AI-103 Practice Question 50: Custom Neural Model Varied Layouts

Document Intelligence's own model-selection guide places "structured and semi-structured" training sets under the custom neural model, and its own example of a semi-structured document type is invoices, the closest analog to this company's purchase orders. Neural models are trained on labeled samples spanning the layout variation and are built to generalize across that variation from a single model, which is exactly the "one endpoint, many supplier layouts" requirement here.

- A is wrong because prebuilt models have fixed schemas that Microsoft defines and maintains; there's no field for a company-specific approval code, and the prebuilt invoice model targets invoices, not purchase orders, so neither the schema nor the document type matches. - B is wrong because the guide places custom template models against "structured, consistent documents with a static layout." A single static layout is exactly what this company doesn't have across 30 suppliers; a template model trained on one layout won't generalize to the others. - D is wrong because composed models are for combining several models each trained on a genuinely different document type under one endpoint; the guide's own example is supply, equipment, and furniture purchase orders as distinct types. That's not this scenario: it's one document type with varying layout, which one neural model already generalizes over without training and maintaining a model per supplier.

Reference: Choose the best Document Intelligence model for your applications and workflows