Free guide: AI
AI prompts for procurement analysts, with tests you can check
AI is useful in procurement for explaining, challenging and drafting. It is dangerous when it calculates or invents. These prompts come from the Starter Kit's 28-prompt library. Each one has a task, an evidence boundary, an output structure and a stopping rule, and most have a known answer on the practice dataset so you can test the AI before trusting it.
Three rules before any prompt
- Compute numbers in Excel or Power BI first. Give the AI verified figures to explain. Never ask it to total a column.
- Check what you may paste. Public tools get public or fictional data only. Confidential supplier, price, contract or personal data goes only into a tool your employer has approved for it.
- Numbers in, numbers out. Every number in the answer must appear in what you gave it. Delete anything that does not.
1. Challenge my conclusion
You are a sceptical finance business partner.
Finding and numbers: [paste].
List the five strongest reasons the conclusion could be wrong or premature, each with the
specific check that would confirm or rule it out, ranked by how likely it is to change the
decision. Do not restate the finding. Do not invent numbers.Test: paste "Professional Services has NGN 115,410,000 off-contract, so run a sourcing event." A good answer asks whether a few large invoices drive it. On the sample, one invoice (NGN 62,500,000) is 54.15% of that amount.
Using these prompts with ChatGPT or Claude
Paste only fictional or approved information into a public AI tool. Start with one of the test cases below, compare its answer with the known result, and then give it your verified figures. Neither tool replaces the Excel or Power BI calculation or your own review.
2. Overlapping exceptions
Two exception groups and their intersection: [amounts and definitions].
Explain the union calculation A + B - intersection, show the arithmetic, and state what the
result does and does not establish. If the intersection is missing, do not add the groups;
ask for the cross-tab. Treat the result as a review population, not loss or savings.Test: non-PO NGN 99,303,500, non-contracted NGN 209,663,500, both NGN 51,478,500. The correct union is NGN 257,488,500.
3. Explain two KPIs that disagree
PO compliance by spend is [82.03%]; by invoice count it is [70.00%]. Department view: [table].
Explain in plain language why they differ, what each answers, and which a policy owner and a
CFO should each watch. Under 150 words. Use only these numbers.4. Practice-project CV bullets
Practice work: [truthful description]. Dataset status: FICTIONAL. Verified findings: [list].
Tools actually used: [tools].
Write two CV project bullets and a short interview explanation, labelled as a practice
project. Do not imply a real employer, an implemented recommendation, savings, promotion,
credential or endorsement. Include the decision reasoning and one limitation.What never to hand to AI
Do not let AI act on its own authority to approve spend or payments, change supplier bank details, score or reject suppliers, issue contractual notices, send anything external unread, report savings Finance has not validated, or accuse an individual. AI may draft and flag. A named person decides.
Why the tests matter. A prompt that sounds right can still produce a confident wrong number. Running it first on data with a known answer is the quickest way to learn where a tool fails.
Go further
The Starter Kit teaches this method end to end on the practice dataset: a 54-page guide, a click-by-click Excel and Power BI build guide, 29 exercises and 3 case studies with worked answers, 28 AI prompts, a self-checking workbook and a 29-KPI dictionary.
CA$15 once (about US$11), plus local sales tax where it applies, shown at checkout. Secure checkout by Stripe. Instant download. All sales final.
Free one-page KPI cheat sheet: download the PDF, no email needed.