AI-103 Practice Question 91: Provisioned Throughput Steady Traffic
Microsoft Azure AI Apps and Agents Developer Associate (AI-103) · Test 10 · question 1 of 10
A logistics company's dispatch assistant runs around the clock and answers drivers in real time. Request volume is unusually stable (close to the same number of requests per minute, hour after hour, day after day), and dispatch leadership is willing to commit budget in exchange for a guarantee that responses won't slow down or get throttled as call volume grows. They ask for the deployment approach that best matches this traffic shape.
Which Microsoft Foundry deployment approach should they choose?
Show answer and explanation
Answer:
Microsoft's own guidance names this exact traffic shape: provisioned throughput is the right choice when an application has "predictable traffic patterns," "latency-sensitive requirements," "production-scale volume," and "real-time or interactive scenarios... where variable response times degrade user experience." A provisioned deployment holds a fixed amount of capacity exclusively for the workload, billed hourly per PTU regardless of tokens consumed, with a defined latency target, trading a locked-in cost for a locked-in guarantee, which is exactly the trade dispatch is asking for.
- A is wrong because standard deployments are explicitly positioned as "the better fit for development, testing, low-volume usage, or highly variable traffic", the opposite of this scenario, and standard carries no latency SLA, so nothing stops it degrading under sustained high volume. - C is wrong because Global Batch has no latency SLA at all and returns results asynchronously; drivers waiting on a dispatch answer cannot use an interface built for bulk, non-urgent processing. - D is wrong because Data Zone Standard is still pay-per-token with no latency SLA; it only changes where requests are routed geographically, which was never the stated requirement.
Reference: Provisioned throughput for Foundry Models