AI Agent Callback Scheduling Integration

Using AI Agents for scheduled callbacks in Amazon Connect.

Rachel K

Published Oct 9, 2026

Last modified Oct 9, 2026

Overview

This document describes how the Amazon Connect AI Agent (chat channel) integrates with the Scheduled Callback system. The AI Agent uses two tools, get_available_slots and schedule_callback, to offer customers the ability to schedule a voice callback at a convenient time during a chat conversation.

The AI Agent detects when a callback is appropriate (customer requests a live agent, issue is too complex for chat, or customer explicitly asks for a phone call), then guides the customer through slot selection, phone number confirmation, and reason capture before creating the reservation.

All callbacks scheduled through chat are recorded with source: "CHAT" for reporting and attribution.

How the tools reach the backend

There is no public HTTP API, no API Gateway, and no internet-facing surface in this path (see design.md, "No Public Surface", property P5). Both the AI Agent tools and the IVR flow invoke the same Scheduling Lambda through Amazon Connect's native Lambda invocation (Lex / Q tool fulfillment for chat; the "Invoke AWS Lambda function" block for IVR).

The Scheduling Lambda does not route on HTTP method or path. It dispatches on an action discriminator carried in the event:

Tool (AI Agent)	Lambda `action`	Purpose
`get_available_slots`	`GET_SLOTS`	Return bookable slots for a date
`schedule_callback`	`CREATE`	Create a reservation

The Lex / Q fulfillment layer maps each tool's parameters onto the Lambda event shape (adding action and source: "CHAT") and maps the Lambda response back into values the AI Agent reads to the customer. The tool definitions the AI Agent sees are described in "Tool Definitions" below; the raw Lambda event/response shapes are in "Lambda Event and Response Contracts".

Timezone: All times are in Central Time (America/Chicago). The AI Agent always labels times with "CT" when presenting them to the customer.

Business Hours: Monday to Friday, 7:00 AM to 6:00 PM CT (excluding holidays).

Intent Detection and Trigger Phrases

The AI Agent offers a callback (Requirement 1.1) when the customer asks for a live person or when the issue is better handled by voice. Treat the following as triggers.

Explicit callback / live-agent requests

Direct phrasings that should immediately lead to a callback offer:

"call me back", "can you call me", "give me a call", "schedule a callback"
"I want to talk to a person", "talk to a human", "speak to someone", "real person"
"connect me to an agent", "I need a live agent", "get me a representative"
"can someone phone me", "I'd rather talk on the phone"

Complexity and frustration signals

Infer that voice is the better channel, then offer a callback:

Multi-step troubleshooting that has already failed (for example account recovery, VPN authentication, device enrollment where the customer has tried several fixes).
The customer expresses frustration with chat ("this isn't working", "I've been going in circles").
The task would be faster with the agent walking through settings in real time.

Offer, do not force

If the customer declines the callback offer, continue the normal chat flow. Do not re-offer repeatedly.
A callback is an option, not a dead end. The customer can keep chatting instead.

Slot Elicitation Order

Collect information in this fixed order. Each step gates the next, so the AI Agent always has what it needs before the CREATE call.

1
date  ->  time  ->  phone  ->  reason  ->  confirm  ->  create

Date: After the customer accepts the offer, call get_available_slots for today and, as needed, the next business days (lookahead is today plus 5 business days). Present days that have open slots. If the customer names a day ("tomorrow afternoon"), check that day first.
Time: Present the open slots for the chosen day (labeled in CT) and let the customer pick one. Only slots returned by the tool are bookable; never invent a time.
Phone: Confirm or collect the callback number (see "Phone Sourcing").
Reason: Capture a brief, free-text reason (max 200 characters) to pass to the agent.
Confirm: Read back date, day of week, time (CT), and phone number, and ask the customer to confirm.
Create: Call schedule_callback only after the customer confirms.

Keep the order even when the customer volunteers information early (for example gives a time before a date). Acknowledge what they said, then still confirm the missing pieces in order before creating the reservation.

Phone Sourcing (Requirement 1.5)

The callback number is resolved as follows.

Authenticated customer (phone on file via Customer Profiles)

If the chat contact is authenticated and a phone number is available from Amazon Connect Customer Profiles, confirm it rather than asking blind:

"I have (615) 555-1234 on file. Should we call you at that number?"

If the customer says yes, use that number.
If the customer wants a different number, collect the new one (treat as the unauthenticated path below).

Unauthenticated customer, or no phone on file

Ask for the number:

"What phone number should we call you at? It needs to be a US or Canadian number."

Validation (applies to both paths)

US and Canada only (country code +1), 10 national digits.
Format as E.164 for the tool call: +1XXXXXXXXXX (for example +16155551234).
If the number is incomplete or clearly invalid, ask the customer to re-state it as a 10-digit number with area code before continuing. The backend also validates and will return INVALID_PHONE if a bad number slips through (see "Error and Status Handling").

Tool Definitions

These are the two tools the AI Agent is configured with. The fulfillment layer translates them to and from the Lambda GET_SLOTS / CREATE contracts.

`get_available_slots`

Retrieves bookable callback time slots for a specific date.

When to use: After the customer accepts the offer, call this for the day(s) you intend to present. The backend already filters to business hours, non-holidays, remaining capacity, and minimum lead time, so every returned slot is bookable.

Parameters

Parameter	Type	Required	Format	Description
`date`	string	Yes	YYYY-MM-DD	The date to check availability for (today through today + 5 business days).

Returns (slots the agent reads back)

1
2
3
4
5
6
7
8
9
10
{
  "date": "2026-07-21",
  "timezone": "America/Chicago",
  "slots": [
    {"time": "14:00", "label": "today at 2:00 PM", "remaining": 2},
    {"time": "15:00", "label": "today at 3:00 PM", "remaining": 3},
    {"time": "09:00", "label": "tomorrow at 9:00 AM", "remaining": 3}
  ],
  "correlationId": "req-8f3a2b1c"
}

Behavior notes

Only slots within business hours for that date are returned (Requirement 1.3).
Only slots with remaining capacity are returned; full slots are excluded (Requirement 1.4).
Holidays and weekends return an empty slots array (Requirement 1.3).
Slots less than 30 minutes from now are excluded (minimum lead time).
Each slot carries a label suitable for reading aloud; prefer it over reformatting the raw time.
An empty slots array means "nothing bookable that day"; offer another day.

`schedule_callback`

Creates a callback reservation.

When to use: Only after the customer has selected a slot, confirmed a phone number, given a reason, and confirmed the read-back.

Parameters

Parameter	Type	Required	Format	Description
`phone`	string	Yes	+1XXXXXXXXXX	Customer phone in E.164 format (US / Canada).
`date`	string	Yes	YYYY-MM-DD	Selected callback date.
`time`	string	Yes	HH:mm	Selected slot start (24-hour, CT). Must be a slot returned by `get_available_slots`.
`reason`	string	Yes	max 200 chars	Brief description of why the customer needs a call (Requirement 1.9).
`name`	string	No	max 100 chars	Customer's preferred name for the callback greeting.

Note: source is set to "CHAT" by the fulfillment layer for every AI Agent booking. Do not include it in the tool call.

Returns (success)

1
2
3
4
5
6
7
8
9
{
  "status": "SCHEDULED",
  "referenceId": "CB-A7X9M2",
  "scheduledDate": "2026-07-21",
  "scheduledTime": "14:00",
  "timezone": "America/Chicago",
  "spokenConfirmation": "your callback is confirmed for today at 2 PM",
  "correlationId": "req-8f3a2b1c"
}

The referenceId always has the form CB-XXXXXX (the CB- prefix plus 6 alphanumeric characters). Always share it with the customer in the confirmation (Requirement 1.7).

Response statuses

`status`	Condition	AI Agent action
`SCHEDULED`	Reservation created	Confirm with date, time (CT), and reference number.
`CAPACITY_FULL`	Slot filled between the availability check and the create call	Offer the `nextAvailable` slot (Requirement 1.8).
`INVALID_DATE`	Holiday, weekend, out-of-range, or non-business-hours date	Ask for a different day and re-check availability.
`INVALID_PHONE`	Phone failed validation (bad format, not US / Canada)	Ask the customer to re-state the number.
`ERROR`	Unexpected backend error	Apologize, offer to retry, or offer a queue transfer.

See "Error and Status Handling" for wording.

Lambda Event and Response Contracts

These are the raw shapes exchanged between the fulfillment layer and the Scheduling Lambda. The AI Agent author does not usually touch these directly, but they define exactly what the tools map to. They match scheduling/models.py and design.md.

GET_SLOTS

Event:

1
2
3
4
5
6
{
  "action": "GET_SLOTS",
  "source": "CHAT",
  "date": "2026-07-21",
  "maxSlots": 5
}

Response:

1
2
3
4
5
6
7
8
9
{
  "date": "2026-07-21",
  "timezone": "America/Chicago",
  "slots": [
    {"time": "14:00", "label": "today at 2:00 PM", "remaining": 2},
    {"time": "15:00", "label": "today at 3:00 PM", "remaining": 3}
  ],
  "correlationId": "req-8f3a2b1c"
}

CREATE

Event:

1
2
3
4
5
6
7
8
9
{
  "action": "CREATE",
  "source": "CHAT",
  "phone": "+16155551234",
  "date": "2026-07-21",
  "time": "14:00",
  "reason": "Can't access VUNet account after password reset",
  "name": "Alex Johnson"
}

Response (success):

1
2
3
4
5
6
7
8
9
{
  "status": "SCHEDULED",
  "referenceId": "CB-A7X9M2",
  "scheduledDate": "2026-07-21",
  "scheduledTime": "14:00",
  "timezone": "America/Chicago",
  "spokenConfirmation": "your callback is confirmed for today at 2 PM",
  "correlationId": "req-8f3a2b1c"
}

Response (capacity conflict):

1
2
3
4
5
{
  "status": "CAPACITY_FULL",
  "nextAvailable": {"date": "2026-07-21", "time": "15:00", "label": "today at 3:00 PM"},
  "correlationId": "req-8f3a2b1c"
}

Response (validation error):

1
2
3
4
5
{
  "status": "INVALID_PHONE",
  "message": "Phone number must be a US or Canada number in E.164 format.",
  "correlationId": "req-8f3a2b1c"
}

Note that the fields are keyed by status, not an HTTP status code. The fulfillment layer never sees a 201 / 409 / 400; it sees SCHEDULED / CAPACITY_FULL / INVALID_* / ERROR.

Conversation Flow

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
┌─────────────────────────────────────────────────────────┐
│ 1. DETECT NEED                                           │
│    Customer requests live agent / complex voice issue /  │
│    explicit callback ask (see Trigger Phrases)           │
└──────────────────────────┬──────────────────────────────┘
                           │
                           ▼
┌─────────────────────────────────────────────────────────┐
│ 2. OFFER CALLBACK                                        │
│    Propose scheduling at a convenient time               │
│    (If customer declines, resume normal chat)            │
└──────────────────────────┬──────────────────────────────┘
                           │ Customer accepts
                           ▼
┌─────────────────────────────────────────────────────────┐
│ 3. PRESENT SLOTS (date -> time)                          │
│    get_available_slots for today + needed business days  │
│    Present open times (labeled CT); customer picks one   │
└──────────────────────────┬──────────────────────────────┘
                           │ Slot selected
                           ▼
┌─────────────────────────────────────────────────────────┐
│ 4. PHONE                                                 │
│    Authenticated: confirm number from Customer Profiles  │
│    Else: ask for a US/Canada number                      │
└──────────────────────────┬──────────────────────────────┘
                           │ Phone confirmed
                           ▼
┌─────────────────────────────────────────────────────────┐
│ 5. REASON                                                │
│    Capture brief reason (max 200 chars)                  │
└──────────────────────────┬──────────────────────────────┘
                           │
                           ▼
┌─────────────────────────────────────────────────────────┐
│ 6. CONFIRM                                               │
│    Read back date, time (CT), phone; ask to confirm      │
└──────────────────────────┬──────────────────────────────┘
                           │ Customer confirms
                           ▼
┌─────────────────────────────────────────────────────────┐
│ 7. CREATE                                                │
│    schedule_callback (source=CHAT set by fulfillment)    │
└────────────┬───────────────────────────┬────────────────┘
             │ status=SCHEDULED           │ CAPACITY_FULL / INVALID_* / ERROR
             ▼                            ▼
┌─────────────────────────┐  ┌────────────────────────────┐
│ 8a. CONFIRM BOOKING     │  │ 8b. HANDLE STATUS          │
│     Reference number     │  │     CAPACITY_FULL -> offer │
│     (CB-XXXXXX), date,    │  │       nextAvailable slot   │
│     time (CT), what next  │  │     INVALID_* -> fix input │
│                          │  │     ERROR -> retry / queue │
└──────────────────────────┘  └────────────────────────────┘

Step details

1. Detect need

Trigger on explicit requests ("call me back", "talk to a person") or complexity / frustration signals. See "Intent Detection and Trigger Phrases".

2. Offer callback

Make the benefit clear: the customer picks the time, there's no hold, and they get a text confirmation and reminder.

3. Present slots (date, then time)

Call get_available_slots for today and the next business days as needed.
Summarize in a friendly way; do not dump raw slot JSON. Group by day.
Always label times "CT".
If a day returns no slots, say so and offer the next day.

4. Phone

Confirm from Customer Profiles when authenticated; otherwise ask. Validate US / Canada and format E.164. See "Phone Sourcing".

5. Reason

Ask for a short description of the problem so the agent has context. Keep it to 200 characters or fewer (Requirement 1.9).

6. Confirm

Read back date and day of week, time with "CT", and the phone number, then ask "Does that all look correct?"

7 and 8. Create and handle result

SCHEDULED: share the reference number (CB-XXXXXX), the date and time in CT, and mention the confirmation text and 15-minute reminder.
CAPACITY_FULL: tell the customer the slot was just taken and offer nextAvailable.
INVALID_DATE: ask for another day and re-check availability.
INVALID_PHONE: ask the customer to re-state the number.
ERROR: apologize, retry once, and offer a queue transfer if it recurs.

Example Conversations

Example 1: Customer requests a callback

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
Customer: I need to talk to a real person about my account access issue.

AI Agent: I understand you'd like to speak with someone about your account
access issue. I can schedule a callback from our team at a time that works
for you, with no waiting on hold. Would you like me to find available times?

Customer: Yes please.

AI Agent: [get_available_slots -> action=GET_SLOTS, date today and next days]

AI Agent: Here are some available times (all Central Time):

Today (Monday, July 21):
- 2:00 PM CT
- 3:00 PM CT
- 4:30 PM CT

Tomorrow (Tuesday, July 22):
- 9:00 AM CT
- 10:30 AM CT
- 1:00 PM CT

Would any of these work? I can check other days too.

Customer: Tomorrow at 9 AM works.

AI Agent: I have (615) 555-1234 on file as your phone number. Should we call
you at that number?

Customer: Yes, that's fine.

AI Agent: Could you briefly describe what you need help with? That gives the
agent some context before they call.

Customer: I can't access my VUNet account after a password reset.

AI Agent: Let me confirm the details:
- Date: Tuesday, July 22
- Time: 9:00 AM CT
- Phone: (615) 555-1234
- Reason: Can't access VUNet account after password reset

Does that all look correct?

Customer: Yes.

AI Agent: [schedule_callback -> action=CREATE, phone="+16155551234",
date="2026-07-22", time="09:00",
reason="Can't access VUNet account after password reset", name="Alex Johnson"]
[response: status=SCHEDULED, referenceId="CB-A7X9M2"]

AI Agent: Your callback is confirmed. Here are the details:
- Reference number: CB-A7X9M2
- Tuesday, July 22 at 9:00 AM CT
- We'll call (615) 555-1234

You'll get a text confirmation shortly, and a reminder 15 minutes before your
call. Is there anything else I can help with today?

Example 2: AI Agent proactively offers a callback (complex issue)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
Customer: I've been trying to set up the Cisco VPN on my new laptop but every
configuration gives me an authentication error. I've reset my password,
reinstalled the client, and checked my 2FA.

AI Agent: You've already done thorough troubleshooting. VPN authentication
issues like this are often faster to resolve over the phone, where our team can
walk through the settings with you in real time. Would you like me to schedule
a callback? You pick the time, and an agent calls you with no hold.

Customer: That would be great, I'm free tomorrow afternoon.

AI Agent: [get_available_slots -> action=GET_SLOTS, date=tomorrow]

AI Agent: Here's what's open tomorrow afternoon (Wednesday, July 23, CT):
- 1:00 PM
- 1:45 PM
- 2:30 PM
- 3:00 PM

Which works best?

Customer: 2:30 PM.

AI Agent: What phone number should we reach you at?

Customer: 615-867-5309

AI Agent: Thanks. Just to confirm the issue: Cisco VPN authentication errors on
a new laptop, after a password reset, client reinstall, and 2FA check. Is that
a good summary?

Customer: Yes exactly.

AI Agent: Here's what I have:
- Date: Wednesday, July 23
- Time: 2:30 PM CT
- Phone: (615) 867-5309
- Reason: Cisco VPN auth errors on new laptop

Should I book this?

Customer: Yes.

AI Agent: [schedule_callback -> action=CREATE, phone="+16158675309",
date="2026-07-23", time="14:30",
reason="Cisco VPN auth errors on new laptop - tried pw reset, reinstall, 2FA"]
[response: status=SCHEDULED, referenceId="CB-K3P8W1"]

AI Agent: All set. Your callback is booked:
- Reference: CB-K3P8W1
- Wednesday, July 23 at 2:30 PM CT
- We'll call (615) 867-5309

You'll get a confirmation text shortly, and a reminder 15 minutes before.
Anything else I can help with?

Example 3: Capacity full, re-offer next slot (Requirement 1.8)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
Customer: Let's do 3 PM.

AI Agent: [schedule_callback -> action=CREATE, time="15:00"]
[response: status=CAPACITY_FULL,
 nextAvailable={"date":"2026-07-21","time":"15:15","label":"today at 3:15 PM"}]

AI Agent: I'm sorry, the 3:00 PM slot just filled up while we were chatting.
The next available time today is 3:15 PM CT. Would that work, or would you
prefer a different time?

Customer: 3:15 is fine.

AI Agent: [schedule_callback -> action=CREATE, time="15:15"]
[response: status=SCHEDULED, referenceId="CB-R2T5N8"]

AI Agent: Done. Your callback is confirmed for today at 3:15 PM CT.
Reference number: CB-R2T5N8. You'll receive a confirmation text shortly.

Example 4: Collecting a phone number (unauthenticated customer)

1
2
3
4
5
6
7
8
9
10
11
12
13
AI Agent: What phone number should we call you at? It needs to be a US or
Canadian number.

Customer: 555-1234

AI Agent: I'll need the full 10-digit number including area code. Could you
provide that?

Customer: Oh sorry, 615-555-1234

AI Agent: Thanks. We'll call you at (615) 555-1234. Is that correct?

Customer: Yes.

Example 5: Invalid date (holiday)

1
2
3
4
5
6
7
8
Customer: Can you call me on July 4th in the afternoon?

AI Agent: [get_available_slots -> action=GET_SLOTS, date="2026-07-04"]
[response: slots=[] because it's a holiday]

AI Agent: July 4th is a holiday, so we won't have agents available that day.
The next business day with openings is Monday, July 6. Would an afternoon time
that day work?

Error and Status Handling Guide

The CREATE response carries a status. Handle each as follows.

CAPACITY_FULL, slot filled

The selected slot filled between the availability check and the booking.

AI Agent behavior:

Tell the customer the slot is no longer available.
Offer the nextAvailable slot from the response.
If the customer declines, call get_available_slots again for that date (or another day) and present fresh options.

Example: "I'm sorry, that time just filled up. The next available time today is 3:15 PM CT. Would that work?"

INVALID_DATE

The requested date is a holiday, weekend, out of range, or outside business hours.

AI Agent behavior:

Explain the day isn't available.
Offer the next business day and re-check availability.

Example: "That day isn't available for callbacks. The next open business day is Monday, July 6. Want me to check times then?"

INVALID_PHONE

The phone number failed validation (format, or not US / Canada).

AI Agent behavior:

Ask the customer to re-state the number.
Confirm it's a 10-digit US or Canada number with area code.

Example: "I wasn't able to use that number. Could you give it as a 10-digit US number, like 615-555-1234?"

ERROR

An unexpected backend error occurred.

AI Agent behavior:

Apologize for the technical difficulty.
Offer to try again.
If it recurs, offer a direct queue transfer.

Example: "I'm sorry, the scheduling system had a hiccup. Let me try that again. If it keeps failing, I can connect you to an agent directly."

Configuration Notes

Connect Instance: 52cec302-bfc8-47ba-8b91-ab622f786605
Invocation: Lex / Q tool fulfillment, which calls the Scheduling Lambda directly (no API Gateway, no function URL). source is CHAT.
Slot Duration: 15 minutes
Capacity per Slot: default 3 (Mondays and Fridays: 2)
Business Hours: Mon to Fri, 7:00 AM to 6:00 PM CT
Minimum Lead Time: 30 minutes (cannot schedule within 30 minutes of now)
Lookahead Window: today plus the next 5 business days
Reference Number Format: CB-XXXXXX (CB- prefix plus 6 alphanumeric characters)
Notifications: SMS confirmation sent immediately after booking; SMS reminder sent 15 minutes before the callback

Tool Definitions (Amazon Connect AI Agent Configuration)

These tool definitions are configured in the Amazon Connect AI Agent (Lex / Q) settings. The fulfillment layer maps them to the Scheduling Lambda GET_SLOTS / CREATE actions and injects source: "CHAT".

get_available_slots

1
2
3
4
5
6
7
8
9
10
11
12
13
14
{
  "name": "get_available_slots",
  "description": "Retrieves available callback time slots for a given date. Returns slots within business hours (Mon-Fri 7AM-6PM CT), on non-holidays, that still have remaining capacity and are at least 30 minutes out. Use this to present scheduling options to the customer.",
  "parameters": {
    "type": "object",
    "properties": {
      "date": {
        "type": "string",
        "description": "The date to check availability for, in YYYY-MM-DD format. Must be today or within the next 5 business days."
      }
    },
    "required": ["date"]
  }
}

schedule_callback

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
{
  "name": "schedule_callback",
  "description": "Creates a scheduled callback reservation. The customer receives a phone call at the specified date and time from a VUIT support agent. Sends an SMS confirmation automatically. Returns a reference number of the form CB-XXXXXX. Source is set to CHAT by the system.",
  "parameters": {
    "type": "object",
    "properties": {
      "phone": {
        "type": "string",
        "description": "Customer phone number in E.164 format: +1 followed by 10 digits (e.g., +16155551234). US and Canada only."
      },
      "date": {
        "type": "string",
        "description": "Callback date in YYYY-MM-DD format. Must be a business day with available slots."
      },
      "time": {
        "type": "string",
        "description": "Callback time slot in HH:mm format (24-hour, Central Time). Must be a slot returned by get_available_slots."
      },
      "reason": {
        "type": "string",
        "description": "Brief description of why the customer needs a callback. Maximum 200 characters. Shared with the agent who calls back."
      },
      "name": {
        "type": "string",
        "description": "Customer's preferred name, used in the callback greeting. Optional."
      }
    },
    "required": ["phone", "date", "time", "reason"]
  }
}

Requirements Traceability

Requirement	Coverage
1.1 (offer callback when a live agent is requested or the issue is complex)	Intent Detection and Trigger Phrases; Detect Need and Offer Callback steps
1.3 (present only business-hours, non-holiday slots)	get_available_slots behavior notes; Present Slots step; INVALID_DATE handling
1.4 (present only slots with remaining capacity)	get_available_slots behavior notes (full slots excluded)
1.5 (confirm phone from Customer Profiles if authenticated, else ask)	Phone Sourcing section; Phone step
1.7 (confirm booking with date, time, and reference number)	Confirm Booking step; SCHEDULED handling; CB-XXXXXX reference format
1.8 (on capacity failure, offer the next available slot)	CAPACITY_FULL handling; Example 3
1.9 (capture a brief reason, max 200 characters)	Reason step; schedule_callback reason parameter

# amazon-connect # ai # ai-agents

Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.

Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article

Builder Center

AI Agent Callback Scheduling Integration

Rachel K

How the tools reach the backend

Intent Detection and Trigger Phrases

Explicit callback / live-agent requests

Complexity and frustration signals

Offer, do not force

Slot Elicitation Order

Phone Sourcing (Requirement 1.5)

Authenticated customer (phone on file via Customer Profiles)

Unauthenticated customer, or no phone on file

Validation (applies to both paths)

Tool Definitions

get_available_slots

Parameters

Returns (slots the agent reads back)

Behavior notes

schedule_callback

Parameters

Returns (success)

Response statuses

Lambda Event and Response Contracts

GET_SLOTS

CREATE

Conversation Flow

Step details

1. Detect need

2. Offer callback

3. Present slots (date, then time)

4. Phone

5. Reason

6. Confirm

7 and 8. Create and handle result

Example Conversations

Example 1: Customer requests a callback

Example 2: AI Agent proactively offers a callback (complex issue)

Example 3: Capacity full, re-offer next slot (Requirement 1.8)

Example 4: Collecting a phone number (unauthenticated customer)

Example 5: Invalid date (holiday)

Error and Status Handling Guide

CAPACITY_FULL, slot filled

INVALID_DATE

INVALID_PHONE

ERROR

Configuration Notes

Tool Definitions (Amazon Connect AI Agent Configuration)

get_available_slots

schedule_callback

Requirements Traceability

`get_available_slots`

`schedule_callback`