AI-103 Practice Question 100: Invoice vs Receipt Model
Microsoft Azure AI Apps and Agents Developer Associate (AI-103) · Test 10 · question 10 of 10
An accounts-payable team ingests two kinds of paper: point-of-sale receipts employees submit for expense reimbursement (merchant name, date, tax, tip, total), and vendor bills from suppliers that carry a purchase-order number, a remittance address, payment terms such as "Net 30," and a due date to match against open purchase orders before payment. Reasoning that "they're both just a statement of money owed," someone configured a single Document Intelligence pipeline using only the prebuilt receipt model for both. Extraction quality is fine for the point-of-sale documents, but the pipeline never returns a purchase-order number, payment terms, or a due date for the vendor bills.
What is the most likely cause, and what should the team do?
Show answer and explanation
Answer:
The two prebuilt models solve two document families that only superficially look alike. The invoice model's supported document types are "invoices, utility bills, sales orders, purchase orders," and it "extracts key information such as customer name, billing address, due date, and amount due", an accounts-payable field set that includes exactly what's missing. The receipt model extracts "merchant name, merchant phone number, transaction date, tax, and transaction total", a point-of-sale field set with no purchase-order, payment-terms, or remittance concept at all, which is why those fields never appear no matter how well the receipt model performs on retail printouts.
- B is wrong because this is exactly the case where a matching prebuilt model already exists; building and labelling custom training data is unnecessary effort for a problem Document Intelligence already solves. - C is wrong because ReceiptType (for example, "Itemized") describes what kind of receipt was recognized; it doesn't unlock invoice-only fields like purchase order or payment terms that the receipt model was never trained to extract. - D is wrong because prebuilt-layout returns structural elements (paragraphs, tables, reading order) not semantic fields like due date or purchase-order number; switching to it would trade a model that already fits for one that returns no such field at all.