PRODUCTION SCENARIO
A logistics company asks an Amazon Bedrock model to compute the cheapest shipping plan across three carriers that each apply tiered discounts. The model states a total immediately, and the total is usually wrong.
Which change to the prompt text is MOST likely to improve the answers?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: Instruct the model to work through the calculation step by step.
Guiding a model to reason before it answers is chain-of-thought prompting, which AWS recommends for multi-step analysis and math problems. Temperature 0 makes the same wrong total repeatable rather than correct. Splitting the request across carriers changes the application rather than the prompt text, and the model still skips tiers inside each call. Examples of finished quotes demonstrate the jump straight to a total.