AR AGEING AI TEST - PROMPT PACK
As-of date: 30 September 2026

1. CONTROL TOTAL
Using the uploaded AR ageing file, calculate the total open amount and show the invoice count and open amount by ageing bucket. Use the Age_Bucket and Open_Amount columns exactly as provided. Show a total row. Do not infer or reclassify any invoice.

2. CUSTOMER CONCENTRATION
Rank customers by total Open_Amount. Show the top 10 customers and, for each customer, split the balance across Current, 1-30, 31-60, 61-90 and 90+ ageing buckets. Include the total open balance and the percentage that is 90+ days overdue.

3. OLD INVOICES WITH STALE CONTACT
Identify all open invoices that are more than 90 days past due and where Last_Contact_Date is more than 30 days before 30 September 2026. Show Invoice_ID, Customer, Open_Amount, Days_Past_Due, Last_Contact_Date, Dispute_Flag and Sales_Owner. Sort by Open_Amount descending. Also show the total exposure represented by these invoices.

4. PASSED PROMISE-TO-PAY DATES
Find every invoice where Promise_To_Pay_Date is before 30 September 2026 and Open_Amount is greater than zero. Show the customer, invoice, open amount, promise-to-pay date, days past the promise date and ageing bucket. Sort by Open_Amount descending and show the total open exposure.

5. DISPUTED OVERDUE BALANCES
Calculate the total overdue Open_Amount where Dispute_Flag = Y. Break it down by customer and ageing bucket. Then list the 20 largest disputed invoices with Invoice_ID, Customer, Open_Amount, Days_Past_Due and Sales_Owner.

6. COLLECTIONS REVIEW LIST
Create a collections review list from the uploaded file. Prioritise invoices using only information available in the file. Consider:
- Open_Amount
- Days_Past_Due
- whether the invoice is 90+ days overdue
- whether the last customer contact was more than 30 days ago
- whether a promise-to-pay date has passed
- whether the invoice is disputed

Return the top 20 invoices for human review. For every invoice, show the exact facts that caused it to be included. Do not invent a reason for non-payment and do not recommend write-off, provisioning or customer action.

7. MODEL CHECK
For every number in your answer, tell me which column(s) and filtering condition(s) you used. If a requested conclusion cannot be supported from the file, say what additional data you would need instead of guessing.