During the quality assurance and testing phase of a digital assistant (agent), latency and token metrics are crucial for evaluating non-functional requirements. Latency directly measures the response wait time from the user's perspective; ensuring this is within an acceptable range is critical for a positive user experience. Tokens are the units of text processed by the underlying Large Language Model (LLM). Since LLM services are typically priced per token, monitoring token usage is essential for understanding and managing the operational financial cost of the agent.

A. Time saved and user adoption are business-level outcomes measured post-deployment, not direct technical metrics like latency or tokens used in pre-deployment testing.

C. Accuracy relates to the correctness and relevance of the agent's response content, which is a separate functional quality metric, not directly measured by latency or token count.

1. Oracle Cloud Infrastructure Documentation

"Generative AI - Pricing": This official pricing guide explicitly states that the service is billed based on the number of input and output tokens. This directly links the "tokens" metric to the "financial cost of running the agent".

Source: Oracle Cloud Infrastructure Official Documentation.

Section: Generative AI > Pricing.

2. Oracle Digital Assistant Documentation

"The Insights User Interface - Conversation-Quality Metrics": This document details the built-in analytics for Oracle Digital Assistant

which includes "Average Skill Response Time (ms)". This metric is a direct measurement of latency

used to confirm that the "agent response wait time is acceptable".

Source: Oracle Help Center

Oracle Digital Assistant Documentation.

Section: Use > Monitor and Analyze Your Digital Assistant > The Insights User Interface > Conversation-Quality Metrics.

3. Oracle Digital Assistant Documentation

"Use the Invoke Large Language Model Component": This section explains how the digital assistant integrates with LLMs. The interaction described is the process that consumes tokens and incurs latency. Testing this component involves evaluating its performance (latency) and resource consumption (tokens)

which translates to cost.

Source: Oracle Help Center

Oracle Digital Assistant Documentation.

Section: Develop > Integrate with Large Language Models > Use the Invoke Large Language Model Component.