AI Customer Support — Usage Guide
Everything you need to know to set up live chat, configure AI auto-replies, manage your FAQ, and handle conversations from your dashboard.
Overview
Live Chat Widget
Floating chat bubble on every storefront page. Customers chat without creating an account.
AI Auto-Replies
Order tracking (WISMO) is rule-based. FAQ answers use RAG from your knowledge base.
Dashboard Inbox
All conversations in one real-time inbox. Reply, resolve, and track unread counts.
AI Customer Support is one feature with three interconnected parts: the storefront widget customers see, the dashboard inbox you use to reply, and the AI engine that answers automatically. You can use any combination — just the widget (manual support only), just the widget + inbox (reactive support), or the full setup with AI enabled (automated + human hybrid).
Enable the Chat Widget
Navigate to your store
From the dashboard, go to your store's overview page.
Open Customer Support
In the Quick Actions grid, find the Customer Support section. Click Widget Settings.
Toggle the widget on
Find the "Enable chat widget" toggle and turn it on. The chat bubble will appear on your storefront immediately.
The widget is disabled by default on every new store. Enabling it takes effect within 60 seconds — the storefront caches widget config for 1 minute.
Configure Widget Settings
Build Your FAQ Knowledge Base
The FAQ Knowledge Base is where you write the content the AI uses to answer questions. Think of it as training the AI — every Q&A pair you add becomes searchable by the AI when a customer asks a related question.
Go to FAQ Knowledge Base
From your store overview, click FAQ Knowledge Base in the Customer Support section.
Create a category
Click "New Category". Name it something descriptive: "Shipping & Delivery", "Returns & Refunds", "Products", "Payments". Categories keep your FAQ organised.
Add questions with "Add question"
Click "Add question" inside any category. Type the question exactly as a customer might ask it. Write a complete, clear answer. Click Save.
Wait ~5 seconds for AI indexing
You'll see a toast: "FAQ item added — AI will learn from it shortly." The embedding is generated in the background. After ~5 seconds, the AI can use this answer.
Write questions the way customers ask them. "What is your return policy?" works better than "Return policy". The AI does semantic matching — phrasing matters for relevance.
You can edit answers at any time. When you save an update, the old embedding is automatically replaced. The AI uses your latest version within seconds.
Empty categories are fine. They won't appear in the storefront FAQ display until they have at least one item.
Suggested FAQ categories to start with:
Shipping & Delivery
How long does shipping take? Do you ship internationally? What carriers do you use?
Returns & Refunds
What is your return policy? How do I start a return? When will I get my refund?
Orders
Can I change my order? Can I cancel? What if my order arrives damaged?
Products
How do I find my size? Are products in stock? What materials are used?
Payments
What payment methods do you accept? Is my payment secure?
Promotions
Do you have a loyalty programme? How do discount codes work?
Enable AI Auto-Replies
Open Widget Settings
Store overview → Widget Settings.
Toggle "Enable AI" on
From this moment, every customer message will be checked for WISMO patterns first, then FAQ matches. Both happen automatically.
The AI only sends a reply if it finds a relevant answer. If a customer asks something completely outside your FAQ and it's not an order tracking question, the AI stays silent — the conversation stays open for you to reply manually.
Requires OpenAI API key on the backend. If OPENAI_API_KEY is not configured in your environment, WISMO will still work (it doesn't use LLM) but FAQ RAG answers will not be generated.
How WISMO Order Tracking Works
WISMO ("Where Is My Order?") is the most common customer support question. The system handles it using fast pattern matching — no AI model, no latency, no cost per query.
Detected patterns (any phrasing is matched)
How it resolves:
e.g. "Check order #1234"
e.g. "[email protected] — where's my order?"
e.g. Customer filled email in the pre-chat form
e.g. "Where is my order?"
The reply includes: order number, current status, payment status, items ordered, total, and all tracking links as clickable URLs. If the order has no tracking yet, this is stated clearly.
Enable the pre-chat form (name + email) so WISMO can look up orders by email even when the customer doesn't include an order number in their message.
How RAG FAQ Answers Work
RAG (Retrieval-Augmented Generation) is how the AI answers FAQ questions. When a customer sends a message that is not a WISMO question, the system:
The AI never makes things up. It only generates an answer when your FAQ has relevant content. This is the critical difference between RAG and a basic chatbot — RAG is grounded in your actual data.
The more FAQ content you add, the better the AI answers. A sparse FAQ (3 items) will only auto-answer 3 types of questions. A rich FAQ (50+ items across multiple categories) will handle most common questions automatically.
Using the Dashboard Inbox
Finding the inbox
Store overview → Chat Inbox (Customer Support section). You'll see a red badge when you have unread conversations.
Status tabs
Message colours in the conversation thread
Replying
Type in the reply box at the bottom of the conversation. Press Enter to send. The message is delivered to the customer's widget in real time via WebSocket.
Resolving conversations
Click Resolve when you're done. The conversation moves to the Resolved tab. You can Reopen it at any time. Messages are never deleted.
The Storefront Widget
What your visitors see
A floating chat bubble in the corner you chose (bottom-right or bottom-left). When clicked, a chat window opens showing your greeting message. If the pre-chat form is enabled, they're asked for name and email first.
Session persistence
Once a visitor has chatted, their session is saved in the browser's localStorage. When they return to the store, they see their previous conversation — they don't have to start over.
Real-time replies
When you reply from the dashboard, the message appears in the customer's widget instantly via WebSocket. When the AI sends a reply, it appears the same way.
The widget has no dependency on third-party scripts. It's part of your storefront build — it doesn't add external domains, doesn't slow page load, and doesn't affect your Core Web Vitals.
End-to-End Message Flow
Here's exactly what happens when a customer sends a message:
Opens the chat widget on your store
If pre-chat form is enabled: shows name/email form. Customer submits → conversation created.
Types a message and hits Enter (or the send button)
POSTs the message to the backend via REST. Optimistic message shown immediately.
Saves the message to the database
Broadcasts the message to the conversation room (you see it in your dashboard instantly)
Notifies your store room (unread badge updates)
Runs asynchronously in the background (does not delay the response to the customer)
Is it an order tracking question? YES → WismoService looks up the order → formats reply → skips RAG
NOT a WISMO question + AI enabled? → searches FAQ embeddings → if matching content found → generates answer
Saves the AI reply to the database → broadcasts to the conversation room
Sees the AI reply in the widget in real time (via WebSocket)
See the full conversation in your inbox. If the AI handled it, you can resolve it. If not, reply manually.
Tips & Best Practices
Enable the pre-chat form to supercharge WISMO
When a customer enters their email in the pre-chat form, WISMO can look up their orders without needing them to type their order number. This dramatically improves the auto-reply rate for order questions.
Start FAQ with your top 10 support questions
Look at your last 50 support emails. What do people ask most? Those are your first FAQ items. Even 10 well-written Q&A pairs will deflect a significant share of repetitive questions.
Write answers in full sentences
The AI uses your FAQ answers verbatim (in part) to generate replies. "3–5 business days" is less useful than "We use standard shipping which typically takes 3–5 business days within the US."
Use the widget's offline message honestly
If you're a solo operator who can only reply in the evenings, say so: "Hi! I'll reply within a few hours. In the meantime, the AI can help with common questions." Sets clear expectations.
Monitor AI replies for the first week
When you first enable AI, check the inbox daily. Are the AI replies accurate? If one is wrong, update the FAQ answer — it'll be re-embedded within seconds.
AI replies appear in the same thread as your replies
Customers see AI replies labeled as AI. You see them labeled in the dashboard too. There's full transparency — no pretending the AI is a human.
Resolve AI-handled conversations to keep your inbox clean
When the AI has fully answered a question and the customer didn't follow up, mark the conversation Resolved. This keeps your Open tab showing only conversations that genuinely need attention.
Canned Responses
Canned responses are saved reply templates. Type a short keyword in the reply box, select your template, and the full answer fills in instantly — so you stop retyping the same phrases dozens of times a day.
Open Canned Responses
Store overview → Customer Support → Canned Responses.
Click "New Canned Response"
Give it a short trigger keyword (e.g. "refund") and write the full response text.
Use it in the inbox
In any conversation reply box, type "/" followed by your keyword. A picker appears — select the response and it fills the box.
Create canned responses for your 10 most common replies: tracking update, refund started, apology for delay, out of stock, discount offer, etc. Most merchants save 30+ minutes a day once they have 10–15 templates set up.
CSAT & Sentiment Analysis
CSAT surveys
When you resolve a conversation, a satisfaction survey is automatically sent to the customer via the widget. They can rate the interaction thumbs up or thumbs down with an optional free-text comment. Scores aggregate into your real-time CSAT score visible in the Analytics dashboard.
CSAT survey flow
Sentiment analysis
Every incoming customer message is automatically analysed for tone. Messages are classified as Positive, Neutral, or Negative. The sentiment badge is shown next to the conversation in your inbox, so you can instantly spot frustrated customers and prioritise urgent conversations.
Sentiment analysis runs asynchronously in the background and does not add latency to message delivery. Results typically appear within 1–2 seconds of the message arriving.
Proactive Chat Triggers
Proactive cues automatically open the chat widget (or send a first message) when a visitor matches conditions you define — without them clicking the chat bubble. Use this to engage visitors who are browsing a product page for a long time, about to abandon checkout, or reaching a scroll threshold on your homepage.
Available trigger conditions
Open Proactive Cues
Store overview → Customer Support → Proactive Cues.
Click "New Cue"
Choose a trigger condition, set the threshold, and write the proactive message (e.g. "Need help finding the right size? I'm here!").
Activate
Toggle the cue on. It starts firing on your storefront immediately.
The most effective proactive cue for e-commerce: trigger on URL containing /checkout + time on page 45 seconds with message "Having any trouble checking out? I can help!" — a well-timed recovery message here can meaningfully reduce cart abandonment.
Business Hours & Auto-Close
Business hours
Configure your support hours per day of week. When a visitor chats outside your hours, they see your configured offline message and your response time expectation. You can choose to disable AI replies outside business hours (so the AI doesn't set a misleading expectation that a human is monitoring) or keep them on 24/7.
Open Business Hours settings
Store overview → Customer Support → Business Hours.
Set hours per day
Toggle each day of the week and set the start/end time. Set timezone to match your location.
Configure outside-hours behaviour
Choose whether to show the offline message, disable AI, or collect email for a follow-up.
Auto-close idle conversations
Conversations that go quiet for a configurable period are automatically moved to Resolved, keeping your inbox from filling with abandoned chats. The default idle timeout is 24 hours for open conversations. You can change this in Customer Support settings.
Set auto-close to 48 hours if your customers often return to a conversation with a follow-up question the next day. Set it shorter (6–8 hours) if you want a clean inbox and most of your conversations resolve in a single session.
Support Analytics
The Support Analytics dashboard shows you how your team and AI are performing. Access it from Store overview → Customer Support → Analytics.
Metrics tracked
Check the "Busiest hours" heatmap after your first week. If you see a clear peak at 7–9 PM, that's when you need to be available — or when your AI should be doing the heaviest lifting.
Smart Replies & AI Summaries
Smart reply suggestions
When you open a conversation, the AI reads the last 10 messages and generates 3 contextually relevant reply suggestions above the reply box. Click any suggestion to fill the reply box — edit as needed, then send. Smart replies dramatically speed up your first response, especially when handling a high volume of similar questions.
AI conversation summaries
After every conversation is resolved, the AI automatically generates a concise summary and saves it as an internal note. The summary includes: what the customer asked, what was resolved, and any follow-up needed. This is particularly valuable for team handoffs — a new agent opening a re-opened conversation immediately understands the full context.
Email transcript
When transcript sending is enabled in your settings, a full email transcript of the conversation is sent to the customer's email address when you resolve the conversation. Customers who want a paper trail of the support interaction particularly appreciate this feature.
AI summaries and email transcripts both require a valid customer email address — either captured via the pre-chat form or linked from their Medusa customer account.
File Attachments
Both customers (via the storefront widget) and agents (via the dashboard inbox) can attach files directly in a conversation — images to show a damaged product, PDFs for invoices or instructions, screenshots of error messages.
Attachment details
Encourage customers to attach a photo when they report a damaged item or the wrong product. This eliminates the back-and-forth of "can you describe the issue?" and lets you process the claim in one step.
File attachments are uploaded directly to your store's R2 bucket — they do not pass through the chat API server. The upload endpoint returns a secure URL that is embedded in the chat message.
Search, Read Receipts & Typing Indicators
Inbox search
Use the search bar at the top of the Chat Inbox to instantly find any conversation. Search works across visitor name, email address, and the full text of every message in the conversation. Results update as you type.
Read receipts
Messages show a read receipt timestamp when the other party has opened the conversation and seen the message. When you read a conversation in the dashboard, the customer's widget updates to show the read time. When the customer opens the widget after you've sent a reply, you see the read time in your dashboard — so you know they've seen your answer.
Typing indicators
When the customer is typing in the widget, a typing indicator (animated dots) appears in your dashboard conversation view. Equally, when you are typing a reply, the typing indicator appears in the customer's widget. Both sides know the other is actively engaged — reducing unnecessary "are you still there?" follow-ups.
Typing indicators and read receipts are delivered via WebSocket in real time. They add no perceptible latency and work on both desktop and mobile browsers.
Ready to enable AI Customer Support?
It takes less than 5 minutes to go from zero to a live chat widget with AI-powered auto-replies.