Question 22
Domain 5: Implement Computer Vision and Document Intelligence SolutionsYou have a Document Intelligence solution that processes expense reports. The reports vary by department — some departments use a table format while others use a list format. Accuracy is poor across all departments. What is the best solution?
Correct answer: C
Explanation
Custom Neural models learn layout-specific patterns, so separate models for the table and list formats improve extraction accuracy when departments use different structures. Composing them lets Document Intelligence route each report to the model that matches its layout, which is the right approach when "accuracy is poor across all departments" due to format variation.
Why each option is right or wrong
A. Use prebuilt Receipt model for all departments
B. Train one Custom Template model on all layouts
C. Train separate Custom Neural models per department layout and compose them
Azure AI Document Intelligence Custom Neural models are trained on labeled examples and are optimized to learn the visual/layout patterns present in a specific document structure; when one department uses a table-heavy expense layout and another uses a list-based layout, a single model is forced to generalize across incompatible formats, which depresses field extraction accuracy. Training distinct models for each layout and then composing them allows the service to automatically evaluate the document and route it to the best-matching model, which is the correct way to handle heterogeneous templates when accuracy is poor across all departments.
D. Use a single Custom Neural model trained on all department layouts