AI-102 Practice Question: Implement knowledge mining and document intelligence solutions
A company uses Azure Document Intelligence with a custom neural model to extract data from purchase orders. The model was trained on 50 labeled samples and performs well on similar documents. However, when processing new purchase orders from a different supplier, the model fails to extract the 'TotalAmount' field accurately. What should you do to improve the model's performance on the new supplier's documents?
⚠ Common exam trap
The trap here is thinking that adjusting confidence thresholds or switching to prebuilt models can fix extraction errors when the model lacks training data for the new layout.
Answer choices
Why each option matters
Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.
Correct answer & explanation
✓
Add labeled samples from the new supplier's purchase orders and retrain the model.
To improve extraction for documents from a new supplier, the custom neural model needs exposure to that supplier's document variations. Adding labeled samples and retraining is the direct method to enhance the model's field extraction accuracy. Other options do not address the model's knowledge gap.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✓
Add labeled samples from the new supplier's purchase orders and retrain the model.
Why this is correct
Custom neural models learn from the training data. Adding labeled samples that represent the new supplier's document variations helps the model generalize to those layouts and field positions. Retraining incorporates this new data, improving accuracy for the 'TotalAmount' field on similar documents from that supplier.
- ✗
Increase the model's confidence threshold for the 'TotalAmount' field.
Why it's wrong here
Adjusting the confidence threshold only changes when the model reports a value as extracted; it does not improve the model's ability to recognize the field. If the model fails to extract accurately, raising the threshold may cause more missing values, not better accuracy.
- ✗
Use the model's composed model feature to combine it with a prebuilt model.
Why it's wrong here
Composed models allow you to combine multiple custom models into one, but they do not inherently improve accuracy for a new supplier unless the individual models are trained on that data. Without adding labeled samples from the new supplier, composing will not resolve the extraction problem.
- ✗
Switch to the prebuilt invoice model and map the 'TotalAmount' field to 'InvoiceTotal'.
Why it's wrong here
Prebuilt models are trained on general invoice formats and may not handle the specific purchase order layout from the new supplier. Mapping fields does not address the underlying extraction issue, and the prebuilt model might not include a 'TotalAmount' field or may extract incorrectly.
Go deeper
Related to this question
About these practice questions
One of 761 original AI-102 practice questions on Courseiva, each with a full explanation and wrong-answer analysis — not exam dumps or protected exam content. Learn why practice questions differ from exam dumps →
JA
Written and reviewed by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
Last reviewed September 2026 · checked against the official Microsoft exam blueprint
This AI-102 practice question is part of Courseiva's free Microsoft certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the AI-102 exam.