AIP-C01 practice question: Cost-effective model choice and tiered usage
Operational Efficiency and Optimization for GenAI Applications A free question from our original AIP-C01 practice exam. Try it, then open the answer to see why each option is right or wrong.
A music streaming service's help desk sends about 1.2 million customer questions a day to a model on Amazon Bedrock. Roughly 70 percent are simple questions about billing dates, password resets and plan limits, and the rest need multi-step troubleshooting of playback and device problems.
The service currently sends every question to its largest model, and its bill has grown faster than its usage. Simple answers must stay accurate and come back within a few seconds, and troubleshooting answers must keep their current quality.
The team wants the cost per question to fall without any drop in quality on the complex questions. It is willing to add one routing step, but it does not want to train or host a new model.
Which TWO actions should the team take? (Choose two.)
Choose two answers.
AA. Add a routing step that sends questions about billing, passwords or plan limits to a smaller model in the same family, and sends all other questions to the largest model.
BB. Keep the largest model for every question, and cache the answers to repeated questions in Amazon ElastiCache, so that each new question is still answered by the most capable model.
CC. Before routing, run an Amazon Bedrock evaluation job that compares the smaller model with the largest one on real simple questions, and route only categories that pass.
DD. Move every question to the smallest model in the family, and send a customer to a human agent whenever the model's reply includes a troubleshooting step that it is unsure about.
EE. Fine-tune the largest model on past tickets so that it writes shorter replies, and keep sending every question to that model, because shorter output tokens should lower the cost of each question.
Show the answer and explanation
Answers: A, C
A. Right: Simple questions go to a smaller, cheaper model, while troubleshooting keeps the largest model, so quality on complex questions is preserved. The routing rules are plain code that the team can adjust without training or hosting anything.
B. Wrong: Caching helps only when questions repeat word for word, and most help desk questions are phrased differently, so the cost per question barely changes. Every new question still runs on the largest model.
D. Wrong: The smallest model is cheapest, but troubleshooting answers would lose quality, which the requirement rules out. Sending many customers to human agents also adds a large staffing cost.
E. Wrong: Fine-tuning adds training cost, and every question still runs on the largest model, so the cost per question does not fall the way the requirement asks. The option does not use routing at all. Learn more: Customize your model (opens in a new tab)
Explanation
Routing simple question categories to a smaller model in the same family lowers the average cost per question, while complex troubleshooting stays on the most capable model. An evaluation on real simple questions first confirms that the smaller model keeps those answers accurate. Caching rarely helps with varied phrasing, and moving everything to the smallest model or fine-tuning the largest one breaks a stated requirement.
Want more on Cost-effective model choice and tiered usage? The 5 full-length practice sets have 5 more Cost-effective model choice and tiered usage questions, 375 in all, none of them in the free exam. One payment of $14.99, no subscription. See the practice sets →