ImpactMe.AI · NECA Accounting Workshop · October 27, 2026

Prompt Library for Accounting and Finance Teams

Built for accounting staff at electrical contracting firms. Works in Microsoft Copilot, ChatGPT, Claude, or Gemini.

How to use this library (and why you'll outgrow it)

Prompts are the fish. Thinking like AI is the fishing.

This library gives you fish: 36 ready-to-use templates for real accounting work. Use them. But know that specific prompts go stale fast. Tools update, features move, and a prompt that worked perfectly in October may need tweaking by November. The part that lasts is the pattern behind each prompt. Once you see it, you can write your own for any task, in any tool.

The pattern behind every good prompt

Almost every prompt below is built from the same five parts:

Part	What it does	Example phrase
1. Role	Tells the tool what kind of expert to act as	"You are a construction accountant..."
2. Context	Gives the background it can't guess	"We're an electrical contractor. This is our March WIP..."
3. Task and format	Says exactly what you want back, and in what shape	"Give me a table with columns for..."
4. Constraints	Sets limits: length, tone, what to avoid	"Under 150 words. Don't change any numbers."
5. Check yourself	Asks it to state assumptions, flag uncertainty, and show work you can verify	"List your assumptions. Flag anything you're unsure of. Show the math."

Part 5 is the one most people skip, and it's the one that matters most in accounting. These tools will happily give you a confident, well-written answer that is wrong. Asking them to show their work and flag doubts makes errors easier to catch.

Three ground rules

Use sample or de-identified data in consumer tools. If you're using a free or personal ChatGPT, Claude, or Gemini account, don't paste real client names, bank details, SSNs, payroll, or unreleased financials. Swap in fake names and round numbers, or use your firm's approved business-tier tool. See the Safe Use Traffic Light handout for details.
Every number gets a human check. Each prompt below that touches numbers has a Verify step. Don't skip it. The AI is not your calculator. Excel is your calculator.
Treat the first answer as a draft. If it's not right, tell it what's wrong and ask again. That's normal, not failure.

How to read each entry

The template goes in a gray box. Replace anything in [BRACKETS] with your details.
Why it works names the pattern, so you can reuse it elsewhere.
Verify tells you how to check the output before you trust it.

1. Month-end close and reconciliation

1.1 Build a month-end close checklist

You are an experienced controller at a [SIZE, e.g. 60-person] electrical contracting company. We close our books in [NUMBER] business days using [ACCOUNTING SYSTEM, e.g. Sage 300 CRE / Viewpoint Vista / QuickBooks].

Draft a month-end close checklist organized by day (Day 1, Day 2, etc.). Include contractor-specific items such as job cost review, WIP schedule, retainage receivable and payable, accrued payroll and union fringe benefits, and equipment cost allocation.

Format as a table: Task | Owner role | Day | Dependency.
List any assumptions you made about our process, and mark items you think may not apply to every contractor.

Why it works: Role plus specific context (industry, system, timeline) gets you a relevant list instead of a generic one. Asking it to mark "may not apply" items invites it to flag uncertainty.

1.2 Investigate a bank reconciliation difference

I'm reconciling our [BANK NAME OR "operating"] account for [MONTH]. The difference between the bank statement balance and the GL balance is [$ AMOUNT]. Here are the unmatched items from both sides (sample data):

[PASTE UNMATCHED ITEMS: date, description, amount]

Suggest the most likely explanations for the difference, ranked from most to least likely. For each, tell me exactly what to look for in our records to confirm or rule it out.
Do not assume any item is resolved. Show any math you do.

Why it works: You give it the specific puzzle and ask for ranked hypotheses plus how to test each one. It becomes a thinking partner, not an answer machine.

Verify: Re-add the unmatched items yourself in Excel and confirm the difference. If it says two items net to the gap, check that sum with a calculator. The fix only counts when your reconciliation actually ties to zero.

1.3 Spot common reconciliation patterns

Here is a list of unmatched transactions from a [CREDIT CARD / VENDOR STATEMENT / INTERCOMPANY] reconciliation (de-identified):

[PASTE LIST]

Look for patterns that commonly cause mismatches: transposed digits, duplicates, timing differences, partial payments, amounts that net together, and sign errors. Group your findings by pattern type.
For every match you suggest, show both amounts side by side so I can check them. Say "no clear match" when you're not sure rather than guessing.

Why it works: Naming the patterns to look for focuses the tool. "Show both amounts side by side" makes its work checkable. Permission to say "no clear match" reduces made-up answers.

Verify: Check every suggested pair against the source documents. A transposition (like 4,520 vs 4,250) always leaves a difference divisible by 9: confirm that with a calculator, not the AI.

1.4 Draft a journal entry with explanation

Act as a senior staff accountant at a construction company. I need to record the following: [DESCRIBE TRANSACTION, e.g. "accrue 2 days of field labor at month-end that will be paid in the first payroll of next month, including union fringe benefits"].

Use these sample figures: [AMOUNTS].
Draft the journal entry (account names, debits, credits) and a one-paragraph memo explaining it for the file.
State which accounting assumptions you're making, and tell me what I should confirm with my controller before posting.

Why it works: Asking for a memo alongside the entry forces it to explain its reasoning, which makes errors easier to spot. "What should I confirm" keeps you in charge.

Verify: Confirm debits equal credits. Recompute any accrual amount yourself. Check account names against your actual chart of accounts. Never post an AI-drafted entry without your normal review.

1.5 Prepare a flux (fluctuation) review list

Here is a trial balance comparison for [CURRENT MONTH] vs [PRIOR MONTH] (sample or de-identified):

[PASTE: Account | Prior | Current]

Calculate the dollar and percent change for each account. Flag any account where the change exceeds [$ THRESHOLD] or [% THRESHOLD]. For each flagged account, suggest 2 to 3 questions I should ask to explain the change, specific to an electrical contractor.
Present as a table. Show the formula you used for percent change.

Why it works: You set the thresholds (a hard rule) and let the AI do the softer work: generating good questions. Asking for the formula makes the math checkable.

Verify: Recompute the changes in Excel. AI tools can make arithmetic mistakes, especially across long lists. Use the AI's questions; use Excel's math. Confirm the flagged list against your threshold with a filter, not by eye.

1.6 Write an accrual or reserve rationale

I need to document the rationale for our [WARRANTY RESERVE / BAD DEBT ALLOWANCE / ACCRUED EXPENSE] at [MONTH-END]. Here is our method and the sample figures: [DESCRIBE METHOD AND NUMBERS].

Write a clear, auditor-friendly memo (under 300 words) explaining the method, inputs, and conclusion. Use plain language.
Do not invent any figures or accounting standards I haven't given you. If the memo needs information I didn't provide, leave a [PLACEHOLDER] and list what's missing at the end.

Why it works: The constraint "do not invent figures or standards" plus the placeholder instruction prevents the tool from filling gaps with plausible-sounding fiction.

Verify: Every number in the memo must match your workpaper. Check any standard or rule it references with your controller or the actual source before it goes to an auditor.

2. Excel formulas and cleanup

Tip: If your firm uses Copilot in Excel, you can ask many of these directly inside your spreadsheet. In other tools, describe your columns clearly.

2.1 Write a formula from a plain-English description

I'm using Excel [VERSION, e.g. Microsoft 365]. My data is in columns:
A = [e.g. Job number], B = [e.g. Cost code], C = [e.g. Amount], D = [e.g. Date]
Data runs from row 2 to row [N].

I need a formula that [DESCRIBE GOAL, e.g. "sums column C for job number in cell G1 and cost code in G2, only for dates in October 2026"].

Give me the formula, then explain each part in plain English. Tell me one simple test I can run to confirm it's working.

Why it works: Exact column layout (context) plus "explain each part" (checkable work) plus "one simple test" (verify step built in).

Verify: Run the test. Then manually add up 3 to 5 rows that should match and compare. A formula that returns a number is not the same as a formula that returns the right number.

2.2 Fix a broken formula

This Excel formula is giving me [ERROR, e.g. #N/A / wrong result / #VALUE!]:

[PASTE FORMULA]

Here's what it's supposed to do: [PLAIN DESCRIPTION]. Here's a sample of the data it's looking at: [DESCRIBE OR PASTE 3 TO 5 SAMPLE ROWS].

Explain the most likely cause, give me a corrected formula, and tell me what to check in my data (like extra spaces, numbers stored as text, or mismatched ranges).

Why it works: Error plus intent plus sample data gives it everything a human expert would ask for. Asking "what to check in my data" catches problems the formula alone won't reveal.

Verify: Test the corrected formula on rows where you already know the answer.

2.3 Clean up a messy export

I exported [REPORT NAME] from [SYSTEM] to Excel and it's messy: [DESCRIBE PROBLEMS, e.g. "merged header rows, subtotals mixed in with detail lines, job numbers with extra spaces, dates stored as text"].

Give me a step-by-step plan to clean it up in Excel, using built-in features (Power Query, Text to Columns, TRIM, filters) rather than macros where possible.
Number the steps. Tell me how to confirm I haven't lost or duplicated any rows along the way.

Why it works: Describing the specific mess gets a specific fix. Asking for a row-count check builds verification into the process.

Verify: Compare the row count and the grand total of the amount column before and after cleanup. If either changed and you didn't mean it to, stop and find out why.

2.4 Build a lookup between two lists

I have two lists in Excel:
Sheet1 has [e.g. vendor invoices: Invoice # in A, Vendor in B, Amount in C].
Sheet2 has [e.g. payments: Invoice # in A, Payment amount in B].

I want to [e.g. "flag invoices that have no payment, and invoices where the payment doesn't equal the invoice amount"].

Give me the formulas (XLOOKUP preferred, with an IFERROR fallback). Explain how to handle duplicate invoice numbers if they exist.

Why it works: Clear structure of both lists, a clear goal, and a nudge about a known risk (duplicates) that the AI might otherwise ignore.

Verify: Spot-check five flagged rows and five unflagged rows by hand. Check whether your data really has duplicates (use COUNTIF) before trusting a lookup.

2.5 Explain a spreadsheet someone else built

I inherited a spreadsheet and need to understand this formula:

[PASTE FORMULA]

Explain what it does in plain English, step by step, as if I'm comfortable with SUM and IF but not with nested functions. Tell me any situations where it could give a wrong or misleading result.

Why it works: Telling it your skill level gets the explanation pitched right. Asking for failure cases turns an explanation into a review.

Verify: Test the explanation against the sheet: pick two rows, predict the result from the AI's explanation, and confirm the formula actually returns that. Use Excel's Evaluate Formula tool (Formulas tab) to step through it.

2.6 Design a reusable template

Help me design an Excel template for [PURPOSE, e.g. "tracking retainage receivable by job and GC"]. Suggest the columns, data types, any formulas, and simple checks (like a total that should tie to the GL).

Keep it simple enough that someone new to the team could maintain it. Present the column layout as a table, then list the formulas separately.

Why it works: "Simple enough for someone new" is a constraint that prevents over-engineering. Asking for built-in checks makes the template self-verifying.

Verify: Once built, enter three test rows where you know the answer and confirm every formula and check behaves correctly.

3. Reporting and variance commentary

3.1 Draft variance commentary

You are a financial analyst writing for the owner of an electrical contracting firm. Here is our [MONTH] actual vs budget summary (sample figures):

[PASTE: Line item | Budget | Actual | Variance]

Additional context I know: [e.g. "Two large jobs started late; we had unplanned overtime on the hospital project; copper prices rose"].

Write 3 to 5 bullet points of variance commentary in plain language, focused on the biggest items. Only use explanations I've given you or that follow directly from the numbers. Mark anything you're inferring with "(confirm)".

Why it works: You supply the why; the AI supplies clear wording. The "(confirm)" tag separates facts from guesses so you can see which is which.

Verify: Check every number in the commentary against the report. Remove or confirm every "(confirm)" item before it goes out.

3.2 Turn a report into an executive summary

Here is our [REPORT NAME, e.g. monthly financial package] (de-identified): [PASTE OR DESCRIBE].

Write a one-paragraph summary for [AUDIENCE, e.g. "our owner, who reads it on his phone between jobsites"]. Lead with the single most important takeaway. Use no more than [NUMBER] numbers. Then list 2 to 3 items needing a decision or attention.

Why it works: A specific audience and a hard limit on numbers forces prioritization, which is the hardest part of summarizing.

Verify: Every number cited must match the source report exactly.

3.3 Suggest KPIs for an electrical contractor

Act as a CFO advisor to specialty contractors. Suggest 8 to 10 KPIs that an electrical contractor with [REVENUE RANGE] and [TYPE OF WORK, e.g. commercial and industrial] should track monthly.

For each: name, formula, why it matters, and a warning sign to watch for. Present as a table.
Do not include industry benchmark numbers unless you're confident they come from a real, named source, and if you do, name the source so I can check it.

Why it works: Asking for formulas makes each KPI concrete and checkable. The benchmark rule addresses a known weakness: AI tools sometimes produce official-sounding statistics that don't exist.

Verify: Look up any benchmark it cites in the named source yourself. If it can't name a source, treat the number as made up.

3.4 Restructure a report layout

Here is how our [REPORT NAME] is currently laid out: [DESCRIBE COLUMNS AND ROWS].
The people who read it are [AUDIENCE] and they mainly want to know [KEY QUESTION].

Suggest a better layout. Explain what you'd move, combine, or drop, and why. Sketch the new layout as a simple table with placeholder values.

Why it works: Starting from the reader's key question gives the AI a goal to design toward, rather than just rearranging.

3.5 Check a report for internal consistency

Review this report text and its numbers for internal consistency (sample data):

[PASTE REPORT]

List any places where numbers don't agree with each other, where the text describes a number differently than the table shows, or where a total doesn't appear to match its parts. Show the specific figures for each issue. Do not fix anything; just list issues.

Why it works: "Do not fix, just list" keeps the AI as a reviewer and you as the decision-maker. Asking for specific figures makes each finding checkable.

Verify: The AI can miss errors and can also report errors that aren't there. Confirm each issue it raises, and still run your normal tie-outs. This is a second set of eyes, not a replacement for yours.

4. Job costing, WIP and progress billing for contractors

4.1 Explain a WIP schedule line by line

You are a construction accountant. Explain how to read this WIP (work-in-progress) schedule row, using cost-to-cost percent complete (sample data):

Contract value: [$]
Estimated total cost: [$]
Cost to date: [$]
Billed to date: [$]

Walk through: percent complete, earned revenue, over or under billing, and estimated gross profit. Show each formula and calculation step. Then explain in plain English what the result means for the business.

Why it works: Asking for each step and formula makes the logic transparent and teachable. The plain-English explanation helps you explain it to others.

Verify: Recompute every figure in Excel. The core formulas: percent complete = cost to date ÷ estimated total cost; earned revenue = percent complete × contract value; over/(under) billing = billed to date minus earned revenue. If the AI's numbers differ from yours, yours win.

4.2 Flag WIP jobs that need attention

Here is our WIP schedule (de-identified job names, sample figures):

[PASTE: Job | Contract | Est. total cost | Cost to date | Billed to date | Prior month est. gross profit %]

Identify jobs that may need attention: large underbillings, gross profit fade from last month, cost to date exceeding estimate, or percent complete that looks inconsistent with billing. For each, explain why it's flagged and what question to ask the project manager.
Show the figures behind each flag.

Why it works: Naming the warning signs you care about focuses the review. "Show the figures behind each flag" lets you check each conclusion.

Verify: Build the same flags in Excel with formulas and conditional formatting. Use the AI to suggest what to look for and what to ask; use Excel to decide which jobs actually meet the criteria. Compare the two lists.

4.3 Draft questions for a project manager review

I'm meeting with a project manager to review job [JOB NAME/NUMBER, de-identified]. Here's the situation: [e.g. "labor costs are 15% over budget at 60% complete, material is on budget, two change orders are pending approval"].

Draft 6 to 8 questions to ask, ordered from most to least important. Keep them non-accusatory and focused on getting accurate estimate-to-complete information.

Why it works: Context plus a tone constraint (non-accusatory) plus a purpose (accurate estimates) gives you questions that get real answers instead of defensive ones.

Verify: Confirm the figures you describe (percent over budget, percent complete, change order status) from the job cost report before the meeting, so the questions rest on accurate numbers.

4.4 Check a pay application before it goes out

I'm preparing a progress billing (in the style of an AIA G702/G703 pay application) for [JOB, de-identified]. Here is the schedule of values with this period's work (sample figures):

[PASTE: Line | Description | Scheduled value | Previous | This period | Materials stored | Retainage %]

Walk me through a review checklist for this pay app: math checks, retainage calculation, change order inclusion, and common reasons GCs reject pay apps. Show any calculations step by step.

Why it works: You get both a reusable checklist and a review of this specific pay app. Showing calculations step by step makes errors visible.

Verify: Recompute every line, the retainage, and the totals in Excel. Confirm retainage percentages against the actual contract terms, not the AI's assumptions. A pay app is a legal billing document: a human signs off, every time.

4.5 Analyze labor cost overruns

Here is labor cost data for [JOB, de-identified] by cost code (sample figures):

[PASTE: Cost code | Budgeted hours | Actual hours | Budgeted cost | Actual cost]

Calculate the variance for hours and cost for each cost code. Separate the cost variance into how much comes from more hours (quantity) versus higher hourly cost (rate). Show your formulas. Then suggest possible causes I should investigate for the largest variances.

Why it works: Asking for the rate vs quantity split gives you more useful analysis than a single variance number. Formulas make it checkable.

Verify: Recompute in Excel. Confirm that rate variance plus quantity variance equals total variance for each code. If it doesn't, something is off.

4.6 Explain over and under billing to a non-accountant

Explain over billing and under billing on a construction WIP schedule to [AUDIENCE, e.g. "a new project manager who came up through the field as an electrician"].

Use a simple example from an electrical job, under 200 words, no accounting jargon. End with why it matters to them personally (cash flow, job profitability, their own review).

Why it works: A specific audience with a specific background gets you an explanation that lands. The "why it matters to them" ending makes it stick.

Verify: If the example includes numbers, check that the math works. Explanations written for beginners sometimes simplify things until they're wrong.

5. Communication

5.1 Request retainage release from a general contractor

Draft a professional email to [GC CONTACT ROLE, e.g. project accountant] at [GC NAME] requesting release of retainage on [PROJECT NAME]. Facts: substantial completion was [DATE], our retainage balance is [$], and we've submitted [e.g. closeout documents, lien waivers, O&M manuals] on [DATE].

Tone: firm but friendly, we want to keep working with them. Under 150 words. Include a clear ask and a specific date for a response.

Why it works: Facts plus relationship goal plus length plus a clear call to action. You control the facts; the AI handles tone and structure.

Verify: Check every date, dollar amount, and document name against your records before sending.

5.2 Follow up on a slow payment

Draft a [FIRST / SECOND / FINAL] follow-up email to [GC or CUSTOMER] about invoice [NUMBER] for [$], which is [NUMBER] days past due. Previous contact: [e.g. "emailed on Oct 1, no reply"].

Give me two versions: one that's a gentle nudge and one that's more direct about next steps. Keep each under 120 words.
Do not mention liens, legal action, or specific deadlines unless I tell you to; I'll add those after checking our contract and state rules.

Why it works: Two versions let you choose the strategy. The constraint about liens and legal language keeps the AI away from areas where wrong advice has real consequences.

Verify: Confirm the invoice number, amount, and days past due. Lien and notice deadlines vary by state and contract: confirm those with your own records and advisor, never with the AI.

5.3 Explain a financial result to a non-finance manager

I need to explain [TOPIC, e.g. "why our cash is tight even though we're profitable this quarter"] to [AUDIENCE, e.g. our operations manager]. Here are the key facts (sample figures): [FACTS].

Write a short explanation (under 200 words) using an everyday analogy. Avoid accounting jargon, or define it in one line if you must use it. End with one thing they can do to help.

Why it works: Audience plus analogy plus jargon rule plus an action step. Turns a lecture into a conversation.

Verify: Make sure every figure matches your actual data and that the analogy doesn't distort the real situation.

5.4 Soften a tough message

I need to tell [AUDIENCE, e.g. "a project manager"] that [MESSAGE, e.g. "his job's margin has faded 6 points and the owner wants a review meeting"]. Here's my rough draft:

[PASTE DRAFT]

Rewrite it to be clear and direct without sounding like blame. Keep my facts exactly as written. Under [NUMBER] words. Then tell me if anything in my draft might land badly and why.

Why it works: "Keep my facts exactly" protects accuracy. Asking how it might land gives you feedback, not just a rewrite.

Verify: Compare the rewrite to your draft and confirm every number and fact survived unchanged. Rewrites sometimes round, soften, or drop a figure.

5.5 Prepare for a difficult conversation

I have a meeting with [ROLE] about [TOPIC, e.g. "repeated late timesheets that are delaying payroll and job costing"]. Help me prepare: list the 3 key points I need to make, 3 objections they might raise, and a calm, specific response to each. Keep it practical.

Why it works: Asking for objections and responses turns the AI into a rehearsal partner, one of the things it's genuinely good at.

6. SOPs and process documentation

6.1 Turn rough notes into an SOP

You are a process documentation specialist. Turn these rough notes into a clear standard operating procedure (SOP):

[PASTE NOTES, e.g. "how I do the weekly AP check run"]

Format: Purpose, Scope, Roles, Step-by-step procedure (numbered), Controls and checks, Common mistakes. Write for someone new to the team.
Where my notes are unclear or a step seems missing, insert [QUESTION: ...] instead of guessing.

Why it works: A standard SOP structure makes output consistent across your team. The [QUESTION] instruction surfaces gaps instead of filling them with invented steps.

Verify: Have someone who actually does the task walk through the SOP step by step. Answer every [QUESTION] before publishing. An SOP with one invented step is worse than no SOP.

6.2 Interview me to build an SOP

I want to document how we [PROCESS, e.g. "process change orders from field request to billing"]. Instead of writing it yourself, interview me. Ask me one question at a time, wait for my answer, and keep going until you have enough to write a complete SOP. Then draft it.

Why it works: Flipping the AI into interviewer mode pulls out knowledge that lives in your head. It's often faster than writing from scratch, and the result reflects how your firm actually works.

6.3 Find gaps and risks in an existing process

Here is our current procedure for [PROCESS, e.g. "adding a new vendor and their banking details"]:

[PASTE PROCEDURE]

Review it as an internal auditor would. Identify control gaps, segregation-of-duties issues, and fraud risks (for example, vendor bank change scams). Rank by risk level. Suggest a practical fix for each that a small accounting team could actually do.

Why it works: The auditor role brings a skeptical lens. "Practical for a small team" keeps recommendations realistic.

Verify: Discuss the findings with your controller. The AI may miss risks specific to your firm and may flag things you already handle.

6.4 Create a quick-reference checklist from an SOP

Here is a full SOP: [PASTE SOP].

Condense it into a one-page checklist someone can print and keep at their desk. Checkbox format. Keep every control step and every approval step, even if you shorten the wording.

Why it works: The "keep every control step" constraint prevents the AI from summarizing away the parts that matter most.

Verify: Compare the checklist to the SOP side by side and confirm no control or approval step was dropped.

7. Learning and explaining

7.1 Learn a concept at your level

Explain [CONCEPT, e.g. "how revenue recognition works on long-term construction contracts"] to me. I'm [YOUR LEVEL, e.g. "an AP specialist with 5 years of experience, but I've never worked on the revenue side"].

Start with the big idea in 2 to 3 sentences, then go deeper. Use an electrical contracting example. End by asking me 2 questions to check my understanding.

Why it works: Stating your level gets the right depth. The quiz at the end turns reading into learning.

Verify: For anything you'll rely on professionally, confirm key points with your controller, CPA, or the actual standard. AI explanations are a great starting point, not the final authority.

7.2 Quiz me

I'm learning [TOPIC, e.g. "the difference between job cost, overhead, and G&A for a contractor"]. Give me 5 practice questions, one at a time. Wait for my answer, tell me if I'm right and why, then give the next question.

Why it works: Interactive, one-at-a-time format makes the AI a patient tutor.

7.3 Compare two approaches

Compare [OPTION A] and [OPTION B] for [PURPOSE], e.g. "percent-complete by cost-to-cost vs by labor hours, for measuring progress on electrical jobs".

Present as a table: how it works, when it fits best, pros, cons, common pitfalls. Then tell me which questions I should answer to decide which fits our situation. Don't pick one for me.

Why it works: "Don't pick one for me" plus "questions I should answer" keeps the decision with you and gives you a framework to decide.

7.4 Ask the AI to critique its own answer

Look at your previous answer. Act as a skeptical senior accountant reviewing it. What's most likely to be wrong, oversimplified, or missing? What would you need to know to be more confident? Rate your confidence in each main point as high, medium, or low.

Why it works: This works after almost any prompt. Asking for a self-critique often surfaces problems the first answer glossed over. Keep in mind the confidence ratings are the tool's own guess, not a guarantee: a "high" can still be wrong.

Verify: Use the low and medium items as your checklist for what to confirm yourself.

Write your own: a blank template

You are [ROLE].
Context: [WHO WE ARE, WHAT THIS IS, WHAT I ALREADY KNOW].
Task: [EXACTLY WHAT I WANT].
Format: [TABLE / BULLETS / EMAIL / LENGTH].
Constraints: [WHAT TO AVOID, WHAT NOT TO CHANGE].
Check yourself: List your assumptions. Flag anything you're unsure of. Show any math step by step so I can verify it.

Then: Verify anything with a hard, checkable answer (totals, thresholds, dates, compliance) using a tool that gives the same answer every time: Excel, your accounting system, or a calculator.

Questions after the workshop? Adam Murphy · ImpactMe.AI · AI Made Useful & Human · adam@impactme.ai · impactme.ai