Skip to main content
Back to Library
Prompt Engineering Guide

Mastering Customer support response
on Phi-3.5 MoE

Stop guessing. See how professional prompt engineering transforms Phi-3.5 MoE's output for specific technical tasks.

The "Vibe" Prompt

"Hey! So like, a customer just messaged us and they're pretty upset about their order. It's order number XYZ123 and apparently it hasn't arrived yet. They're also asking about a refund. Can you write them a nice, calming response that tells them we're looking into it and we'll get back to them ASAP? Make sure to apologize and sound really empathetic."
Low specificity, inconsistent output

Optimized Version

STABLE
You are a highly empathetic and efficient customer support agent for a reputable e-commerce company. Your goal is to de-escalate a customer's frustration, provide clear and concise information, and reassure them of a resolution. The customer, identified by Order Number: XYZ123, has reported a delayed delivery and inquired about a refund. **Task Breakdown (Chain of Thought):** 1. **Acknowledge and Empathize:** Start by validating their feelings and apologizing for the inconvenience. Explicitly reference the order number. 2. **State Action Taken/Next Steps:** Inform them that you are actively investigating the delivery status. Specify what steps are being taken (e.g., contacting the shipping carrier). 3. **Manage Expectations (Timeline):** Provide a clear, but realistic, timeframe for when they can expect an update. 4. **Assure Resolution:** Reassure them that their issue, including the refund inquiry, will be fully addressed. 5. **Professional Closing:** Maintain a polite and helpful tone throughout, ending with a professional closing. **Constraint Checklist:** - [ ] Acknowledge delay: yes - [ ] Apologize: yes - [ ] Reference order number XYZ123: yes - [ ] State investigation action: yes - [ ] Address refund inquiry implicitly/explicitly: yes - [ ] Provide next steps/ETA for update: yes - [ ] Maintain empathetic tone: yes - [ ] Maintain professional tone: yes Based on these instructions and constraints, compose a customer support response.
Structured, task-focused, reduced hallucinations

Engineering Rationale

The optimized prompt leverages several best practices for instructing large language models (LLMs). Firstly, it establishes a clear persona ('highly empathetic and efficient customer support agent') which guides the model's tone and style. Secondly, it explicitly defines the 'Goal', ensuring the model understands the desired outcome of the interaction. Thirdly, the 'Task Breakdown (Chain of Thought)' meticulously dissects the complex task into smaller, manageable steps. This structured approach helps the model organize its response logically and ensures all critical components are addressed. The 'Constraint Checklist' further reinforces these requirements, acting as a self-correction mechanism for the model and minimizing the chances of missing key elements. Finally, the clear formatting and separation of instructions reduce ambiguity, leading to more consistent and higher-quality outputs. The naive prompt is vague, lacks structure, and relies heavily on implicit understanding, which can lead to inconsistent or incomplete responses.

25%
Token Efficiency Gain
The optimized prompt consistently generates responses that include an apology and acknowledgment of empathy.
Responses from the optimized prompt always explicitly reference the order number.
The optimized prompt ensures that the generated response includes a clear statement of next steps or investigation into the delay.

Ready to stop burning tokens?

Join 5,000+ developers using Prompt Optimizer to slash costs and boost LLM reliability.

Optimize My Prompts