Check an expense report against policy in a verdict table

About this exercise

You review expense reports before the manager approves them. Write a prompt for a table with columns Line | Amount | Verdict | Note, one row per expense line, with each verdict exactly one of OK, Over limit, Missing receipt or Not allowed, using the policy including its September update. Then a last line starting "Reimburse now:" that adds up the OK lines as math. Under 200 words.

Common mistakes when you ask AI for this

• Using the old hotel limit. The AI applies the $200 rule and flags the hotel, missing the September update. Try: Say: "Apply the Sept 2026 update: $225 a night in Columbus."
• Free-form verdicts. The AI writes 'looks fine' or 'questionable', which nobody can act on. Try: Give the labels: "Verdict is exactly one of OK, Over limit, Missing receipt, Not allowed."
• No total, or a total with no working. The AI gives a number without showing which lines it added. Try: Ask: "Last line: Reimburse now: with the OK amounts added as math."
• Trusting the AI to count words. Reviews come with intros, summaries and recommendations that blow the limit. Try: Say: "Each note under 8 words. Output only the table and the last line."

What a strong prompt includes

• Names the columns
• Gives verdict labels
• Applies the update
• Asks for the total as math
• Keeps it short

Practice this now ↑

Related exercises

• Accept a job offer and confirm the exact terms (Writing AI · Job seekers)
• Accrue goods received but not invoiced at PO price (Finance AI · Accountants)
• Age open requisitions as of today and flag the old ones (Finance AI · FP&A and financial analysts)
• Age reconciling items and escalate those over 60 days (Finance AI · Accountants)
• Allocate shared rent by headcount using the right roster (Finance AI · Accountants)