Each prompt matches a job from the article: checking delivery, comparing a renewal with the signed contract, reconciling a spreadsheet before a customer sees it, prepping and grading a meeting, and reading a vendor quote. Paste the common rules first, then one prompt, then attach the inputs. Use it with whatever AI tool your company already approves. You don't need to build anything.
Download the 2-page PDF · Open copy-ready text
Before you start
- Use a company-approved tool, and only with data it's approved for. If you aren't sure whether contracts, customer data or call recordings are allowed, ask first, or practice on sample or redacted files.
- Decide the permissions up front: what it can read, what it can draft, and what only a person can send, sign or pay.
- Try each prompt on a past example where you already know the answer before you use it on live work.
Common rules (paste these first)
Work only from the files and systems I give you access to here.
Before you start, list the inputs you received and anything that's missing.
For every finding, cite where it came from: file, page or section, sheet and cell, ticket, or timestamp.
Keep what the source says separate from what you're assuming, and label every assumption.
If you can't find something, write "I can't find that." Don't guess or fill the gap.
Use spreadsheet formulas I can check for any arithmetic.
Work in a copy of supplied files. You may create the requested draft analysis and working files. Don't change source systems or send, sign, pay, post or approve anything.
End with the questions a person needs to answer before acting on this.
Prompt 1: Delivery check: what's been delivered, and what hasn't?
Use when: Before a status meeting with an engineering team, agency or outsourced provider.
Inputs: Ticket or project-board export; commit or release notes; the team's updates and emails; QA or test report; the acceptance criteria you agreed.
Trace every work item in [scope or milestone] through the tickets, release records, team updates and test report. Return a table: item | status claimed | evidence found, with source | live where the customer will use it? | gaps or contradictions | what the team still needs from me (decision, test data, acceptance criteria) | next step.
Treat "code complete" and "ready for a customer" as different statuses. Flag any item marked Done whose own notes list checks that haven't happened. Then draft a short "done means" list for each open item so we can agree on it before more work starts.
Human review point: Walk through the table with the team as a shared starting point, not a verdict. Confirm any gap with its owner before anyone calls it late.
Prompt 2: Contract or renewal comparison
Use when: A supplier sends a renewal, order form, addendum or amendment.
Inputs: The signed agreement and every amendment; the new document; prior order forms; your notes from the sales call; recent invoices.
Compare the new document with the signed agreement and prior orders. Return: item | signed version, with section | new version, with section | what changed | money or date impact | question for the supplier.
Check specifically: renewal and notice deadlines, including time zone; dates and signatories; how any "credit" or discount is calculated; every clause in the new document you can't find in the signed one; and anything in the old agreement or orders that the new document doesn't cancel or replace, such as a support charge that could restart. If my notes disagree with the documents, list the conflict instead of choosing a side.
Human review point: Read every flagged section in the original yourself. This is a comparison to prepare questions, not legal advice. Route material terms to legal or procurement before anything is signed.
Prompt 3: Spreadsheet reconciliation and customer-demo check
Use when: Someone else built the file, or a customer is about to see the numbers.
Inputs: The file to check; the source export or system records; how each total is defined; the screens or reports the customer will see.
Validate this file against the source records and add a tab that shows your work: each source row, the matching row in the file, the difference, and the formula used. Flag unit, currency, decimal and transcription errors (for example, 842.17 that became 842,170). Reconcile totals by month and category, and list every difference you can't explain.
Then list every screen, report, field and scorecard the customer will actually look at. Confirm each one is populated and correct, not just that the totals tie. Report anything blank, defaulted or still a placeholder.
Human review point: Spot-check a sample of rows yourself. Before the meeting, walk the exact path the customer will see, on the account you'll be using.
Prompt 4: Meeting prep, then an independent grade
Use when: Before and after a sales call, supplier review or presentation.
Inputs: Notes from prior meetings; public information your company policy allows you to use; your own scoring rubric; the transcript, recorded only with everyone's permission.
PREP: Summarize what we've discussed before, what they said they need, open commitments on both sides, and relevant public news. Mark which points come from my notes and which from public sources. Suggest three questions and one clear ask.
GRADE (run in a new chat or a different model): Grade this meeting against my rubric, not against any plan written before the meeting. Quote the transcript for every deduction. Separately, assess whether the meeting reached its goal, such as an agreed next step. If the plan was the problem, say so.
Human review point: Check every deduction against the transcript. When the grade is wrong, fix the rubric, not just the score. Don't let whoever wrote the plan be the only one grading it.
Prompt 5: Vendor pricing: what did the supplier say, and what are you assuming?
Use when: A quote, rate card or volume tier needs checking against what you actually need.
Inputs: The quote and rate card; your volumes and contract term; supplier emails or clarifications; your requirements.
Break the quote into every price line and tier. Return: line | unit and billing period | what the supplier stated, with source | what you're assuming | cost per unit at our volume | total for the term.
Keep annual and monthly charges separate. Calculate cost per unit as dollars divided by units, and show the formula. Flag minimums, overage rates, auto-renewal, price increases and anything not priced. End with the questions the supplier must answer in writing before we compare options.
Human review point: Recalculate two of the numbers yourself. Get the supplier to confirm every assumption in writing before it goes into a business case.
Reusable loop template
The second time you type the same instructions, stop and write them down as a loop. Fill this in once, save it wherever your approved tool keeps saved instructions, projects or custom agents, and run it by name.
Loop name and trigger phrase: A phrase you'll actually remember, such as "run the renewal check loop."
Purpose: The decision or deliverable this loop supports, and who uses it.
Inputs: The files and systems it needs, and where they live.
Permissions: Can read: ... / Can draft: ... / Only a person can send, sign, pay or change: ...
Steps: Numbered, in order. Paste the common rules and one prompt from this sheet as a starting point.
Expected output: Format, columns, length, and who it's for.
Source and assumption checks: What must be cited, what counts as an assumption, and what to do when evidence is missing.
Acceptance criteria: How you'll know it's right. For example: totals tie to the source, every item has a source, unknowns stay visible.
Human review point: Who reviews it, when, and what they check before anything goes out.
Corrections log: Date | what went wrong | the new rule | re-tested on the same example? (yes/no)
Saving corrections: When it gets something wrong, fix the loop, not just the answer. Write the correction as one plain rule, add it to the saved loop, rerun the same example, and log whether the rule worked.
Rules worth saving, from the article:
- "Keep annual and monthly charges separate."
- "Missing evidence is unresolved, not a three."
- "A roadmap feature is not a capability available at go-live."
These prompts help you find questions and gaps. They don't replace legal, financial, security or other professional review, and no prompt guarantees accurate output. Check the work against the source before you rely on it.
Read the companion article: How I Actually Use AI to Run ScaleX