

Below is a vendor transaction list for [YEAR] (client name removed). Columns: Date | Transaction Type | Vendor | Account | Amount

Review the list and do the following:

1. Total all payments by vendor for the year.
2. Flag vendors who received $600 or more total who appear to be individuals, sole proprietors or LLCs based on name format — no Inc., Corp., Ltd. or clearly national brands.
3. List vendors who are clearly ineligible (incorporated entities, national chains, utilities) and why.
4. List vendors where eligibility is uncertain and explain what information is needed to confirm.
5. Note any vendors with payments categorized to rent, legal fees or medical payments — those have different 1099 type rules.

Format the output as three sections: LIKELY NEED 1099 | CLEARLY INELIGIBLE | NEEDS VERIFICATION

For each flagged vendor: Vendor Name | Total Paid | Reason for Flag

[PASTE YOUR VENDOR TRANSACTION LIST HERE]

Expected Outcome

AI returns three sorted lists: vendors likely requiring a 1099, vendors clearly exempt and vendors needing W-9 verification. A 90-minute manual review becomes a 20-minute verification task focused only on the uncertain cases.

S

Practitioner Perspective

Sydney Smart, CPA, MST

Run this in mid-December, not January. You want time to chase W-9s before the deadline. The uncertain list is where the work actually is — most vendors are obvious once AI separates them out. The ones that are not obvious are usually vendors who have been around for years without anyone ever asking for a W-9.

Known Constraints

  • AI reads vendor names to guess entity type — it cannot verify actual tax classification. Smith Consulting LLC might be a disregarded entity that needs a 1099 or a C-corp that does not. Every flagged vendor still requires W-9 confirmation.
  • QBO sometimes creates duplicate vendor records for the same payee under slightly different names. AI will treat Acme Plumbing and Acme Plumbing Co. as separate vendors. Review payment totals for near-duplicate vendor names before submitting to AI.
  • Payments categorized to the wrong expense account may be included or excluded incorrectly. If a contractor payment was accidentally coded to Materials instead of Contract Labor, it may not appear in the transaction list at all.
  • Attorney fees require a 1099-MISC box 10 even when paid to a corporation — AI will flag law firms as clearly ineligible unless you specifically instruct it to separate legal services payments.
  • State 1099 filing requirements vary. Some states require 1099 copies with lower thresholds or different forms. AI will not flag state-specific requirements unless you include your state rules in the prompt.

Security Deployment

The vendor transaction list contains vendor names, payment amounts and account classifications — no Social Security numbers or EINs unless W-9 details are included in the export. Remove the client company name from the report header before uploading. Do not paste vendor W-9 information into the AI prompt — that escalates this to red tier. Use a paid business AI tier (Claude for Work, ChatGPT Business, Microsoft Copilot M365) or Ollama locally. Free consumer tools are not appropriate for client transaction data under AICPA Rule 301.

Claude for Work$25/user/mo
ChatGPT Business$25/user/mo
Microsoft Copilot M365Included with most M365 Business plans
Ollama (local)Free
12
Tier 2: Anonymize

AR Collection Email Drafts

Turn your AR aging report into tiered collection emails without writing each one from scratch.

accounts receivablecollectionsAR agingQuickBooksclient communication

The Problem

When the AR aging report shows 15 clients past due, most firms send the same generic reminder to everyone — or nothing at all because writing individual emails takes too long. A 30-day past-due client needs a different tone than one at 90 days. Writing tiered personalized emails for each situation takes 45 minutes. Most of that time is spent on tone, not content.

What You Need

  • QuickBooks Online: Reports → Accounts Receivable Aging Summary. Run as of today.
  • Customize columns: Customer, Invoice Number, Invoice Date, Amount Due, Days Past Due.
  • Export to Excel or copy the table directly.
  • Remove your firm name from the report header before copying.
  • Have your payment terms and accepted payment methods handy.

Execution Roadmap

1

Export the AR aging summary

In QuickBooks Online, go to Reports and search Accounts Receivable Aging Summary. Run it as of today. Customize to show customer name, invoice number, invoice date and balance by aging bucket: Current, 1-30, 31-60, 61-90, 90+. Export to Excel or copy the table. Replace your firm name in the report header with Firm A before pasting.

2

Paste and specify the tone for each bucket

Tell AI your firm name, payment terms (e.g., Net 30) and accepted payment methods (check, ACH, credit card, online portal URL). Ask it to draft one email per client in the aging report, with tone calibrated by bucket: friendly reminder at 1-30 days, direct and specific at 31-60 days and firm with escalation language at 61+ days. Ask it to include the invoice number and amount in each email.

3

Review and send from your email client

AI returns a draft email for each client grouped by aging tier. Review each one for tone and accuracy — especially anything at 61+ days where escalation language should reflect your actual relationship with that client. Paste the final version into your email client, add the client name back in and send. You are reviewing and sending, not writing.

The Engineered Prompt

Paste directly into any AI chat
You are helping an accounting firm draft collection emails for past-due invoices.

Firm name: [FIRM NAME]
Payment terms: [e.g., Net 30]
Accepted payment methods: [e.g., check, ACH, online payment portal at URL]

Below is an accounts receivable aging summary (client names are placeholders — I will swap them back before sending). Columns: Customer | Invoice # | Invoice Date | Amount Due | Days Past Due

[PASTE AGING REPORT HERE]

Draft one collection email per customer. Calibrate tone by aging tier:

1-30 days: Friendly, assumes oversight. Remind, include invoice details, provide payment options.
31-60 days: Direct. Reference that payment was expected by [DATE]. Ask for payment or a status update within 3 business days.
61-90 days: Firm. Note that the account is significantly past due. Request payment or a call to discuss. Mention that continued non-payment may affect future services.
90+ days: Escalation. State clearly that the account requires immediate attention. Ask for payment in full or a payment plan discussion within 5 business days.

Format each email with: Subject line | Body | [CUSTOMER PLACEHOLDER] so I can find and replace the name.

Group emails by tier.

Expected Outcome

AI returns one draft email per customer grouped by aging tier with subject lines included. A 45-minute writing task becomes a 10-minute review and send.

S

Practitioner Perspective

Sydney Smart, CPA, MST

The 31-60 day tier is where this actually saves time. Everyone knows how to write a gentle reminder and everyone knows what a final notice looks like. The middle-tier email is the one nobody wants to write because the tone is hard to calibrate. Let AI draft it, then adjust for the relationship.

Known Constraints

  • AI does not know your actual client relationship. A 90-day past-due client who is also your largest account needs a different tone than an unknown client at 30 days. Review the escalation-tier emails carefully before sending.
  • The aging report shows balances, not context. AI will not know that a client disputed an invoice, is waiting on a credit memo or had a verbal payment arrangement. Cross-reference flagged clients against any open disputes before sending.
  • If the aging report groups multiple invoices per client into one line, AI may not have enough detail to reference specific invoice numbers. Use the Accounts Receivable Aging Detail report instead of the summary for more granular data.
  • Email tone for collections varies by industry and relationship. The default escalation language AI uses may feel too aggressive for professional services clients where the relationship matters. Soften the 61+ bucket if your firm values long-term retention over immediate collection.
  • This workflow produces draft emails — it does not send them. Build in a partner review step for anything in the 61+ tier before it goes out.

Security Deployment

The AR aging report contains client names and invoice amounts. Remove your firm name from the report header before uploading. Use customer placeholder labels if you want to further anonymize, then swap real names back in before sending. Do not include client Social Security numbers, EINs or tax identification data in this workflow. Use a paid business AI tier or Ollama locally. Free consumer tools are not appropriate for client financial data under AICPA Rule 301.

Claude for Work$25/user/mo
ChatGPT Business$25/user/mo
Microsoft Copilot M365Included with most M365 Business plans
Ollama (local)Free
13
Tier 2: Anonymize

Expense Report Policy Audit

Scan submitted expense reports for policy violations before reimbursing.

expense reportsinternal controlsreimbursementConcurQuickBooks

The Problem

Reviewing expense reports for policy compliance before approval is tedious and inconsistent. Most reviewers catch obvious issues — a $400 dinner for two — but miss subtler ones: missing receipts on expenses just under the receipt threshold, the same meal expensed by both a manager and a direct report or a pattern of weekend charges from someone who should not be traveling on weekends. A manual audit of 30 line items takes 20 minutes and depends heavily on who is reviewing.

What You Need

  • Export the expense report from your platform: Concur, Ramp, Emburse, QuickBooks or a CSV/Excel file.
  • Include columns: Date, Employee Name or ID, Merchant, Category, Amount, Receipt Attached (Y/N), Notes.
  • Have your firm expense policy ready — specifically: per diem limits by category, receipt threshold, approved categories and travel day rules.
  • Replace employee names with IDs (Employee 1, Employee 2) before uploading if reviewing a multi-employee report.

Execution Roadmap

1

Export the expense report and note your policy rules

Pull the expense report from your system — Concur, Ramp, Emburse or export from QuickBooks. Make sure the export includes date, merchant, category, amount and whether a receipt is attached. Before pasting, replace employee names with Employee 1, Employee 2 to reduce identifiable data. Note your key policy limits: meal per diem, receipt threshold, restricted categories and whether weekend travel requires pre-approval.

2

Paste the report and your policy rules

Give AI your policy rules first, then paste the expense report. Be specific: Meals $75 per person per day. Receipts required for all expenses over $25. No alcohol reimbursed. Weekend travel requires manager pre-approval. Ask AI to flag every line that appears to violate a rule or warrants a closer look, and explain why.

3

Review flagged items before approval

AI returns a list of flagged items with a reason for each flag. Pull up the original receipts or documentation for each flagged line. Most flags will have an explanation — a working dinner that went over or a client meal with a note. The ones without a clear explanation are the ones to hold before approving reimbursement.

The Engineered Prompt

Paste directly into any AI chat
You are helping an accounting manager audit expense reports for policy compliance before reimbursement.

Expense policy rules:
[PASTE YOUR POLICY RULES HERE — e.g., Meals: $75/person/day, Receipts required over $25, No alcohol, Weekend travel requires pre-approval]

Below is an expense report for review. Employee names have been replaced with IDs. Columns: Date | Employee | Merchant | Category | Amount | Receipt (Y/N) | Notes

[PASTE EXPENSE REPORT HERE]

For each expense line, flag any that:
1. Exceed per diem or category limits
2. Are missing a receipt where one is required
3. Fall into a restricted category (alcohol, personal entertainment, etc.)
4. Appear on a weekend or holiday without an explanation
5. Have the same date, merchant and similar amount to another line — possible duplicate
6. Have no merchant name or description

For each flag:
— Show the line: Date | Employee | Merchant | Amount
— State the policy rule it may violate
— Assign: HOLD (requires documentation before approval) or REVIEW (explain or verify)

If no flags, confirm the review was completed and list the policy rules checked.

Expected Outcome

AI returns a flagged list organized by issue type with each item labeled HOLD or REVIEW. A 20-minute manual audit becomes a 5-minute exception review focused only on the flagged lines.

S

Practitioner Perspective

Sydney Smart, CPA, MST

The most useful part of this is not catching the obvious violations — a reviewer would catch those. It is the pattern flags: the same employee with missing receipts every month or two people expensing the same dinner. Those patterns are hard to see one report at a time.

Known Constraints

  • AI can only detect patterns in the data you give it. If the receipt column says Y but the receipt is fraudulent or for a different amount, AI has no way to flag it. This workflow audits for policy compliance, not receipt authenticity.
  • Duplicate detection works on exact or near-exact matches. If a manager and a direct report both expensed the same client dinner, they will have the same date and merchant but different amounts. AI will flag same-date, same-merchant entries but will not know they are the same event without context.
  • Category misclassification is common — a team lunch coded as Office Supplies will pass the meal per diem check. AI audits categories as submitted; it does not reclassify.
  • This workflow is designed for pre-approval review. If you are auditing already-reimbursed reports, any recoveries require a separate conversation with the employee — AI flags will not automatically reverse payments.
  • Per diem limits vary by city for travel expenses. If your policy uses city-specific rates, include those in your policy rules section or AI will apply your standard rate to all locations.

Security Deployment

Expense reports contain employee names and spending data. Replace employee names with IDs before uploading. The report should not contain Social Security numbers, payroll information or direct deposit details — if your export includes those fields, remove them before pasting. Use a paid business AI tier or Ollama locally. Free consumer tools are not appropriate for employee data under AICPA Rule 301.

Claude for Work$25/user/mo
ChatGPT Business$25/user/mo
Microsoft Copilot M365Included with most M365 Business plans
Ollama (local)Free
14
Tier 2: Anonymize

Tax Projection Client Letter

Turn a tax projection into a plain-English client letter in under 5 minutes.

tax projectionclient communicationestimated taxyear-end planningtax letter

The Problem

Most clients receive their tax projection as a spreadsheet or a PDF with numbers and no explanation. They do not know what the numbers mean, why their tax changed from last year or what they need to do before December 31. Writing a clear personalized explanation takes 20 minutes per client and usually gets deprioritized during busy season — so clients call later asking the same questions the letter would have answered.

What You Need

  • Complete the tax projection in your software (Lacerte, Drake, ProConnect or Excel) and note: prior year tax, projected current year tax, payments made to date, balance due or refund projected, Q4 estimated payment amount and due date.
  • Note the primary driver of any change from prior year (income increase, deduction loss, rate change, new income source).
  • Do NOT include the client Social Security number or EIN in the draft input — use their first name only.
  • Have the next quarterly due date handy: April 15, June 16, September 15 or January 15, 2027.

Execution Roadmap

1

Pull the key numbers from your projection

From your tax software or projection model, note six numbers: prior year total tax, projected current year total tax, estimated payments made so far, remaining balance due or refund expected, recommended Q4 estimated payment and the next due date. Also note the primary driver of the change — an income increase, a large deduction that went away or a new filing situation like a business sale or Roth conversion.

2

Paste the numbers and ask for a client letter

Give AI the client first name, the six projection numbers, the change driver and your firm name. Ask for a letter that explains what the numbers mean in plain English, why the tax changed from last year, what the client needs to do and by when and a one-line note on what to watch for before year-end. The letter should read like it was written by someone who knows the client, not a form letter.

3

Review for accuracy and send on letterhead

Read the letter once for factual accuracy — AI can misstate a number if the input was ambiguous. Check that the payment amount and due date are correct. Paste onto firm letterhead, add the client mailing address and send. The letter should take under 5 minutes from paste to send.

The Engineered Prompt

Paste directly into any AI chat
You are helping a CPA draft a plain-English tax projection letter for a client.

Client first name: [FIRST NAME]
Firm name: [FIRM NAME]
Letter date: [DATE]

Projection data:
- Prior year total tax: $[AMOUNT]
- Projected current year total tax: $[AMOUNT]
- Estimated payments made to date: $[AMOUNT]
- Remaining balance due (or refund expected): $[AMOUNT]
- Recommended Q4 estimated payment: $[AMOUNT]
- Q4 due date: [DATE — e.g., January 15, 2027]

Primary reason tax changed from prior year: [e.g., rental income increased by $40,000 or mortgage interest deduction was lost or Roth conversion of $75,000]

Any year-end action items: [e.g., consider charitable contribution before December 31 or none at this time]

Write a 3-4 paragraph letter that:
1. Opens warmly and states the purpose in plain English
2. Explains what the projection shows and why it changed from last year — no tax jargon
3. States clearly what the client needs to do and by when (payment amount and due date)
4. Closes with one sentence on any year-end opportunity or a simple reach-out invitation

Do not use the words herein, aforementioned, pursuant or enclosed please find. Write like a person, not a form letter.

Expected Outcome

AI returns a 3-4 paragraph plain-English client letter ready to paste on letterhead. A 20-minute writing task per client becomes a 3-minute review.

S

Practitioner Perspective

Sydney Smart, CPA, MST

Send this in October or November, not December. Clients who get the letter in October have time to act on it. December letters arrive when most clients have already spent the money. The letter itself is 10% of the value — the timing is the other 90%.

Known Constraints

  • AI uses only the numbers you provide. If you enter the projected tax incorrectly or confuse balance due with total tax, the letter will state the wrong figures. Read the letter once specifically for number accuracy before sending.
  • Safe harbor rules differ by income level. Clients with prior year AGI over $150,000 need to pay 110% of prior year tax to avoid underpayment penalties — AI will not apply this distinction unless your payment recommendation already accounts for it.
  • State estimated tax rules and due dates vary significantly. This prompt produces a federal projection letter. If the client has material state liability, add the state numbers and due dates explicitly — do not assume AI knows your state rules.
  • This letter is for informational client communication, not a formal tax opinion or engagement deliverable. Use it as a cover letter alongside your formal projection, not instead of it.
  • Year-end planning opportunities require your judgment, not AI. What to recommend — Roth conversion, charitable giving, income deferral — depends on the client full picture, which AI does not have.

Security Deployment

This workflow uses the client first name and projection numbers only — no Social Security number, EIN or full tax return data. Do not paste a PDF of the client tax return or a full Lacerte or Drake report into the prompt. If your projection export includes SSN or EIN fields, remove them before copying. Use a paid business AI tier or Ollama locally. Free consumer tools are not appropriate for client financial data under AICPA Rule 301.

Claude for Work$25/user/mo
ChatGPT Business$25/user/mo
Microsoft Copilot M365Included with most M365 Business plans
Ollama (local)Free
15
Tier 2: Anonymize

New Client Intake Summary

Turn onboarding questionnaire responses into a structured client profile before the first meeting.

onboardingnew clientintakeclient profilepractice management

The Problem

After a new client submits their intake questionnaire, someone on the team has to read through it, extract the key information and turn it into a usable reference. Entity type, filing requirements, fiscal year, current software, open questions — it is all in there but buried in free-text responses and inconsistent formatting. That first-pass extraction takes 30 minutes and produces a document nobody looks at again.

What You Need

  • Export or copy the completed intake questionnaire — from Typeform, Google Forms, Practice Ignition, Karbon or a PDF or email response.
  • Do not include prior year tax returns or financial statements in this prompt — that is a separate higher-security workflow.
  • Have your firm standard service list and pricing tiers available for cross-reference.
  • Note any verbal commitments from the intake call that are not captured in the written questionnaire.

Execution Roadmap

1

Copy the completed questionnaire responses

Pull the client intake questionnaire from your intake platform — Practice Ignition, Karbon, Typeform, Google Forms or a PDF email response. Copy the full text of their answers. You do not need to clean it up or reformat it. Leave the client name in — you are creating an internal reference document. Do not include prior year tax returns, financial statements or any document with Social Security or EIN numbers in this step.

2

Paste and ask for a structured client profile

Tell AI this is a new client intake questionnaire for an accounting firm and ask it to extract a structured profile. Specify the fields you want: entity type and filing requirements, fiscal year, current accounting software, services requested, key contacts, known pain points or open questions and any flags that need follow-up before the first meeting.

3

Review and save to your practice management system

AI returns a clean structured profile. Review it for accuracy — free-text questionnaire answers sometimes use ambiguous language AI will interpret literally. Add any notes from the intake call, then paste the profile into your practice management system (Karbon, Canopy, Financial Cents or your CRM). Use it as the reference document for the first client meeting.

The Engineered Prompt

Paste directly into any AI chat
You are helping an accounting firm create a structured client profile from a new client intake questionnaire.

This is an internal reference document. Extract the following fields from the questionnaire responses below:

1. Client name and primary contact
2. Entity type (sole proprietor, LLC, S-corp, C-corp, partnership, individual)
3. State of formation and states of operation
4. Fiscal year (calendar year or other — note month-end if other)
5. Current accounting software and version
6. Services requested (bookkeeping, tax prep, payroll, CFO advisory, etc.)
7. Current situation: what is working, what is not and why they are changing firms
8. Key deadlines or urgent items mentioned
9. Open questions or items that need follow-up before the first meeting
10. Any red flags noted (e.g., behind on filings, unresolved IRS issues, books not current)

Format as a clean reference document with labeled sections. If a field is not answered or unclear, mark it as [NOT PROVIDED — follow up].

Questionnaire responses:
[PASTE INTAKE QUESTIONNAIRE RESPONSES HERE]

Expected Outcome

AI returns a structured one-page client profile with labeled fields and flagged gaps. A 30-minute extraction task becomes a 5-minute review before the first meeting.

S

Practitioner Perspective

Sydney Smart, CPA, MST

The red flags section is the most useful part. Clients almost never say directly that they have unfiled returns, but they do say things like we have been a little behind on our taxes or we had a disagreement with our last CPA. AI will flag those for follow-up. That is the conversation you want to have in the first meeting, not after you have started the work.

Known Constraints

  • Clients fill out questionnaires inconsistently. We use QuickBooks with no version specified or we are an LLC without clarifying the tax election are common. AI will mark these as incomplete but you need to follow up — it will not guess.
  • The intake questionnaire captures what the client thinks is relevant. It will not surface issues they do not know about — unfiled returns, misclassified workers, sales tax exposure. The structured profile is a starting point, not a risk assessment.
  • Entity type answers are often wrong. A client who says we are an S-corp may be an LLC with an S-corp tax election, a C-corp that should have elected S status years ago or a sole proprietor who formed an LLC but never filed Form 2553. Verify entity type against formation documents before relying on the profile.
  • This workflow is for the intake questionnaire only — not prior year returns or financial statements. Those documents contain SSNs and EINs and require red-tier handling. Keep them out of this prompt.
  • Practice management tools vary in how they accept structured data. A formatted AI profile may not import cleanly into Karbon or Canopy field-by-field — you may need to copy sections manually.

Security Deployment

The intake questionnaire typically contains the client name, contact information and business details — but not Social Security numbers, EINs or financial statements. Keep it that way: do not include prior year tax returns, financial statements or payroll records in this prompt. If the client submitted supporting documents with their questionnaire, review those separately. Use a paid business AI tier or Ollama locally. Free consumer tools are not appropriate for client business information under AICPA Rule 301.

Claude for Work$25/user/mo
ChatGPT Business$25/user/mo
Microsoft Copilot M365Included with most M365 Business plans
Ollama (local)Free
16
Tier 3: Local / Enterprise

AR Risk Summary via AI Assistant

Ask an AI assistant connected to live AR data for a quarter-end risk summary instead of pulling and formatting an aging report by hand.

accounts receivableAI agentMCP connectordata access controlscollectionsAR aging

The Problem

Vendors are now wiring live accounts receivable data straight into general AI assistants like Claude and Microsoft Copilot through MCP connectors, so a CFO or controller can ask for an AR risk summary instead of exporting and formatting an aging report before a quarter-end call. That is a real time save. It also means client-named financial data now moves through a channel most firms have not reviewed, and the same feature name can mean read-only access for one user and collections-outreach access for another.

What You Need

  • Confirm with your AP/AR platform admin, or the client's, whether an AI-assistant connector is already enabled and for which users.
  • Get the connector's permission scope in writing: read-only access versus write-enabled actions like collections outreach or payment application.
  • Identify which staff accounts currently have access and whether that list matches who should have it.
  • Confirm the connector runs through an enterprise-provisioned seat, not a personal or free AI account.

Execution Roadmap

1

Ask for the risk summary, not the raw account list

Request a summary, not a data dump: total outstanding, aging buckets, and the largest at-risk balances. Asking the assistant to "list everything" pulls more customer-level detail into the conversation than a summary needs.

2

Check what permission level actually answered you

If the assistant offers to draft or send a collections email, that is a write action, not the read-only summary you expected. Stop and confirm your firm or the client actually enabled that before it happens once by accident.

3

Verify one number against the source platform

Before repeating a figure in a client conversation, log into the AR platform directly and confirm the total outstanding and the largest balance match. The connector's sync schedule may not be as real-time as the summary implies.

4

Log the query if it feeds a client deliverable

If the AR risk summary ends up in a client report or board memo, note that it was AI-assisted from live connector data, and who reviewed it, per your firm's AI documentation practice.

The Engineered Prompt

Paste directly into any AI chat
Context: [client or company name], AR platform: [Billtrust / NetSuite / QuickBooks, etc], as-of date: [date]

Summarize accounts receivable risk for [client name] as of [date]. Include:
1. Total outstanding AR
2. Aging buckets (0-30, 31-60, 61-90, 90+ days)
3. The five largest outstanding balances by customer
4. Any accounts that moved into a worse aging bucket since [prior date]

Do not take any action on any account. This is a read-only summary request.

Expected Outcome

AI returns a quarter-end AR risk snapshot in seconds instead of the 20 to 30 minutes it takes to pull and format an aging report by hand. It replaces the manual pull, not the judgment call about which accounts actually need a phone call.

S

Practitioner Perspective

Sydney Smart, CPA, MST

This is worth testing before a client asks about it, not after. The read-only summary is genuinely useful heading into a quarter-end call. What I check every time: who actually has this connector turned on, and whether the number it just gave me matches the platform itself. A sync lag has a way of turning an innocent AI answer into an awkward client conversation.

Known Constraints

  • Read-only access and the ability to draft or send collections outreach are sometimes bundled under one feature name by the vendor. Confirm which is actually turned on for which user, not just what the marketing copy implies.
  • The connector syncs on the vendor's schedule, not always truly real-time. A summary marked "as of today" may reflect last night's batch. Confirm the sync cadence before quoting a number in a client meeting.
  • A general-purpose assistant does not know your firm's materiality threshold or which client relationships are politically sensitive. It will flag the largest dollar balance, not necessarily the account that matters most.
  • When the assistant references a customer by name in a shared or logged workspace, that is client financial data moving through a new channel. Confirm your data processing agreement actually covers this specific connector, not just the base AI platform contract.
  • Giving several staff members assistant access to the same live financial data changes who can pull sensitive client information, even if nobody changes a single number. Your access review needs to include this connector, not just logins to the AR platform itself.

Security Deployment

This is red tier: live, customer-named accounts receivable data moving through a general AI assistant. Free consumer accounts are never appropriate here, and personal business logins are not either. This needs an enterprise-provisioned Claude or Copilot seat tied to a signed data processing agreement that specifically names the AR connector, not just the underlying model. Before this workflow goes live for any client, get the connector's permission scope in writing and confirm access matches your firm's authorization list. Under AICPA Rule 301, client financial data leaving the practitioner's direct control still requires client consent and a vendor that has committed in writing not to train on it.

Claude for Enterprise (admin-provisioned)Custom enterprise pricing
Microsoft Copilot M365 (tenant-scoped)Included with qualifying M365 plans
Vendor-native AI connector with signed DPAVaries by platform
The Nexairi Dispatch

New Workflows. Every Week.

New practitioner-reviewed workflows for tax planning, audit automation, and firm ops — delivered every Tuesday. Subscribe to stay ahead.

Join Nexairi Dispatch

AI is showing up everywhere you live and work. A short dispatch, 3x per week, so you see it coming.

By subscribing, you accept our Terms & Privacy Policy. Unsubscribe anytime.