AI Guidance Best Practice

Discover how to leverage AI Guidance efficiently to enhance performance and ensure more relevant responses.

With AI, you can customize its responses to customers, making sure they are precise, reflect your brand voice, and stay consistent with your support policies. To achieve optimal outcomes, use the following best practices when crafting guidance for AI.

Start with the outcome in mind

Before creating AI Guidance, think about the specific outcome you aim to achieve. Then, work backwards from this goal to develop clear and actionable instructions.

❌ Bad example (vague and all in one-sentence)

“Know about our different tiers of business plans before answering.”

✅ Good example (clear and with steps)

"If a customer asks about buying business plans, first ask them what their specific needs are, and then recommend based on the specific content and needs of different plans after matching them."

Use accurate and straightforward wording

Be direct, specific, and easy to understand.

❌ Bad example (unclear)

"Be friendly and professional."

✅ Good example (specific and actionable)

"Adopt a tone that is both professional and friendly. Keep replies brief, avoid jargon, and use comforting language to ease customer concerns."

Provide concrete examples

Use words like “if,” “when,” and “then” to define conditions, and include clear examples.

❌ Bad example (talk in generalities)

"When customers ask about refunding, answer correctly."

✅ Good example (detailed and with examples)

"If a customer asks about refunding, first check if they mention a specific reason. If they don't, ask why they want a refund."

Speak directly to AI with your guidance

Don’t talk about AI as an third person or suggest how it should change its replies. Instead, speak directly to the AI, giving it clear, direct instructions on what actions to take.

❌ Bad example (indirect)

"If the AI can not find answers in the knowledge base, let the AI tell customers that it is going to transfer the conversation to a human agent."

✅ Good example (direct)

"If can't find answers in the knowledge base, tell customers that the conversation will be tansfered to a human agent."

Ensure the consistency of the language

❌ Bad example (multiple languages mixed in one guidance)

"If the customer asks about refund-related matters and doesn't mention the specific reason for the refund, ask '¿Por qué quieres un reembolso?'"

✅ Good example (keep the consistency of the language)

“If the customer asks about refund-related matters and doesn't mention the specific reason for the refund, ask why he needs a refund.”

Continuously fine-tune your guidance

Just like training your employees, guidance is an ongoing process.

• Always test AI responses first in the preview window to confirm that it works as you expected.

• Check customers' feedback regularly and decide how to refine your guidance.

• Remove or update conflicting or redundant guidance to keep AI's behavior consistent.