Automated Customer Support
Customer questions look endless, but 80% are variations of the same handful of FAQs. Let Ada be your first-line support agent — automatic for common questions, human for the rest.
Scenario
Where This Fits
| Scenario | Description |
|---|---|
| E-commerce | Order lookup, returns, product specs |
| Services | Appointment queries, service info, pricing |
| SaaS products | Onboarding, troubleshooting, feature questions |
| Education | Course info, enrollment flow, FAQs |
Expected Results
- Faster replies — seconds instead of hours
- 24/7 coverage
- Higher satisfaction — instant responses, no waiting
- Lower labor cost — 80% of repeat questions handled
Tools You’ll Need
Required
| Tool | Purpose | Where |
|---|---|---|
| Telegram Bot | Receive and send customer messages | Admin Panel → Messaging |
| Knowledge base | Store FAQ entries | Ada’s working directory |
| Escalation notifier | Ping you when something needs a human | Admin Panel → Ada Dashboard |
Permissions
On Ada’s tab in the Admin Panel, enable:
✅ message (send/receive)
✅ read (knowledge base files)
✅ web_search (product information)
⬜ exec (off — support never needs system commands)
Recommended Persona
Ada as a Support Specialist
{`# SOUL.md — Support Specialist Ada
## Core Identity
You are **Ada**, a warm and professional support specialist.
Your mission is to give every customer fast, accurate, friendly service.
## Personality Traits
- Patient and kind, even under pressure
- Detail-oriented — confirm the question before answering
- Proactive — anticipate what the customer needs next
## Domain Expertise
- Product specifications and usage
- Order lookup and shipping tracking
- Return and exchange policies
- Common troubleshooting
## Communication Style
- Warm, polite English
- Reply structure: greet → confirm the question → answer → ask if more is needed
- When out of scope, politely route to the right team
## Operating Principles
1. **Confirm first**: verify you understand before answering
2. **Don't over-promise**: look things up; don't guess
3. **Log what matters**: feedback and special requests go in the record
4. **Escalate fast**: if unable to help, hand off to a human within 10 minutes`}
Step-by-Step Setup
Step 1: Build the Knowledge Base
Give Ada FAQs to lean on:
{`# FAQ Library
## Orders
Q: When will my order arrive?
A: Share your order number and I will check the shipping status. Orders typically arrive 1–3 business days after dispatch.
Q: How do I check my order status?
A: You can:
1. Sign in to your account and view order details
2. Share your order number with me and I'll look it up
3. Click the tracking link in your shipping confirmation email
Q: Can I cancel my order?
A:
- Not yet shipped: yes, cancellable directly; refund in 3–5 business days
- Already shipped: not cancellable, but you can refuse delivery or request a return
## Returns
Q: Can I return an item?
A: Yes — we offer a 7-day return window. Please make sure:
- The item is in new condition
- Packaging is intact (including accessories and gifts)
- The return form is filled out
Q: Who pays for return shipping?
A:
- Defective product: we cover shipping both ways
- Change of mind: the customer covers the return shipping
## Products
Q: Is [product name] in stock?
A: Let me check the inventory system… [connects to inventory]
Q: Does the product come with a warranty?
A:
- Electronics: one-year warranty
- Accessories: three-month warranty
- Full terms on the product page`}
Step 2: Set Up the Telegram Bot
- In Telegram, search for @BotFather
- Create a new bot and get a Bot Token
- In the Admin Panel → Messaging → Telegram, paste the token
- Send a test message to confirm it works
{`/newbot
# Enter the bot's display name
MyShop Customer Service
# Enter a username (must end with bot)
myshop_cs_bot
# Receive the token
Use this token to access the HTTP API:
123456789:ABCdefGHIjklMNOpqrSTUvwxyz`}
Step 3: Configure the Auto-Reply Flow
{`{
"trigger": "Telegram DM received",
"steps": [
{
"action": "Classify intent",
"agent": "Ada",
"description": "Identify what the customer is asking"
},
{
"action": "Search knowledge base",
"agent": "Ada",
"description": "Find the matching answer"
},
{
"action": "Reply to customer",
"agent": "Ada",
"description": "Deliver the answer"
},
{
"action": "Log the conversation",
"agent": "Ada",
"description": "Store the exchange in the transcript"
}
],
"escalation": {
"condition": "Cannot answer OR customer requests a human",
"action": "Notify admin and hand over the conversation"
}
}`}
Step 4: Test and Launch
- Internal test: use a second account to exercise a range of questions
- Soft launch: limited release to a small group
- Collect feedback: note what Ada could not handle
- Keep iterating: update the knowledge base, refine the persona
Full Configuration Example
{`# ============ SOUL.md ============
You are **Ada**, a warm and professional support specialist.
Your mission is to give every customer fast, accurate, friendly service.
## Personality
- Patient and kind
- Detail-oriented
- Proactive
## Domain Expertise
- Product specifications and usage
- Order lookup and shipping tracking
- Return and exchange policies
## Communication Style
- Warm, polite English
- Reply structure: greet → confirm the question → answer → ask if more is needed
- When out of scope, politely route to the right team
## Operating Principles
1. Confirm first: verify you understand before answering
2. Don't over-promise: look things up; don't guess
3. Escalate fast: hand off to a human within 10 minutes when needed
# ============ USER.md ============
- **Name:** [Company name]
- **Industry:** [E-commerce / services / SaaS]
- **Business model:** [short description]
## Support Policy
- Service hours: 24/7 auto-reply
- Human support: weekdays 9:00–18:00
- Emergency contact: [admin Telegram ID]`}
Escalation Configuration
{`# Settings for when a human needs to step in
escalation:
triggers:
- keyword: ["human", "agent", "representative", "complaint"]
- sentiment: "negative" # Negative sentiment detection
- confidence: "low" # Ada is unsure
notification:
method: "telegram"
target: "@admin_username"
message: |
🚨 Customer support needs a human
Customer: {customer_name}
Last message: {last_message}
Conversation: {conversation_link}
Please follow up as soon as possible.
auto_reply:
message: |
Apologies — I'd like a specialist to handle this one for you.
I've handed it over; someone will reach out in 5–10 minutes.
Thanks for your patience!`}
Advanced
Integrating With an Order System
{`// Order lookup API integration example
async function queryOrder(orderId) {
const response = await fetch('https://api.myshop.com/orders/' + orderId, {
headers: {
'Authorization': 'Bearer ' + process.env.SHOP_API_KEY
}
});
const order = await response.json();
return {
status: order.status,
items: order.items,
total: order.total,
shipping: order.shipping_tracking,
estimated_delivery: order.estimated_delivery
};
}
// Ada can call this function to look up an order`}
Customer Satisfaction Tracking
{`At the end of a conversation, Ada sends:
---
Was this answer helpful?
👍 Helpful / 👎 Not helpful
Your feedback helps us improve.
---
Clicks feed into:
- Satisfaction reports
- Ada's learning loop
- Admin dashboard`}
FAQ
Will Customers Notice It’s an AI?
You can choose:
- Transparent mode: introduce Ada as an AI in the first message
- Hybrid mode: stay neutral, reveal when stuck
- Recommendation: for e-commerce, transparency builds trust
How Do I Keep Ada From Making Things Up?
- Set clear knowledge-base boundaries
- Auto-escalate when confidence is low
- Periodically review transcripts and correct errors
How Many Simultaneous Conversations Can Ada Handle?
In theory unlimited, but in practice:
- Monitor reply quality
- Queue new conversations when load spikes
- Prioritize urgent issues
Summary
Automated customer support is one of Ada’s best use cases:
- Configure the knowledge base and persona
- Connect Telegram or another messaging platform
- Build an escalation path for exceptions
- Iterate continuously to raise reply quality