AI Agent for Communication: Handle Customer Inquiries and Bookings

Let’s start building your AI agent. Communication & Booking Agent will process inquiries, coordinate bookings, and handle scheduled messages and follow-ups. You'll approve customer messages and booking commitments before they go out. Let's build the complete workflow step by step.

By the end of the lesson, routine conversations can keep moving while every customer-facing message and business commitment stays under your control.

Choose the Right Job For an Agent

Customer inquiries may arrive through the same channels, but they don’t all follow the same route. The next step changes with what the customer says, what the business already knows, and what it can currently offer.

Before you set up the agent, check which parts of this workflow make an agent useful.

Let's check out the nail salon example, which shows how these features work together in one customer conversation.

That is the workflow you’re about to build: move a routine inquiry toward a booking or a clear owner handoff, using the information and authority the business has provided. Start by defining what the agent may use and do, which decisions stay with you, and how it knows the inquiry is finished.

Set Up Your Agent

Step 1. Define the Agent's Job and Boundaries

Use the builder and setup method you chose in the previous lesson. Now decide what the agent is responsible for — and which job must remain with you.

Capture these decisions in an agent setup plan, which is also called an agent contract. It defines the result the agent should work toward, the information and actions it may use, where your approval is required, when it should hand the work to you, and how it knows the job is finished.

You'll use the same plan whichever setup method you choose:

• With direct on-screen setup, fill in the agent's instructions, information sources, actions, approvals, and finish conditions.
• With text or voice setup, include these decisions in your description, then check that the builder's proposed setup matches them.

The setup plan for the agent in this lesson is:

Your builder may use different names for these settings. For example, information may appear under Knowledge or Sources, while connected actions may appear under Tools or Integrations. So, match each decision to what the setting does.

Now use this plan with the setup path you chose above. After the initial setup is created, you'll use the same plan to connect the right information and actions, set approvals, and test the complete workflow.

Check the prompts for each setup method below. Before you use a prompt, replace every bracketed phrase with the decisions from your agent contract.

Path A: Configure the Agent Directly

In a direct-setup builder, you add the agent's main instruction yourself. Then you connect its business information and actions, choose how it starts, set approvals, and test the complete workflow.

Path B: Describe the Agent in Text or by Voice

In a conversational builder, you explain the business job in everyday language. The builder turns that description into a proposed setup. You then review its instructions, connections, triggers, approvals, and finish points before you test anything.

These two prompts are different on purpose. The creation brief tells a conversational builder what to create. The core instruction tells the finished agent how to behave each time it works. Whichever path you choose, you will confirm the same foundation next.

Step 2. Connect Information And Actions

Connect only the business sources and accounts this job needs. Keep payment details inside your booking or payment system; the agent should never ask a customer to send card details in a message.

Set up these capabilities using the terms in your chosen builder:

If the schedule only starts the agent, also connect the customer system it will use to find follow-ups due that day. Keep services, appointment lengths, booking rules, and approved message templates in the source the agent is allowed to use.

Use dedicated test records while you build. Give each connection only the access required for this job.

Step 3. Set Approval For Each Action

Your builder may use different words for permissions. What matters is the rule behind each action.

Let approved read actions, such as finding the client record or checking live availability, run automatically. Keep record updates under approval while you test; later, you can allow updates only to the fields you have verified. Require your approval before every customer message and booking commitment.

For an event-triggered run, make sure you know where the approval request will appear. Add an owner alert if needed. An approval gate works only when you can see it and respond.

You now have the foundation of a working agent: a defined job, trusted information, connected actions, and clear approval rules. Next, you’ll give the agent a state system so it can continue the same conversation.

Make the Agent Act at the Right Moment

Use Status to Guide the Next Step

In the previous lesson, the Marketing Campaign Agent could return to the same opportunity when facts changed, or you requested a revision, rather than start a new proposal from scratch.

A state is the saved stage of a job. It tells the agent what has already happened, what the work is waiting for, and which action may come next.

Customer conversations also need this continuity: an inquiry may pause while the agent waits for missing information, your approval, a customer reply, or a scheduled follow-up.

For this Communication & Booking Agent, save the state as a status in the existing inquiry or client record. When the work restarts, the agent reads the current status, keeps what is already known, and continues from the right point.

Use the current status field in your customer system, or add one status field to the existing inquiry record. Match your current labels to the meanings below. Allow the agent to change that status and only the other approved fields.

Each label below shows how far the inquiry has progressed and which next action the agent may take.

Owner approval is a checkpoint, rather than a customer state. The inquiry keeps its current state while the external action waits for your decision.

Build the On-Request Path First

Now turn the states above into instructions the agent can follow. Add the prompt below to the agent's instructions, replacing each bracketed phrase with the matching field, state, or rule from your setup.

Add the state table to the agent’s instructions or workflow. Then add the prompt below so the agent knows how to save and update the current state.

Teach the Agent How the Conversation Moves

Now test these instructions before adding automatic triggers. Start the agent yourself and give it the first inquiry. For example, the nail salon can tell the agent: the client wants a gel manicure next Friday after 4:00 PM but hasn't said whether gel removal is needed. The owner’s job is to check whether the agent keeps up with the details, changes the status to Needs information, comes up with the question to the client, and waits for the approval. They should review the agent's activity and any changes it makes. If everything works correctly — the first version is ready.

This is the on-request version of the agent. It can now understand where the conversation stands and continue from the same record. It’s already useful for inquiries you want to review. Next, you’ll give it the events that restart the workflow at the right moment.

Start the Agent Automatically

Add an Event Trigger

The on-request path now works. Next, give the same agent two automatic ways to resume the job: when a new message or reply arrives, and when a scheduled follow-up becomes due.

Let a business event start the same job. Connect a dedicated test inbox, test form, or test-only channel, and choose its new-message event as the trigger. Turn the trigger on, then send the test inquiry through that channel. Confirm that the event starts the agent and opens the correct inquiry.

Keep every customer-facing message, record change, and booking commitment behind approval while you test. The trigger is active, but the agent still waits for your go-ahead before taking an external action.

If your tool treats new inquiries and replies as different events, add and activate a second trigger for replies. At the beginning of each run, have the agent find the open inquiry using the stable customer or conversation ID supplied by the connected app.

Add the Scheduled Follow-Up

A message event restarts the work when the customer does something. A schedule restarts it when follow-up work becomes due, even if no new message has arrived. When an approved question or option is sent, and the inquiry moves to Waiting for customer, save the next follow-up time in the existing customer record. Use an existing follow-up field when your system already has one.

If your agent builder does not include scheduling, a separate automation tool can start the agent at the chosen time.

Turn on a scheduled trigger that checks for follow-ups whose time has arrived. Before preparing anything, the agent reads the latest conversation and confirms the customer hasn't replied, booked, declined, or opted out.

For this lesson, use one follow-up only. After it is approved and sent, clear the time and move the inquiry according to your business rule. A reply, booking, decline, opt-out, or move to human review also clears the scheduled follow-up.

The agent now knows how to start, continue, pause, finish, and bring you in. The final step is proving that each route works.

Test the Complete Agent

Test the Decision And the Action

A polished message is one part of a successful run. Your test should also prove that the agent selected the right state, used current information, took the intended action, and respected the approval point.

Keep the event and scheduled triggers active only for your dedicated test source and test records. Run these four cases:

Finish with one end-to-end run. Confirm that removal is needed, approve the 5:30 PM option, and have the client accept it. Let the agent check the slot again, then approve the booking action. Verify that Booked appears only after the calendar confirms the appointment. Then check that the confirmation matches the approved template and still waits for your approval.

Then run the scheduled trigger for an inquiry in Waiting for customer. Confirm that its follow-up waits for approval and that a reply or opt-out closes that path.

Improve the Part That Needs Attention

Each test result gives you a clue about which part of the agent needs attention. Use the activity history or run log and compare the business record before and after the run.

• If the agent chooses the wrong next step, improve the instruction or decision rule.
• If it uses the wrong facts, correct the trusted source or required input.
• If the decision is right but the action fails, check the connection, permission, or field mapping.
• If the work repeats, skips a step, or gets stuck, check the trigger, saved state, and finish rule.

Ask AI to Help You Improve the Setup

AI can help you find the part of the workflow that needs attention. Give it the expected result, the actual result, and evidence from the run. Use a dedicated test record or remove sensitive details before sharing a log in a separate AI chat.

Path A: Diagnose a Directly Configured Agent

Use this prompt when you manage the agent through settings, forms, or a visual workflow. AI will recommend a change, and you will update the relevant setting yourself.

Path B: Improve an Agent Through Text or Voice

Use this prompt when the builder can review and update the agent through conversation. The builder should show you the proposed change before applying it.

Whichever path you use, change one part at a time. Rerun the case that exposed the problem, then rerun one case that already passed. This shows whether the change solved the issue while keeping the rest of the workflow working.

Move into supervised use

Your Communication & Booking Agent is ready for supervised use when it can:

• Start from a direct request, new-inquiry event, and scheduled follow-up
• Keep one client record current as the conversation develops
• Prepare the right next message and wait for approval
• Create a booking only after the live calendar confirms it
• Route owner decisions and tool interruptions with useful context

After all these checks are passed, you’re ready to use your agent on a platform that fits you best. You already made this decision for your first agent. Use the same approach here: keep this agent in the builder if that platform fits your work, or ask how to import or connect it to another platform.

Once the agent is saved in the platform where you’ll use it, connect one approved live inquiry source and activate the triggers you tested. Keep customer-facing messages, record changes, and booking commitments behind approval while you review its early activity. If the agent selects the wrong record or state, misses a handoff, or acts outside a rule, pause the live trigger and return to test records.

Look at what you have built. Your second agent is ready for supervised work! It has a clear job, trusted information, connected actions, ways to start, and a saved state that lets it continue the same conversation. You tested various paths to make sure that your agent works as you intended.

This agent is the next working part of your AI-powered business system. Every agent in this course follows the same build flow:

Define the job and what “done” means → connect trusted information and actions → set permissions, approvals, stop rules, and handoffs → decide where progress is saved → choose what starts the work → test and improve → choose where you will use the agent and move it into supervised use.

In the next lesson, you’ll build a Weekly Planning Agent that helps plan the week and checks that every priority has the information and preparation it needs. Your AI-powered business system will then support timely marketing, customer communication, and the work required to deliver on your commitments.