Excel Copilot for Accountants
Practical workflows for faster workbooks—without giving up accounting judgment or control.
Educational content — not software advice or a substitute for professional judgment. Microsoft features, names, licensing, and interfaces change frequently. Soft-label vendor claims on this page as vendor documentation unless independent evidence is cited. Confirm behavior in the Excel version your organization uses before relying on a specific capability.
Product freshness: Updated September 2026. Copilot names, interfaces, licensing, and availability change often. Some releases previously marketed direct editing as Agent Mode; that naming may still appear in some UIs even when the capability is folded into ordinary editing with Copilot.
Why This Matters
Copilot can help an accountant create and explain formulas, highlight exceptions, summarize data, build PivotTables and charts, and complete multi-step workbook edits. Microsoft documentation also instructs users to review, edit, and verify what Copilot creates—because it can produce inaccurate or inappropriate content.
The right goal is not to let AI "do the accounting." The goal is to use Copilot as a fast workbook assistant while the accountant retains control of the source data, accounting logic, calculations, controls, review, and final conclusion.
One big idea
Put durable accounting logic in transparent Excel formulas and controlled data processes. Use AI around that logic—not as the logic itself.
Learning Objectives
After completing this lesson, you should be able to:
- Distinguish Copilot Chat, editing with Copilot, and advanced analysis with Python.
- Prepare an accounting workbook so Copilot can work with it effectively.
- Write prompts that identify the source, scope, accounting rule, output, and required controls.
- Use Copilot in practical workflows involving formulas, reconciliations, exceptions, variances, PivotTables, and reporting.
- Verify AI-assisted work before it affects a journal entry, reconciliation, financial report, or decision.
- Protect confidential financial information and follow organizational policies.
- Recognize tasks that should remain in native Excel formulas or under direct professional review.
Important Product Change
The COPILOT worksheet function is gone
Microsoft discontinued the preview COPILOT worksheet function on September 14, 2026. Existing cached results can remain visible, but recalculation of a cell containing the former function returns a #NAME? error because Excel no longer recognizes it.
Do not design a new workbook around formulas such as =COPILOT(...). Use the Copilot pane for summarization, classification, content generation, and workbook assistance. Continue to use native Excel formulas—such as SUMIFS, XLOOKUP, IF, ROUND, and EOMONTH—for calculations and controlled lookups.
Design principle: durable accounting logic belongs in transparent Excel formulas; AI assists around that logic.
Copilot Today
The exact interface depends on Excel version, account, platform, license, administrator settings, language, and rollout channel. In current Microsoft documentation, the central experiences are Copilot Chat, editing with Copilot, and advanced analysis with Python.
| Experience | What it does | Best accounting use | Main caution |
|---|---|---|---|
| Copilot Chat | Answers questions and discusses the open workbook without necessarily changing it | Explaining formulas, brainstorming tests, summarizing visible patterns, drafting instructions | A fluent answer can still be wrong; verify it against the workbook |
| Editing with Copilot | Makes direct workbook changes using Excel features such as tables, formulas, formatting, charts, and PivotTables | Building helper columns, exception views, management reports, and multi-step workbook structures | Review every edit, formula, range, filter, and assumption before accepting the result |
| Advanced analysis (Python) | Uses Python in Excel for deeper analysis and may create a new analysis sheet containing Python code | Statistical exploration, anomaly analysis, forecasting experiments, and large-scale summaries | Python availability is required; methods, inputs, and outputs still need independent review |
| Former COPILOT function | Previously generated AI responses inside cells | None for new workbooks | Discontinued September 14, 2026; recalculation returns #NAME? |
Microsoft previously marketed some direct-editing capabilities as Agent Mode and now integrates them into the regular editing experience in some releases. Other documentation and interfaces may still display Agent Mode as a tool, so the name visible to one user may differ from the name visible to another.
What It Can Do
Microsoft documentation describes editing with Copilot as able to add, rename, and delete worksheets; modify cells and ranges; create tables; apply formatting and data validation; generate and apply formulas; and create charts and PivotTables. Vendor materials also describe planning and executing multi-step tasks directly in the open workbook.
Copilot can also identify patterns and present insights as charts, PivotTables, summaries, trends, or outliers. Quality improves when the prompt names the relevant table, range, and column headers and specifies the desired form of the output.
These capabilities are useful, but they do not prove that the resulting workbook is accurate, complete, compliant, or appropriate. Microsoft states that Copilot cannot evaluate accuracy and that users must review, edit, and verify AI-generated work.
Before Opening Copilot
Complete the technical and accounting checklists before a serious accounting workflow. AutoSave writes cloud changes as you work—so a deliberate working copy is a control, not optional polish.
Prep progress: 0/18
Technical checklist
0/10 completeAccounting checklist
0/8 completeIf the accounting items are unclear, a longer prompt will not fix the underlying problem.
The CLEAR Prompt Method
Strong accounting prompts do more than describe a desired chart. They tell Copilot where to work, which accounting rules to follow, what output to create, and how the result will be checked.
Interactive CLEAR Prompt Builder
Step through Context, Locate, Execute, Accounting rules, and Review requirements. Compare a weak request with a stronger accounting prompt, then copy the complete template.
C — Context
State the business purpose, entity, period, and user.
Weak
Analyze expenses.
Better
You are assisting with the March 2026 month-end review for ABC Coffee Shop. The controller needs an operating-expense variance report for management.
Complete CLEAR template
Context: You are assisting with [entity, period, and accounting objective]. Locate: Use [worksheet/table/range] and these fields: [column names]. Execute: Create [specific formula, analysis, report, PivotTable, chart, or exception list]. Accounting rules: Apply [definitions, mappings, sign convention, materiality threshold, exclusions, and period rules]. Review requirements: Do not overwrite source data. Add [control totals and exception checks]. Return [form of output] and explain all formulas, filters, and assumptions.
The Accountant's 6-Stage Workflow
A controlled Copilot workflow has six stages. Professional guidance on AI-assisted accounting work emphasizes that AI remains a resource rather than a replacement for professional judgment. Material AI-assisted work should receive thorough human review, and primary authoritative sources should be checked directly when AI generates research, technical explanations, or citations.
Stage 1
Preserve
Keep the original source and create a working copy.
Stage 2
Prepare
Clean the workbook structure before asking AI to interpret it.
Stage 3
Prompt
Define the task, data, accounting rules, output, and controls.
Stage 4
Inspect
Review what Copilot plans and what it changes.
Stage 5
Prove
Recalculate independently and reconcile source to output.
Stage 6
Approve
Have the responsible accountant make and document the final decision.
Ten Practical Workflows
Expand each card for the prompt text, review points, and controls. Copy prompts into your organization's approved Excel environment—and verify every edit.
One of Copilot's safest high-value uses is explaining an existing formula before changing it.
Prompt
On the Reconciliation worksheet, explain the formula in cell K12 one function at a time. Identify every referenced range, explain the business purpose of the formula, and list three conditions that could cause a misleading result. Do not edit the workbook.
Better follow-up
Now propose a clearer version using LET, but do not insert it. Preserve the existing logic and show how the old and proposed formulas can be tested on five boundary cases.
What to inspect / review
- Does the explanation match the actual formula?
- Are absolute and relative references described correctly?
- Does the formula omit blanks, errors, or duplicate records?
- Is the period boundary correct?
- Could hidden rows or hardcoded values affect the result?
This workflow uses AI to accelerate comprehension while leaving the decision and edit with the accountant.
Lab: ABC Coffee Shop (March 2026)
The following is a fictional teaching scenario. The numbers are practice data, not research findings or industry benchmarks.
Assignment
ABC Coffee Shop's controller receives four files for March 2026:
GL_March.xlsx— general-ledger detail.Bank_March.xlsx— bank transactions.Budget_2026.xlsx— monthly budget by account.COA.xlsx— chart of accounts and management-report mapping.
The controller needs a bank reconciliation, expense-variance report, journal-entry review queue, and management summary.
Step 1 — Protect the sources
- Store the originals in a read-only source folder.
- Create a separate working workbook.
- Record the file names, extraction dates, periods, and row counts.
- Import each source into a separate worksheet.
- Convert each source population into a named Excel Table (
tblGL,tblBank,tblBudget,tblCOA). - Remember AutoSave: edits to the working copy in OneDrive or SharePoint persist as you go—do not edit the read-only originals.
Step 2 — Add a control sheet
Create a worksheet named Control with formula-driven totals. Do not ask Copilot to invent these totals. Ask it to insert native formulas referencing the named tables, then independently inspect each formula.
| Control | Source | Report | Difference |
|---|---|---|---|
| GL row count | Formula | Formula | Formula |
| GL debits | Formula | Formula | Formula |
| GL credits | Formula | Formula | Formula |
| Bank row count | Formula | Formula | Formula |
| Bank amount | Formula | Formula | Formula |
| Budget amount | Formula | Formula | Formula |
| Unmapped accounts | Formula | 0 expected | Formula |
Step 3 — Validate the general ledger
Use Copilot Chat or editing with Copilot to design an exception list—without changing tblGL.
Review the structure of tblGL without changing it. Identify blank required fields, invalid dates, nonnumeric amounts, duplicate JournalID-LineNumber combinations, entries outside March 2026, and AccountNumbers that do not appear in tblCOA. Return a proposed exception design on a new worksheet named GL_Validation: - One row per exception with JournalID, LineNumber, PostingDate, AccountNumber, Amount, and ExceptionReason. - ExceptionReason must list every rule triggered. - Add population row count, exception row count, and a control that sums exception Amounts for high-dollar blanks or invalid amounts. - Do not overwrite source data. Explain every formula, filter, and assumption before applying edits.
Independently sample flagged and unflagged rows. Unmapped accounts must also update the Control sheet's Unmapped accounts difference.
Step 4 — Bank recon assist
Use tblBank and cash-related GL lines (or a prepared tblCashGL). Never force matches to clear a difference.
Create a new worksheet named Bank_Recon_Work. Do not change tblBank or the cash GL population. Propose exact matches when amount is equal and either the reference matches or the dates are within three days. Do not automatically force one-to-many or many-to-one matches. Place uncertain and unmatched items in separate sections. For every proposed match, show the bank reference, GL journal ID, both dates, both amounts, date difference, amount difference, and matching reason. Add control totals for the original bank population, original GL population, matched items, and unmatched items. Reconcile those controls to the Control sheet bank row count and bank amount.
- Investigate deposits in transit and outstanding checks.
- Separate bank fees, interest, voids, and stale items.
- Document why each unmatched item remains open at March 31, 2026.
Step 5 — Expense variance
Build an operating-expense variance for March 2026 using tblGL, tblBudget, and tblCOA. Avoid causal language.
Create a worksheet named Expense_Variance for ABC Coffee Shop March 2026. Locate: use tblGL, tblBudget, and tblCOA. Match on AccountNumber. Execute: show AccountNumber, AccountName, Actual, Budget, DollarVariance, and PctVariance for operating-expense accounts only. Accounting rules: exclude income tax and owner distributions. Treat expense above budget as unfavorable (positive). Flag absolute variance greater than $1,000 or 10%. If ABS(Budget) is less than 100, show PctVariance as N/M. Do not state a cause unless the workbook contains supporting evidence. Add a Questions for Management section for items that need operational explanation. Review requirements: do not overwrite sources. Add source-row count, source actual total, report actual total, and a difference check that must equal zero.
Step 6 — Journal-entry review queue
Organize investigation leads. Flags are not fraud findings.
Using the March 2026 journal population derived from tblGL (and preparer/approver fields if present), create JE_Review_Queue on a new worksheet. Flag entries that meet any of these rules: - Posted on a weekend. - Manual source and amount exceeds $25,000. - Prepared and approved by the same user (if both fields exist). - Description blank or shorter than 10 characters. - Account on a SensitiveAccounts list you create from tblCOA (cash, related-party, and equity). - Posted in the final two days of March and reversed in the first two days of April (if reversal markers exist). Show the exact rule triggered. Do not describe an entry as fraudulent. Include population count and debit/credit control totals before and after filtering.
Step 7 — Management summary + prove controls
Draft a one-page management summary from verified report sheets only—not from raw, unreconciled transactions. Then prove and approve.
Create Management_Summary for March 2026 using only verified outputs from Bank_Recon_Work, Expense_Variance, and JE_Review_Queue. Include: - Bank reconciliation status: matched count, unmatched bank items, unmatched GL items, and unresolved dollar difference. - Top five expense variance flags with dollar and percentage variance (facts only). - Count of JE review flags by rule, without fraud language. - A Questions for Management list limited to items needing operational evidence. - Links or cell references to each supporting worksheet. Do not invent metrics. Do not insert synthetic sample data. Add an As-of date from the workbook period (March 31, 2026). Explain every formula and assumption.
Prove and Approve (required)
- Recalculate Control sheet differences; all material differences must be explained or zero.
- Tie management summary bank and expense totals to the supporting worksheets.
- Sample at least three matched bank items, three unmatched items, three variance flags, and three JE flags to source evidence.
- Document the reviewer name, date, and conclusion on a Signoff tab.
Deliverables checklist
0/8 readyKnowledge Check
Five questions on the discontinued COPILOT function, CLEAR prompts, reconciliation discipline, flags versus fraud, and durable native formulas.
Question 1: What happened to the preview COPILOT worksheet function as of September 14, 2026?
Question 2: In the CLEAR prompt method, what does the letter A represent?
Question 3: When Copilot proposes bank-reconciliation matches, which practice is required?
Question 4: A journal-entry review queue flags a weekend posting over $25,000. What does that flag mean?
Question 5: Where should durable accounting calculation logic live after the COPILOT function change?
Key Takeaways
- Put durable accounting logic in transparent native Excel formulas; use Copilot around that logic—not as
=COPILOT(...)cell logic (discontinued September 14, 2026). - AutoSave and cloud working copies are a control risk: preserve read-only sources before AI edits.
- CLEAR prompts (Context, Locate, Execute, Accounting rules, Review) beat vague requests that invite wrong ranges or unsupported conclusions.
- Never force reconciliation matches merely to clear a difference; keep unmatched items visible.
- Exception flags organize investigation—they are not proof of fraud or error.
- Avoid causal language in variance analysis unless the workbook contains supporting evidence; use Questions for Management instead.
- Agent Mode naming may still appear in some UIs; treat it as editing-with-Copilot and review every change.
Faster workbooks are valuable. Controlled workbooks—with preserved sources, native formulas, proved totals, and human approval—are what stand up to review.
Sources & Further Reading
Selected Microsoft support documentation, Journal of Accountancy AI policy guidance, and NIST human-AI interaction material. Product pages change; recheck official docs when evaluating a tool.
- Microsoft Support — Get started with Copilot in Excel
- Microsoft Support — Frequently asked questions about Copilot in Excel
- Microsoft Support — COPILOT function (discontinued preview)
- Microsoft Support — Edit with Copilot in Excel
- Microsoft Support — Advanced data analysis with Copilot in Excel and Python
- Microsoft Support — How Copilot Chat works in Microsoft 365 apps
- Microsoft Support — Get started editing with Copilot in Office (Agent Mode note)
- Microsoft Support — What is editing with Copilot in Excel?
- Microsoft Support — Copilot in Excel tips
- Microsoft Support — Get insights about numerical data with Copilot in Excel
- Microsoft Support — What you should know about AutoSave
- Microsoft Support — Turn on AutoSave in Microsoft 365 apps
- Microsoft Support — Generate formula rows and columns with Copilot in Excel
- Microsoft Support — Visualize your data with Copilot in Excel
- Journal of Accountancy — Drafting an AI policy that actually works
- NIST — AI Risk Management and Human-AI Interaction (Appendix C)
Ready to Practice?
Take CLEAR prompting and prove-before-approve habits into the AI Practice Arena—judgment first, automation second.
Open AI Practice ArenaWhat's Next?
Continue with prompt design and AI failure modes, or return to the AI hub and related practice topics such as AI in bookkeeping.
Using AI for Reconciliations
Exact rules first, candidate queues, and human sign-off
AI in Accounting Hub
Browse all AI pillar topics