The most useful chatbot examples for a consumer brand are its everyday service conversations: "Where's my order?", "Can I change the address?", "It arrived cracked", "Where's my refund?". Below are seven of them, written out on WhatsApp, Instagram, website chat and email, each shown three ways: what a scripted bot, an AI chatbot and an AI agent would do with the same message.
They're written for support teams at consumer brands in India and the US, from the daily work of ecommerce customer service, and the AI agent's wording is ready to reuse in your own bot. Four brand examples (Klarna, IKEA, Amazon and Flipkart), from the brands' own pages, follow.
In short
- A scripted bot follows menus and keywords, an AI chatbot answers from your policies, and an AI agent checks the customer's order and acts within your rules.
- On the same message, the first two can explain your policy. Only an AI agent can say where this customer's parcel is, or change her order.
- The channel changes the conversation: on WhatsApp the brand can write first only with an approved template, on Instagram it can't write first at all, and website chat ends when the tab closes.
- Every example needs a way out: the rule that hands the conversation to a human agent, and what that agent receives.
Three kinds of chatbot, one message
| Scripted bot | AI chatbot | AI agent | |
|---|---|---|---|
| How it works | Menus, buttons and keywords, drawn as a flowchart | Understands free text and answers from your policies and FAQs | Also looks up the customer's order and takes the actions your rules allow |
| What it needs | A path for every question | Up-to-date policies and product content | Your policies, access to order, delivery and payment systems, and a rule for each action |
| "Where's my order?" | Asks for the order number and sends the courier's tracking link | Explains usual delivery times and links the tracking page | Finds the order and says where the parcel is and when it will arrive |
| When it hands over | When the customer steps off the menu | When the answer isn't in your content | When a rule says a human agent must decide |
A scripted bot handles only what's on its chart; anything else dead-ends, loops or lands with a human agent. That's why the scripted versions below keep sending customers to email. An AI chatbot escapes the chart but not the policy page: it can explain a rule, not apply it to this customer's order. The AI agent finishes the request, the idea behind agentic AI for customer service. To compare products rather than conversations, see our list of AI chatbots for customer service; for the technology, see chatbot vs conversational AI.
Flowcall's AI Agent is the third kind. It runs on the SOP Engine: it follows your written procedure step by step while talking naturally, your rules make each decision, and order, delivery and payment details come from your systems, such as Shopify and logistics platforms and couriers like ClickPost, Shiprocket and Delhivery. Once a workflow is set up correctly, it runs the same way every time. Ready-made workflows cover returns, refunds, exchanges, cancellations, COD and damaged products, and order and delivery questions are answered from the live shipment. The same workflows run on WhatsApp, Instagram DMs, Messenger, Live Chat and email.
The seven examples at a glance
| Situation | Channel | Scripted bot | AI chatbot | AI agent |
|---|---|---|---|---|
| 1. Order status | Tracking link, after an order-number prompt | Usual delivery times and the tracking page | Where the parcel is and the new delivery date | |
| 2. Address change | Website chat | "Please email support" | Explains the before-dispatch rule | Checks it hasn't shipped, collects the new address, updates the order |
| 3. COD confirmation | WhatsApp (the brand writes first) | Confirm and Cancel buttons only | Understands a typed reply but can't change the order | Confirms, handles "send it to my office", offers to pay online |
| 4. Return or exchange | Instagram DM | Link to the returns portal | Explains the policy | Finds the order, checks the window and stock, books the exchange |
| 5. Damaged item | WhatsApp, with a photo | "Email us a photo" | Asks for photos, promises a reply in two days | Checks the photo and the claim window, sends a replacement |
| 6. Refund status | Auto-acknowledgement | Quotes the refund policy | Explains this refund, line by line | |
| 7. Product question | Instagram DM | Link to the product page | Answers from one product page | Searches the catalogue and suggests products in stock |
Seven chatbot examples from consumer-brand support
The first three happen between order and delivery, the next three after delivery, and the last before the customer buys. Each gives the customer's message, what each kind of chatbot does with it, the AI agent's side written out to reuse, and when to hand over to a human agent. The brands, customers and order details are made up; the situations aren't.
1. Order status on WhatsApp: "Where's my order?"
"Where is my order?", or WISMO for short, is the easiest message to answer and the easiest to answer badly, because the reply depends on one customer's parcel. At 9:10 pm, a customer writes to an electronics brand:
"Hi, my earbuds were supposed to come yesterday. Where are they?"
Scripted bot. It shows a menu ("1 Track order, 2 Returns, 3 Something else"), asks for the order ID and sends the courier's tracking link. If that page says "delayed", she's back where she started.
AI chatbot. It apologises, explains that orders usually arrive in three to five working days, and links the tracking page. It can't see the parcel.
AI agent. WhatsApp shows it her phone number, so it looks for her order; if the number matches no customer, or more than one, it asks for the order number instead of guessing. Then it checks the courier:
AI agent: Hi Aditi, sorry for the wait. Your earbuds (order #48213) reached the Pune hub this morning, a day late. The courier shows them out for delivery tomorrow, before 7 pm.
Customer: ok thanks. can you tell me when it's out?
AI agent: Yes, I'll message you here when it's out for delivery tomorrow.
Hand over when the parcel has been stuck longer than your limit (say three days), the courier marks it delivered but she says it hasn't arrived, or it's her third message about the same order. The AI agent opens a ticket for the team that deals with the courier, with the tracking history attached, and tells her when she'll hear back.
What it needs: the order and live courier tracking, a delay limit, and a rule for "delivered but not received".
2. Address change in website chat: "Can I send it to my office?"
A shopper opens the chat on a footwear brand's order-status page:
"I ordered an hour ago but I'll be at work on Thursday. Can I change the delivery address?"
Scripted bot. The menu has no "change address" option, so she gets "Please email support with your order number." By the time a human agent replies, the order may have shipped.
AI chatbot. It explains that addresses can be changed until dispatch and asks her to contact support. It's right about the policy and can't apply it.
AI agent. A website chat starts without knowing who the visitor is, so it asks for the order number and checkout email, and changes the address only if they match:
AI agent: Yes, as long as it hasn't been packed. What's your order number and the email you used at checkout?
Customer: 10442, jess.m@example.com
AI agent: Thanks, Jess. It hasn't been packed yet, so we can change it. Please add the new address here. [address form]
Customer: [submits 221 Market St, 5th floor, San Francisco, CA 94105]
AI agent: Done. Order #10442 will now go to 221 Market St, 5th floor, San Francisco, CA 94105, still arriving Thursday. You'll get the updated confirmation by email.
The form catches a missing unit number or postcode, and asking for the email first means the conversation can continue by email if she closes the tab.
Hand over when the order has shipped and your courier offers no redirect, the email doesn't match the order, or this order's address has already been changed once. For human agents on website chat, see our live chat examples.
What it needs: the order's fulfilment status, permission to edit the shipping address, a delivery check for the new pincode or ZIP code, and a rule for proving the order is hers.
3. COD confirmation on WhatsApp: the brand writes first
Cash on delivery (COD) means the customer pays the courier when the parcel arrives. Brands that offer it, in India especially, often confirm each COD order before dispatch, because a refused parcel travels both ways at the brand's cost (return to origin, or RTO).
Because the brand writes first, the first message has to be one of its approved WhatsApp message templates. Meta treats a message that confirms an order with its details as a utility template, as long as it contains no promotion; add a discount for paying online and Meta classes it as marketing. That's a trade-off. A discount inside the confirmation reaches the customer while they decide, as a marketing template: the marketing rate, and no delivery to US numbers. Made in the conversation after the customer replies, as below, it leaves the confirmation utility.
Brand (template): Hi Rahul, thanks for your order #55120: 1 × cotton kurta, ₹1,299, cash on delivery to Andheri East, Mumbai 400069. Please confirm so we can ship it. [Confirm order] [Cancel order]
Scripted bot. The buttons work. Anything typed ("confirm but send it to my office") gets "Please choose Confirm or Cancel", and the order waits, unconfirmed.
AI chatbot. It understands the typed reply but can't change the order, so it asks him to contact support.
AI agent. It treats a typed reply like any other request:
Customer: Confirm, but can you send it to my office? Nobody's home in the day
AI agent: Confirmed. Please add your office address here. [address form]
Customer: [submits an office address in Lower Parel, Mumbai 400013]
AI agent: Updated: order #55120 will go to your office in Lower Parel, cash on delivery, ₹1,299. If you'd rather not keep cash ready, you can pay ₹1,249 now by UPI instead. [Pay ₹1,249]
On WhatsApp, forms like this one are built with WhatsApp Flows. If the customer doesn't reply at all, send one reminder template after a few hours, then follow your rule: cancel, or leave the order for a human agent to call.
Hand over when he disputes the amount, wants to change the items, or his number has refused COD parcels before.
What it needs: new-order events from your store, an approved template (utility, or marketing if it carries the offer), permission to edit the order, a payment link, and reminder and cancellation rules. See our guide to COD orders for the policy and our WhatsApp order template examples for wording.
COD confirmation in Flowcall's WhatsApp Marketing confirms cash-on-delivery orders from Shopify on WhatsApp and offers a discount to pay online in the same message, and customers confirm or cancel in the chat. A reply to the template can start an AI Agent workflow, so a customer who types "send it to my office" gets the address form, not a button reminder. If the template gets no reply, Flowcall sends another template, runs an action, or both.
4. Return or exchange on Instagram: "It doesn't fit"
A customer writes to a fashion brand's Instagram account:
"Hi! The linen dress is too big 😩 can I exchange it for an S?"
Scripted bot. An Instagram auto reply sends the returns portal link, where she starts again with her order number and email.
AI chatbot. It explains the policy (exchanges within 15 days, unworn, with tags) and links the portal. It can't see her order or the stock.
AI agent. Instagram gives it a handle, not a phone number or email, so it asks for one first:
AI agent: Of course. What's the phone number or email you used at checkout?
Customer: 98200 xxxxx
AI agent: Found it: the sage linen dress in M, delivered on Monday. You're within the 15-day exchange window and the S is in stock. Shall I book the exchange? A courier collects the M, and the S ships as soon as the pickup is scanned.
Customer: yes pls
AI agent: Booked. Pickup is on Thursday between 10 am and 6 pm from your delivery address; please keep the tags on. Your exchange number is EX-5521.
Hand over when she's outside the window and asks for an exception, or disputes a final-sale rule. If the S is out of stock, the AI agent first offers what your policy allows, such as a refund, store credit or another colour.
What it needs: order lookup by phone or email, the return window and exclusions, stock by size, and pickup booking with your returns partner.
5. Damaged item on WhatsApp, with a photo
A customer writes to a homeware brand the evening her order arrives:
"The glass jars came today and one is cracked" [photo]
Scripted bot. "Sorry for the inconvenience! Please email a photo and your order number to support." She has to send the photo again, somewhere else.
AI chatbot. It apologises, explains the 48-hour reporting rule, and says the team will review it within two working days. The photo is in the chat, but nothing looks at it.
AI agent. It checks the photo, matches it to the order and applies your damaged-item rule:
AI agent: I'm so sorry, that's not how they should arrive. I can see the crack on one of the 1-litre jars from order #30918, delivered this afternoon. Would you like a replacement jar or a refund for it?
Customer: replacement
AI agent: Done. A new 1-litre jar ships tomorrow at no cost, order #30918-R. You don't need to send the cracked one back; please wrap it before you throw it away.
Hand over when the photo shows no damage or a different product, it's her second claim within your limit, the item is worth more than the AI may replace on its own, or she mentions an injury. An injury goes straight to a trained human agent.
What it needs: AI that can read the photo, the delivery date, your claim window and value limit, the customer's claim history, and permission to create a replacement order.
Damaged products are one of Flowcall's ready-made returns, refunds and exchanges workflows. The AI Agent reads the photo the customer sends and assesses the damage, then your rules decide: the claim window, the value limit, and whether a second claim needs approval. When approval is needed, the human agent opens a ticket in Flowcall's AI Helpdesk with the photo, the order and the checks already done, and approves with a ticket action: the AI fills the replacement form from the conversation and the order, and the human agent checks it and clicks once.
6. Refund status by email: "Why is my refund less than I paid?"
A customer emails a clothing brand:
Subject: Refund
"I returned one of the two shirts on the 3rd and got $36 back, but the shirt was $42. Where's the rest?"
Scripted bot. Email has no menu, so the scripted version is an auto-acknowledgement: "We've received your email and will reply within 48 hours."
AI chatbot. It drafts a reply from the policy: refunds go to the original payment method within five to seven business days, and return shipping may be deducted. It's all true, and she still has to work out whether that explains the $6.
AI agent. It reads the order, the return and the refund, and answers her actual question:
Hi Maria,
Your refund for the white Oxford shirt (order #7781) was issued on 5 March: $42.00 for the shirt, less the $6.00 return shipping fee that applies to returns that aren't faulty, so $36.00 went back to your Visa ending 4417. Banks usually take 5–10 business days to show it.
If the shirt was faulty, reply with a photo and we'll refund the $6 as well.
Hand over when the refund failed at the payment gateway, it hasn't arrived after your bank timeline, or she says she was never told about the fee, which is a goodwill decision.
What it needs: order, return and refund records from your store and payment gateway, your deduction rules, and a decision on whether AI email replies go out directly or wait for review. For wording, see our return email templates.
7. Product question on Instagram: "Will this suit oily skin?"
Questions before buying arrive in the same inbox. After watching a reel, a shopper writes to a skincare brand:
"Is the vitamin C serum ok for oily skin? And do you have anything without fragrance?"
Scripted bot. It matches neither of its keywords ("price", "link"), so it replies "Thanks for reaching out! We'll reply soon."
AI chatbot. It answers from the serum's product page ("suits all skin types") but can't search the catalogue for fragrance-free options.
AI agent. It searches the catalogue by skin type and fragrance, and shows what's in stock:
AI agent: The 10% vitamin C serum is a light gel made for oily and combination skin, but it does contain a mild fragrance. Two fragrance-free options for oily skin:
- Niacinamide 5% serum, ₹549, in stock [product link]
- Vitamin C + E gel, ₹699, in stock [product link]
Would you like me to compare them?
Hand over when the question is about a medical condition, pregnancy or an allergic reaction. Those go to a trained human agent, or get the answer your policy sets, such as "please check with your doctor first".
What it needs: a catalogue with the attributes customers ask about (skin type, fragrance, size, fit), live stock, and rules for health questions. If this is the main job, see our guide to the best ecommerce chatbot.
Flowcall's AI shopping assistant answers these from your catalogue. Customers describe what they want, and it finds, filters and compares products by size, colour, price and other attributes, and sends each one with its price and a buy link, as in the reply above; on WhatsApp and Live Chat, it shows them as product cards with images. It can also find products from a photo or screenshot and, in India, send a payment link or UPI in the chat. It's the same AI Agent that handles orders and returns, on the same WhatsApp number and Instagram account, so a shopper who asks about a serum today and about her delivery next week talks to one assistant.
Four brand chatbot examples, as the brands describe them
Four consumer examples, described only as far as each brand's own pages go (checked on 30 September 2026):
Klarna's AI assistant (an AI agent for service). In February 2024, Klarna, the buy-now-pay-later company, said its AI assistant had handled 2.3 million conversations in its first month, two-thirds of its customer service chats. It handles refunds, returns, payment issues, cancellations, disputes and invoice inaccuracies, in 23 markets and more than 35 languages, and customers can still choose a live agent (Klarna, 27 February 2024).
IKEA's Billie (an AI chatbot). IKEA retailer Ingka Group says its AI chatbot Billie resolved about 47% of the customer enquiries it received from 2021 to 2023, some 3.2 million interactions, and that 8,500 call-center co-workers were reskilled for work such as remote interior design advice (Ingka Group, 29 June 2023).
Amazon's Rufus, now Alexa for Shopping (a shopping assistant). Amazon describes Rufus as a generative AI shopping assistant trained on its product catalogue, customer reviews, community Q&As and information from across the web, answering questions such as "is this jacket machine washable?". Amazon renamed it Alexa for Shopping on 13 May 2026 (About Amazon).
Flipkart's Flippi (a shopping assistant). Flipkart describes Flippi, the generative AI chat assistant in its app, as guiding shoppers to personalised product recommendations based on their tastes, budgets and preferences, in English for now (Flipkart Stories).
Each works from the brand's own data (Klarna's payments and refunds, the retailers' catalogues), and Klarna and IKEA kept human agents in the picture.
How the channel changes the conversation
The WhatsApp and Instagram rules below come from Meta's developer documentation, checked on 30 September 2026.
| Instagram DMs | Website chat | |||
|---|---|---|---|---|
| Who can write first | The customer; the brand only with an approved template | The customer only | Either, while the visitor is on the site | Either |
| How long the brand can reply freely | 24 hours from the customer's last message; after that, templates only | 24 hours; a human agent can reply later using Meta's human agent tag | While the visitor is on the page | No limit |
| What the chat can show | Buttons, lists, forms, photos, documents, payment links | Photos, videos, quick replies | Buttons, forms, product cards | Long text and attachments |
| What the bot knows about the customer | Usually the phone number (see below) | An Instagram handle | Nothing, unless they're logged in or fill in a pre-chat form | The email address |
| Examples above | 1, 3, 5 | 4, 7 | 2 | 6 |
Three things follow:
- The WhatsApp phone number isn't guaranteed any more. WhatsApp is rolling out usernames during 2026, and a customer who adopts one can hide her number. Meta still passes it on if the brand has exchanged messages or calls with it in the last 30 days or has it in its contact book, so recent order updates keep lookups working; keep "what's your order number?" as the fallback. These rules come with the WhatsApp Business API. For more, see our WhatsApp chatbot guide, and for template costs, our WhatsApp API pricing guide.
- An Instagram bot starts every order request with a question. An Instagram chatbot sees only a handle, so its first reply to "where's my order?" asks for a phone number, email or order number. Write that question well.
- Website chat needs a way to continue. Collect an email or phone number early, as in example 2, or the conversation ends with the tab. (Email is the opposite: no time limit, but one email can hold several questions, and each needs an answer.)
How to test a chatbot with these examples
Use the seven situations as a test script: take real messages from last month for each, add one that should break a rule (a return outside the window, a second damage claim), and run them through your bot or a trial of the one you're evaluating. Score each conversation:
| Check | Pass | Fail |
|---|---|---|
| Did it use this customer's data? | Names the order, the parcel's position, the refund amount | Quotes the policy page |
| Did it finish the request? | The address, exchange or replacement is done and confirmed | "Please contact support" |
| Did it follow the rule on the exception? | Declines, offers what your policy allows, or hands over | Approves it anyway, or loops |
| Did the handover carry the details? | The human agent gets the order, photos, checks and reason | The human agent asks the customer to start again |
| Did it respect the channel? | A template for the first COD message; asks Instagram customers who they are | Free text outside WhatsApp's window; can't find an Instagram customer's order |
What the human agent should receive at handover is covered in chatbot vs live chat, and who owns each case after it in escalation management.
In Flowcall, each workflow is tested against simulated customer conversations before it goes live and after every change, with AI-written tests for the happy path, edge cases and errors, and real failures become new tests. After launch, AI Auto QA checks what the AI Agent said and did (wrong, incomplete or made-up answers, loops, missed or late handoffs), and automation gap detection groups the requests it couldn't resolve into the workflows worth adding next.
How Flowcall handles these conversations
Flowcall puts an AI Agent and an AI Helpdesk on the same WhatsApp, Instagram, Messenger, Live Chat and email conversations. The AI Agent answers product questions and completes order, delivery, return and refund requests within your rules, and when a rule says a human agent should decide, the AI Helpdesk gives them a ticket with the conversation, the order and what's left to do. Flowcall's team sets up the workflows with you and keeps them current as your policies change.
If you'd like to see Flowcall handle a COD confirmation, a damaged-item claim or an Instagram exchange like the ones above, book a demo. Qualifying brands can also try Flowcall on one of their own service problems in a free proof of concept.
Frequently Asked Questions
What is a chatbot, with an example?
A chatbot is software that replies to messages automatically, in a website chat window, a messaging app such as WhatsApp, or email. A simple customer service example asks for your order number and sends a tracking link; a more capable one finds your order from your phone number and tells you when the parcel will arrive. For the basics, see what is a chatbot.
What are the top AI chatbots?
The best-known AI chatbots are general assistants such as ChatGPT, Gemini, Claude and Microsoft Copilot, which answer questions on almost any topic. Customer service chatbots are narrower: they answer from one brand's policies and, at their best, act on its orders, like the Klarna, IKEA, Amazon and Flipkart examples above.
Is Siri a chatbot?
Siri is a voice assistant, a chatbot you speak to rather than type to. Like ChatGPT, it's general-purpose; a brand's customer service chatbot works from that brand's own policies, catalogue and orders.
What are the most common chatbot use cases in ecommerce?
After purchase: order status, address changes, COD confirmation, returns and exchanges, damaged items and refund status, the seven examples above. Before purchase: product questions, recommendations and abandoned-cart reminders. For the wider picture, see customer service AI use cases.
