Why are confidence scores useful in extraction?
Why D is the answer
They route uncertain results to a person
Microsoft notes confidence scores support straight-through processing with less manual review.
In detail
- Confidence scores show which results are uncertain.
- Those can go to a person.
- Set the threshold from real results.
Use case
An illustrative scenario, not a real incident.
Most fields extract cleanly, but a smudged total scores low. It goes to a person, and the rest flows through.
See it in different ways
flowchart TD
A{"Source content"} -->|"Structured documents, exact fields"| B["Document Intelligence"]
A -->|"Documents, images, audio, video"| C["Content Understanding"]
B --> D["Structured output"]
C --> D
D --> E["Search index or agent"]stateDiagram-v2 [*] --> Extract Extract --> Review: low confidence Extract --> Index: high confidence Review --> Index: corrected Index --> [*]
flowchart TD
Q{"How confident is the extraction?"}
Q -->|"High"| A0["Straight through"]
Q -->|"Low"| A1["Human review"]Key point: Confidence scores route uncertain results.
Why the other options are wrong
- A. They raise the quota on the deployment
- They do not affect quota.
- B. They encrypt the output of the service before it reaches storage
- They do not encrypt anything.
- C. They set the price of each document
- They do not set prices.