Once you have mastered the 4-part framework and built working prompts for your Conversation AI or Voice AI agents, it is time to level up. This guide covers advanced techniques that help your AI agents handle complex scenarios, maintain consistent tone, and deliver results that rival human team members.
Multi-step prompting
Many business tasks are too complex for a single prompt instruction. Multi-step prompting breaks a complex task into a sequence of smaller steps the agent follows in order.
When to use multi-step prompts
Use multi-step prompting when your agent needs to:
- Collect multiple pieces of information before taking action
- Follow a decision tree based on user responses
- Complete a workflow with validation checkpoints
- Handle a process that has a specific required order
How to structure multi-step prompts
Number each step clearly and include transition language so the agent moves naturally from one step to the next.
Multi-step prompt example
LEAD QUALIFICATION PROCESS:
Follow these steps in order. Do not skip steps.
Step 1 — INTRODUCTION
Greet the prospect warmly and ask what they are looking for.
"Hi there! Thanks for reaching out to Apex Marketing. I'd love to
learn a bit about your business so I can point you to the right
solution. What kind of marketing help are you looking for?"
Step 2 — UNDERSTAND THE NEED
Based on their response, ask one clarifying question to understand
their primary goal. Do not list all your services — focus on what
they mentioned.
Step 3 — QUALIFY BUDGET
Once you understand their need, ask about budget naturally:
"To make sure I recommend the right package, do you have a monthly
marketing budget in mind? Our plans typically start around $500 per
month."
Step 4 — QUALIFY TIMELINE
Ask about their timeline:
"And when are you hoping to get started? Are you looking to launch
something in the next couple of weeks, or are you still in the
planning phase?"
Step 5 — CAPTURE CONTACT INFO
"This sounds like a great fit for our team. Let me have one of our
strategists reach out to you with a personalized plan. Could I get
your name, email, and the best phone number to reach you?"
Step 6 — CONFIRM AND CLOSE
Read back their information and let them know what to expect next:
"Perfect — I've got [name], [email], and [phone]. One of our
strategists will reach out within 24 hours with some ideas tailored
to your goals. Is there anything else I can help with?"
Include explicit instructions like “Do not skip steps” and “Complete Step 2 before moving to Step 3” to prevent the agent from rushing through the qualification process.
Using knowledge base references in prompts
Your Knowledge Base stores the detailed information your agent needs — pricing, FAQs, policies, product specs. Your prompt should tell the agent exactly how and when to use this knowledge.
Direct the agent to the knowledge base
Knowledge base reference pattern
ANSWERING QUESTIONS:
- When a customer asks about pricing, services, hours, policies,
or product details, ALWAYS check your knowledge base first.
- If the information is in your knowledge base, use it to answer
accurately. Do not paraphrase in a way that changes the meaning.
- If the information is NOT in your knowledge base, say: "I want
to make sure I give you the right answer on that. Let me connect
you with our team."
- Never combine knowledge base information with guesses to fill
in gaps. If you only have partial information, share what you
know and offer to connect them with someone who has the full
picture.
When to reference the knowledge base vs. the prompt
| Put in the prompt | Put in the knowledge base |
|---|
| Personality and tone instructions | Product and service details |
| Workflow steps and processes | Pricing tables and packages |
| Escalation and handoff rules | Business hours and locations |
| Response length and format rules | FAQs and common answers |
| Channel-specific guidelines | Policies (return, cancellation, etc.) |
As a rule of thumb, if the information changes frequently (pricing, hours, availability), put it in the knowledge base. If the instruction is about how the agent should behave, put it in the prompt.
Conditional prompt logic
Conditional logic lets your agent respond differently based on what the user says. This is one of the most powerful techniques for creating agents that feel intelligent and responsive.
If-then patterns
Conditional logic examples
CONDITIONAL RULES:
IF the customer mentions an emergency (keywords: "emergency,"
"urgent," "broken," "leak," "flood," "no heat," "no power"):
- Prioritize their request immediately
- Skip the standard qualification steps
- Provide the emergency contact number: (555) 911-0000
- Say: "That sounds urgent — let me get you help right away."
IF the customer mentions a competitor by name:
- Do not disparage the competitor
- Acknowledge their comment neutrally: "I appreciate you sharing
that."
- Redirect to your differentiators: "What sets us apart is
[key differentiator]. Would you like to learn more about that?"
IF the customer says they are "just browsing" or "not ready yet":
- Do not push for a sale or appointment
- Offer helpful information: "No problem at all! I'm happy to
answer any questions you have. And whenever you're ready, we're
here."
- Offer to send information via email for later reference
IF the customer mentions a price concern or says "too expensive":
- Acknowledge the concern: "I understand — budget is important."
- Highlight value rather than discounting: "Our plan includes
[key benefits]. Many clients find the ROI covers the cost
within the first month."
- Offer a lower-tier option if available
- If still hesitant, offer to schedule a call with sales to
discuss flexible options
Nested conditions
For more complex scenarios, you can nest conditions:
IF the customer wants to book an appointment:
Ask what type of service they need.
IF they need a standard service (cleaning, consultation):
Proceed with normal booking flow.
IF they need an emergency service:
IF it is during business hours (Mon-Fri, 8 AM - 5 PM):
Offer same-day appointment slots.
IF it is outside business hours:
Provide the emergency hotline number and offer to book
the earliest available slot for the next business day.
IF they are unsure what service they need:
Ask 2-3 clarifying questions about their situation, then
recommend the appropriate service type.
Tone calibration strategies
Getting the right tone is critical. Too formal and your agent sounds robotic. Too casual and it may seem unprofessional. Here are strategies for dialing in the perfect tone.
Tone spectrum
Define where your agent falls on the spectrum and give concrete examples:
YOUR TONE: Warm professional
This means:
- Use contractions naturally ("I'm," "you'll," "we'd")
- Use first names when the customer provides them
- Be empathetic but not overly emotional
- Use conversational transitions ("Sure thing," "Great question,"
"Happy to help")
- Avoid corporate jargon ("leverage," "synergy," "circle back")
- Avoid slang or overly casual language ("yo," "dude," "no worries
bro")
TONE EXAMPLES:
Too formal: "I would be delighted to assist you with scheduling
a consultation at your earliest convenience."
Too casual: "Yeah totally, let's get you booked! When works?"
Just right: "I'd be happy to help you schedule a consultation.
Do you have a preferred day or time?"
Too formal: "We regret to inform you that this falls outside the
scope of our current service offerings."
Too casual: "Sorry, can't help with that one!"
Just right: "That's a great question, but it's outside what I can
help with. Let me connect you with someone on our team."
Adapting tone to the situation
Situational tone adjustment
TONE ADJUSTMENTS:
When the customer is frustrated or upset:
- Shift to a more empathetic tone
- Validate their feelings before problem-solving
- Example: "I completely understand your frustration, and I'm
sorry for the inconvenience. Let's get this sorted out."
When the customer is excited or enthusiastic:
- Match their energy slightly (but stay professional)
- Example: "That's great to hear! I think you're going to love
this. Let's get everything set up for you."
When the customer is confused:
- Shift to a patient, clear tone
- Break information into smaller pieces
- Example: "No worries — let me walk you through it step by step."
When the customer is in a hurry:
- Be brief and direct
- Skip pleasantries and get to the point
- Example: "Absolutely. I just need your name and preferred time
and I'll have you booked in 30 seconds."
Handling objections and pushback
Real customers push back, negotiate, and raise concerns. Prepare your agent for the most common objections.
Objection handling patterns
COMMON OBJECTIONS AND RESPONSES:
"That's too expensive."
- Acknowledge: "I hear you — price is an important factor."
- Reframe value: "Our [plan/service] includes [key benefits],
and most clients see [specific result] within [timeframe]."
- Offer alternative: "We also have a [lower tier] option at
[price] that might be a better fit to start."
- Escalate if needed: "Would it help to speak with someone on
our team about flexible payment options?"
"I need to think about it."
- Respect: "Of course — take all the time you need."
- Provide value: "In the meantime, I can email you a summary
of what we discussed so you have it handy."
- Keep the door open: "Feel free to reach out anytime. I'm here
to help whenever you're ready."
"I'm already working with [competitor]."
- Do not criticize: Avoid any negative comments about competitors.
- Differentiate: "That's great that you have a solution in place.
What a lot of our clients appreciate about us is [unique
differentiator]."
- Plant a seed: "If you ever want to compare options, we'd be
happy to put together a quick proposal."
"Can you give me a discount?"
- Check your guidelines: Only offer discounts if authorized.
- If authorized: "I can offer [specific discount]. Would that
work for you?"
- If not authorized: "I appreciate you asking! Our pricing is
set to reflect the value and quality of our service, but I can
connect you with our sales team to discuss options."
Never instruct your AI agent to promise discounts, make commitments, or agree to special terms unless you have explicitly authorized it. Unauthorized promises can create legal and customer service problems.
Prompt versioning and A/B testing
As your AI agent handles more conversations, you will want to test different prompt approaches to see which performs better.
Version your prompts
Keep a record of each prompt version so you can track what changed and how it affected performance.
Prompt versioning example
Version 1.0 (Launch) — Jan 15
Baseline prompt with standard greeting, booking flow, and
escalation rules.
Version 1.1 (Feb 1) — Added objection handling
Added responses for "too expensive" and "need to think about
it." Result: 12% fewer handoffs to human agents.
Version 1.2 (Feb 15) — Shortened greeting
Reduced greeting from 3 sentences to 1. Result: 8% improvement
in customer engagement within first 30 seconds.
Version 2.0 (Mar 1) — Complete rewrite
Restructured prompt using multi-step qualification flow.
Result: 25% increase in qualified leads captured.
A/B testing with workflows
You can test different prompts by creating two versions of your bot and routing traffic between them using HoopAI workflows:
- Create two bot versions with different prompts (Version A and Version B)
- Set up a workflow that randomly assigns incoming conversations to one of the two bots
- Monitor performance in the Conversation AI Dashboard for each version
- Compare metrics like resolution rate, handoff rate, and customer satisfaction
- Promote the winner and iterate further
Next steps